Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Google announced on 9 September 2026 that it will invest at least €13 billion in Finnish data centers and supporting infrastructure across 2027 and 2028, and signed a 22-year power purchase agreement with Fortum covering up to 50 percent of the Loviisa nuclear power plant. The four locations are Hamina, Kajaani, Muhos and Vaala. Google described the package as its largest single investment in Europe.

• The Fortum agreement releases €700 million of capital spending that carried no investment decision. Fortum’s Loviisa life-extension program runs to approximately €1 billion through 2050, and about 80 percent of its individual projects remained unsanctioned before the contract. Fortum states that Loviisa could not continue operating beyond 2030 without that program.

• Goldman Sachs estimates that Google agreed to pay €32 per megawatt hour above the forward curve, or 1.6 times the current 2030 Finnish power price. J.P. Morgan puts the contract at roughly €80 to €90 per megawatt hour against Nordic forwards in the 50s. Neither the tariff nor the escalation formula appears in any public release.

Part One: The Record

Google announced on 9 September 2026 that it will invest at least €13 billion in digital infrastructure in Finland across 2027 and 2028.

Fortum published a stock exchange release the same day confirming a 22-year power purchase agreement with Google covering up to 50 percent of the capacity of the Loviisa nuclear power plant.

The capital funds data centers and supporting infrastructure at four locations: Hamina, Kajaani, Muhos and Vaala.

The announcement discloses no electrical demand, no compute capacity and no construction schedule by site.

Sites And Use Of Proceeds

Google has operated in Hamina since 2009, when it bought a former paper mill and converted it into a data center.

Kajaani, Muhos and Vaala are new sites. Google says its Hamina operations supported more than 600 Finnish suppliers between 2023 and 2025, including DNA, Elisa and Nokia.

The new infrastructure will serve Gemini, Search, Maps and YouTube.

Google projects an average annual contribution of €3.6 billion to Finnish gross domestic product during the 2027 and 2028 construction phase, and more than 37,000 jobs supported nationwide.

Google committed a further €31 million over four years to community programs across the four municipalities.

That package includes €10 million for research and innovation.

The Loviisa Power Purchase Agreement

The agreement begins in 2028 at a smaller volume and reaches 50 percent of Loviisa’s capacity for the years 2030 to 2049. Loviisa holds two pressurized water reactors of 507 MWe each, commissioned in 1977 and 1980.

The plant produces about 8 terawatt hours a year, roughly 10 percent of Finland’s electricity, and employs about 580 Fortum staff.

Fortum runs an investment program of approximately €1 billion to extend operations to 2050.

About 80 percent of those projects and €700 million of the capital spending remain pending investment decisions, which Fortum will take separately.

Fortum states that the plant could not continue operating beyond 2030 without the program, and that the Google contract provides the revenue certainty to complete it.

The agreement is expected to enable a further 10 MW power increase on top of a 38 MW uprate already planned for 2028.

Fortum expects the contract to lift group comparable return on net assets by about 1.4 percentage points once half the plant is contracted.

The long-term target is 14 percent, against 11.0 percent for the twelve months to the end of June 2026. Neither party disclosed the contract price.

Wind, Storage And Grid Partnerships

Google signed onshore wind power purchase agreements with Valorem and Suomen Hyötytuuli, bringing its contracted new-to-grid onshore wind capacity in Finland to 629 MW.

A contracted 94 MW battery system near the Kajaani site is expected to connect to the grid in late 2027, and Fortum has agreed to optimize its dispatch.

The two companies also signed a memorandum of understanding covering new nuclear capacity, renewable capacity, flexibility solutions and energy portfolio management services.

Google is working with Fingrid and Business Finland to identify connection locations. Loviisa sits about 55 kilometers from the Hamina data center.

Part Two: The Read

Google Paid 1.6 Times The Curve

Goldman Sachs estimates that Google agreed to pay €32 per megawatt hour above the forward curve, or 1.6 times the current 2030 Finnish power price.

Coverage has carried that premium as an energy cost and left its mechanism undrawn: the premium is what converts four fifths of an unsanctioned capital program into a financeable one on someone else’s balance sheet.

A reader who treats this as procurement will misprice every European life-extension contract that follows it.

The Offtake Is The Investment Decision

Fortum holds a license to run Loviisa to 2050 and an investment program of about €1 billion to make that possible.

Four fifths of the projects in it, and €700 million of the money, had no investment decision behind them before 9 September.

Fortum takes each of those decisions separately, and each one needs a revenue case that Nordic wholesale prices were not producing.

Fortum’s comparable operating profit fell to €106 million in the second quarter of 2026 from €115 million a year earlier.

The company pointed to a lower achieved power price.

Google does not own the plant, does not fund the capex and does not sit on the committee that approves it.

It supplies the one input that was missing, which is 20 years of contracted revenue on half the output. Europe has a fleet of reactors facing the same arithmetic on the same timetable.

The next 12 to 24 months will show whether the marginal underwriter of European nuclear life extension is a state, a utility balance sheet, or a compute buyer with a 20-year horizon.

The Premium Prices Distance And Firmness

J.P. Morgan puts the contract at roughly €80 to €90 per megawatt hour against Nordic forwards in the 50s.

That gap pays for firmness through a windless Nordic January, and for carbon-free power carrying no interconnection queue and no construction risk.

It also pays for location, which is what the 55 kilometers between Loviisa and Hamina buy.

The premium is now a visible reference point for every European nuclear offtake negotiated after it, which is the part that travels beyond Finland.

One caution applies to the number itself.

The tariff and the escalation formula sit in a contract no outside party has read, so 1.6 times the curve is a bank estimate on undisclosed terms.

J.P. Morgan and Goldman Sachs both moved Fortum to neutral on the news, and neither moved to buy.

Power Came Before The Load Disclosure

Google contracted roughly 507 MW of nuclear capacity share, 629 MW of onshore wind and 94 MW of battery storage.

It disclosed no site demand, no compute capacity and no per-site construction schedule alongside any of it.

The same announcement paired Google with Fingrid and Business Finland to find connection points that avoid network reinforcement.

Read the sequence. In a market where the interconnection queue is the binding constraint, the supply is contracted first and the load is fitted to what the grid can already carry. Site selection in Europe will increasingly follow contracted generation.

Operators who treat power procurement as a step that follows land control are working the order backwards.

What Each Segment Prices Without The Tariff

Private Capital. European nuclear life extension has become a contractable cash flow, and the asset to look for is an operating plant with a pending capex decision and an investment-grade buyer inside transmission reach.

That describes a short list across the Nordics, Iberia and Central Europe, and it is being worked now. The cost of waiting is arithmetic.

The premium is wide because this is the first contract of its type in Europe.

After two or three more of these, the spread over the curve compresses toward the cost of capital of whoever moved first.

Public Markets. Fortum closed at €24.74 on 9 September, up 15.82 percent, and two banks raised targets to €24.30 and €25.50 the following day. What repriced is an estimate.

The tariff, the indexation and the volume ramp are all undisclosed, and Fortum’s own language is that the 1.4-point return lift arrives over time, with full volume starting in 2030.

Holders should price the ramp and treat the next quarterly hedge-ratio disclosure as the first real read on the contract.

Operators. The reference price for firm, carbon-free, adjacent European power now sits near €80 to €90 per megawatt hour on a 22-year term.

An operator without an investment-grade counterparty behind it cannot sign at that level, which narrows the field for every existing plant that comes up for a life-extension decision.

The move available is co-location next to a reactor facing that decision, negotiated before the fleet is spoken for.

The Next Contract Prices Off This One

Google did something narrower than building power and more durable than buying it.

It bought two decades of an operating asset’s remaining life, paid above the curve to do it, and took no ownership.

Through 2027 every European operator with a pending life-extension decision will test that template, and the binding constraint will be the number of reactors with a creditworthy compute buyer within transmission reach.

The first three contracts will set the band that the rest of the fleet trades in.

One question decides whether 1.6 times the curve was expensive. A premium fixed for 22 years and a premium that escalates with Nordic power are different assets, and only one of them transfers the price risk Fortum was carrying.

Ask for the escalation formula before pricing the premium.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

In this week’s issue:

First, Google paid to keep a nuclear plant alive. The company committed at least €13 billion to Finland across 2027–2028 and signed a 22-year contract for up to half of Fortum’s Loviisa output. About 80% of the €700 million life-extension programme lacked an investment decision, raising a key question: what happens when a tech company becomes the credit behind a national power asset?

Then, more than ten US states pulled back. Ohio’s data center sales-tax exemption cost over $1.5 billion in 2025, more than ten times what lawmakers projected, and new applications are frozen. New Jersey enacted a repeal in August.

Finally, the lenders changed. Vantage is seeking $2 billion from PIMCO and PGIM, and Blue Owl is planning a public REIT seeded with roughly $6.5 billion of its own assets.

Let us get into it.

Why the Tax Break Is Now a Credit Question

Part of a data center’s return sits on a contract and part sits on a policy, and this week the two came apart.

A sales-tax exemption is not a construction subsidy.

Servers are a large share of project cost and get replaced every three to five years, so the exemption is a recurring operating line inside the hold-period model.

The counterparty on that line is a legislature.

More than ten US states have now paused or cancelled those exemptions. Ohio’s reached over $1.5 billion in 2025, and Governor Mike DeWine froze new applications.

New Jersey enacted a repeal in August. A tenant covenant survives an election. A tax exemption does not.

Google showed the other route in the same week. It committed €13 billion to Finland and signed a 22-year contract for up to half of Loviisa’s nuclear output, which is what let Fortum sanction a life-extension programme that was about 80% unapproved.

Google did not ask Finland for a concession. It bought a position in Finland’s power system.

So split your model. Take the part of the return that sits on a signed contract with a named counterparty, then take the part that sits on a policy, and underwrite the second at zero.

If the deal still clears, you own an asset. If it does not, you own a lobbying position.

Where the capital went when the terms stopped holding is below.

THIS WEEK BY REGION

The week’s biggest moves — what happened and what it signals.

North America

The terms that shaped the last U.S. data center cycle are being rewritten. States are pulling back tax breaks, with more than ten states pausing or cancelling exemptions, while Texas Republicans are turning against the sector ahead of November. Lenders are adjusting with them.

The financing rotated in the same week. Vantage is seeking $2 billion from PIMCO and PGIM, a move the paper attributes to political and community opposition facing bank lenders.

Blue Owl is planning a listed data center REIT seeded with roughly $6.5 billion of its own assets. Microsoft reportedly plans about 38GW of capacity by 2032, a figure I could not confirm elsewhere.

If you run an infrastructure fund, the incentive line in your model now has a political counterparty, and private credit is pricing that before the banks do.

Europe

Europe sold power certainty this week and got capital for it. Google committed at least €13 billion across 2027 and 2028 to Hamina, Kajaani, Muhos and Vaala, its largest single investment in Europe, alongside a 22-year contract for up to half of Fortum’s Loviisa output.

Bloomberg reports the contract unlocks about €1 billion of investment needed to keep the plant running.

The package also carries 629MW of new onshore wind contracts and a 94MW battery near Kajaani due online in late 2027. Equinix and CPP closed their acquisition of atNorth .

If you invest in real estate, the Nordic bid is no longer about cheap power. It is about power a counterparty will commit to for two decades.

Asia-Pacific

Capacity across the region is being contracted by parties that are not hyperscalers. TCS plans to invest up to $7.4 billion in a one-gigawatt campus in southern India.

Firmus agreed to supply OpenAI with compute capacity in Malaysia, in a market where officials say data center power draw is climbing as temperatures rise.

Nvidia is expanding data centre capacity in Australia to meet AI demand.

For tech and venture investors, the offtaker here is increasingly a services firm or a model developer, not a cloud platform, and the credit behind the lease differs accordingly.

PUBLISHED THIS WEEK

• Saudi Arabia’s LEAP 2026: $15 Billion, Two Hyperscaler Regions, One Sovereign Operator — Two hyperscalers land six weeks apart in a market where one state-backed operator already holds the live GPU clusters.

• Redata Runs Five Years. Three Of Its Four Waivers Run Four Months. — Brazil passed a five-year data center regime in which three of the four federal waivers expire on December 31.

• How Long Does an AI Chip Last (And Why Two of the Biggest Companies Disagree) — Meta and Amazon assigned opposite useful lives to the same AI hardware in the same month, a depreciation gap worth billions a year.

• PPAs, Leases, and Pre-Commitments: How Demand Gets Locked Before Concrete — How powered shell, turnkey, and take-or-pay structures each lock demand before construction, and where the counterparty credit sits in each one.

Sponsored By: Global Data Center Hub

Your trusted source for global AI infrastructure analysis. Every week, we break down the trends shaping the future of data centers, cloud, and AI, from emerging markets to hyperscale battlegrounds.

📰 Subscribe to get expert insights delivered to your inbox.

Where the Capital Went When the Terms Stopped Holding

Fortum had roughly €700 millions of Loviisa life-extension work on the books and had not sanctioned about 80% of it.

A 22-year revenue covenant from one customer is what makes that capital expenditure financeable.

Deliveries start in 2028 at reduced volume and scale to about half the plant’s output across 2030 to 2049, which is the window the modernisation has to serve.

The plant supplies more than 10% of Finland’s electricity and was facing closure in 2030.

The exposure runs both directions, and neither party disclosed the pricing. Fortum’s investment case now rests on one counterparty’s 22-year covenant.

Google’s Finnish power rests on two 1970s-era reactors completing an uprate programme, with a 38MWe increase already under way and due in 2028.

The US ran the opposite experiment. Ohio built a hub on an exemption that cost more than ten times its original estimate, and legislators there have been reported as wanting to reopen agreements already signed with Amazon, Meta and Google.

Nothing signed has been clawed back, but risk pricing has already shifted. That is why Vantage turned to PIMCO and PGIM over bank syndicates, while Blue Owl is seeking public shareholders to fund its next growth phase.

When I run the split across the three, the answers differ.

Google converted a political dependency into a 22-year contract, so the policy share of its Finnish return is close to zero.

Vantage moved its funding to lenders who price political risk instead of avoiding it, which costs more and clears faster.

Blue Owl is the one I cannot answer from disclosure, because the seeded portfolio has not been itemised and I do not know how much of its value sits on incentives a state can reopen.

That is a finding, not a gap. The document to watch is Blue Owl’s registration statement when it files, and specifically whether the seeded assets disclose incentive agreements by state and expiry.

NOTABLE TRANSACTIONS

Key structures and capital moves from this week’s deal tape.

Google and Fortum: 22-year power purchase agreement covering up to 50% of the Loviisa nuclear plant

The 22-year contract scales to half of Loviisa’s output from 2030–2049, supporting a €700 million life-extension programme. A single corporate offtaker now underpins revenue for an asset supplying over 10% of the country’s electricity.

If you underwrite power, the mechanism is a compute contract standing in for a regulated revenue base, and it is the first time that has happened in Europe.

Vantage Data Centers: seeking $2 billion from PIMCO and PGIM

Vantage is seeking $2 billion from institutional lenders after $7.4 billion in recent financing, backed by 17 campuses and over 4GW of leased IT load. The FT cites growing political and community opposition to data center lending.

For operators, the mechanism is a spread you pay to a lender who will price local opposition rather than decline it, and that spread is the new cost of building where permission is contested.

Blue Owl: publicly traded data center REIT seeded with roughly $6.5 billion

Blue Owl plans to seed a listed REIT with $6.5 billion in data center assets, then raise capital for acquisitions. Unlike Blackstone’s $2 billion blind pool, it launches with an operating portfolio, signaling demand for public-market liquidity.

If you allocate to public markets, the mechanism is seeded-asset pricing, and your diligence question is what share of that portfolio’s cash flow depends on incentives a state can reopen.

If you are new to Global Data Center Hub, visit our Start Here page for more information about the newsletter.

If you found this newsletter useful, share it with a colleague.

Have a great week.

— Obinna

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

This article is the 13th article in the series: Ground to Grid: A Free 21-Lesson Guide to Mastering Data Center Development

Join Global Data Center Hub

A data center development becomes financeable at the moment someone signs a contract to use it.

Before that signature, the project is a plan with land, drawings, and a power application. After it, the project has a revenue stream a lender can measure.

This is the commercial layer of development, between site control and construction financing.

Three documents drive it: the lease or pre-lease, power purchase agreement, and pre-commitment or build-to-suit arrangement.

Each one converts an assumption into an obligation.

The developer assumes demand will arrive. The contract replaces that with a named party who has agreed to pay whether it does or not.

Pre-development is funded almost entirely with equity, typically 50 to 70 percent of the stack at entitlement stage, because nothing is contracted yet.

Senior construction debt becomes available once a signed lease and a guaranteed maximum price EPC contract are in place.

Pre-leasing now runs three to four years ahead of delivery, at 70 to 90 percent of capacity.

The Three Contracts

The lease or pre-lease. Data center space is leased by critical IT capacity, measured in MW or kW per month, rather than square footage.

A powered shell lease delivers the building and high-voltage power, while the tenant installs electrical and cooling systems.

These run 15–20 years, cost $11–14 million per MW, and price at a 7.0–8.0% cap rate.

A turnkey lease delivers fully commissioned server halls, typically for 10–15 years at $100–150 per kW per month, costing $15–20 million per MW and pricing at 5.5–6.5%.

The power purchase agreement. A PPA is typically a 5–20-year contract to buy electricity at a pre-agreed price, locking in the asset’s highest operating cost.

Physical PPAs deliver power to a named grid node, while virtual PPAs settle financially and hedge costs without physical delivery.

Behind-the-meter arrangements use on-site generation; front-of-the-meter arrangements rely on the public grid.

The pre-commitment or build-to-suit. The tenant commits before design is finalized and shapes the specification around its own requirements.

The term underneath is take-or-pay: the tenant pays a fixed capacity charge on the full reserved load from rent commencement, whether or not a single server has been racked.

How the Contracts Allocate Risk

A long lease moves absorption risk from the developer to the tenant.

The developer no longer carries the cost of an empty building.

A PPA moves electricity price and supply risk to the generator and gives the lender a fixed input cost.

A build-to-suit or take-or-pay contract moves demand risk almost entirely to the tenant, which supports the highest leverage.

Risk does not disappear at the developer’s end.

Turnkey leases carry a contractual power-usage effectiveness cap, and any drift above it is the developer’s cost.

On a 100 MW campus at $0.08 per kilowatt-hour, a 0.10 drift runs to more than $7 million a year.

Delay damages typically run $1,000 to $5,000 per MW per day past target COD.

Long-stop dates must be sequenced so the contractor’s deadline falls before the tenant’s termination right, which falls well before loan maturity.

Why the Counterparty Matters as Much as the Contract

The contract is only as bankable as the party signing it.

Senior lenders model a minimum debt service coverage ratio of 1.15 to 1.25 times against an investment-grade tenant.

Against a weaker counterparty, or a contract permitting volume flexibility, that floor rises to 1.35 to 1.50 times.

Higher coverage means less debt, and more equity for the same building.

A hyperscaler lease sits at the strong end: investment-grade balance sheets and often a parent guarantee on rent commencement and termination fees.

An AI-native tenant sits differently.

Demand may be intense, and the term may be long, but the credit is thinner.

Lenders respond by tightening structure: no walk-away rights, must-pay economics, bankruptcy-remote special purpose vehicles, and receivables assigned in advance.

NVIDIA now provides take-or-pay commitments, and minimum revenue guarantees on neocloud capacity, reaching $36 billion by late July 2026.

That gives lenders a creditworthy name in place of the operator.

A Worked Example

In September 2024, Microsoft signed a 20-year PPA for 100 percent of the 835 MW output of Three-Mile Island Unit 1, now the Crane Clean Energy Center.

The estimated all-in price is $100 to $115 per MWh, and the contract supports a $1.6 billion restart of a reactor that would otherwise have remained shut down.

Constellation did not finance that restart against a power price forecast.

It financed against a 20-year obligation from an investment-grade buyer. The cash flow reads like a bond to a lender.

Structure mattered as much as credit. Microsoft took delivery at the front of the meter under standard PJM tariffs.

Talen Energy and Amazon Web Services went behind the meter at Susquehanna, and FERC rejected that interconnection agreement in November 2024.

Strategic Implications

For developers. Secure the anchor tenant before pursuing debt. The lease determines how much capital you can raise and at what cost. A weak tenant raises lender coverage requirements, reduces debt capacity, and forces more expensive equity. Accepting a weak tenant just to reach financial close is often worse than delaying.

For investors. Assess the counterparty as closely as the contract. A long lease with a weak tenant may be short-lived. Look beyond the headline term: review cancellation rights, parent guarantees, termination fees, and regulatory dependencies. Two contracts can model identically but carry very different risks.

For operators. Lease and PPA terms govern revenue and major costs for years. Efficiency covenants become operating obligations, while PUE caps create liability when performance slips. Capacity charges begin at rent commencement, so commissioning delays reduce margins. Involve the operating team before terms are finalized.

For policymakers. PPA structures influence which generation gets built and when. Long-term offtake from a creditworthy buyer can finance major generation projects. Behind-the-meter versus front-of-the-meter structures also determine who bears transmission costs, driving requirements for minimum capacity commitments and upfront infrastructure funding.

Key Takeaway

You do not finance a data center. You finance the contracts behind it.

The lease and the PPA are the project’s real collateral. The building is what they are attached to.

That changes what you examine first. The megawatt figure and the design tell you what is being built.

The lease term, the coverage ratio, the termination rights, and the delivery structure tell you whether it gets built at all.

The tenant’s credit is what makes the paper worth anything. The debt quantum, the equity requirement, the cost of capital, and the timeline to close are all priced off that one signature.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Useful life is the number of years a company assumes a computer will keep earning its keep, and it is an estimate rather than a measurement. Meta, Amazon, Microsoft, Alphabet, and Oracle each publish this number in their annual filings with the US Securities and Exchange Commission. The number decides how much of the purchase price is charged against profit each year.

• Meta extended the assumed useful life of most of its servers and network equipment to five and a half years effective 1 January 2025 and expected the change to cut that year’s wear-and-tear charge by about $2.9 billion. Amazon moved the opposite way on the same date, shortening the assumed life of a portion of its servers and networking equipment from six years to five. Amazon gave the reason in its filing as the increased pace of technology development, particularly in artificial intelligence and machine learning.

• The profits reported by the companies running artificial intelligence rest on an assumption about how long a computer keeps working, and the companies do not agree on the assumption. Amazon disclosed that its change raised 2025 depreciation and amortization expense by $1.4 billion and cut net income by $1.0 billion. Both the longer estimate and the shorter one are permitted under the same accounting rules.

Every computer you have ever owned got slower, then got replaced.

Nobody wrote the date down in advance.

The companies building artificial intelligence do write it down.

They are required to, and the number they choose moves how much money they appear to make.

That number is called useful life. It is the number of years a company assumes a machine will keep doing useful work before it is retired.

It sounds like a technical detail buried in a filing. Yet it is one of the largest open questions in the AI boom, because companies buying the same equipment have written down different answers.

Every one of them is guessing. That is the part worth sitting with.

Meta’s wear-and-tear bill on servers alone reached $13.36 billion last year

Start with what the number does.

When a company buys something that lasts for years, it spreads the cost across all the years it expects to use the thing, rather than charging the whole amount in the year it writes the check.

That spreading is called depreciation, and most people already hold the idea from cars.

A car bought for $30,000 and expected to last ten years loses value on a schedule.

If you assume it lasts five years instead, the schedule doubles, and every year of ownership looks more expensive.

Servers work the same way, at massive scale. In 2025, Meta reported $13.36 billion in depreciation on servers and network assets, up from $11.34 billion in 2024 and $7.32 billion in 2023.

That figure covers wear alone; on machines Meta had already bought and paid for.

It is the portion of that existing equipment the company counted as used up inside twelve months, on a single line of a single company’s accounts. In two years, it has grown close to double.

Change the assumed life and that line moves by billions without a single machine being bought, sold, or touched.

The thing most readers would get wrong here is treating wear as a physical fact. The machine ages at whatever rate the machine ages. The number on the page is a choice made in advance about a future nobody has seen yet.

Two companies, the same machines, the same January

On 1 January 2025, Meta and Amazon both changed their answer. They changed it in opposite directions.

Meta extended the estimated useful life of most of its servers and network assets to five and a half years, up from a prior range of four to five years.

The company expected the change to reduce its 2025 depreciation expense by roughly $2.9 billion.

Longer life, smaller annual charge, higher reported profit, and not one new machine required to produce the effect.

Amazon went the other way on the very same effective date.

It shortened the assumed useful life of a portion of its servers and networking equipment from six years back to five.

Amazon’s filing named the reason directly: the increased pace of technology development, particularly in artificial intelligence and machine learning.

In plain terms, Amazon told its shareholders that the hardware is going out of date faster than it used to.

The cost was disclosed too. For 2025, Amazon reported that the change increased depreciation and amortization by $1.4 billion and reduced net income by $1.0 billion, or $0.10 per share, primarily affecting Amazon Web Services.

Two companies. The same core supplier. The same technology cycle. The same calendar date. Opposite conclusions, both signed off by auditors, both entirely legal.

If you have been reading headlines about AI profits and taking them as measurements, this is the moment to notice how much of those figures is an estimate about time.

Amazon changed its own answer three times in four years

The disagreement runs inside the companies as well.

Amazon lengthened the assumed life of its servers from four years to five effective January 2022.

It lengthened them again, from five years to six, effective January 2024, and disclosed that the change cut depreciation and amortization by $3.2 billion and added $2.5 billion to net income for that year, twenty-three cents a share.

Twelve months later it reversed course and went back to five.

Read the three moves together and you have a company that stretched the number twice while the machines looked durable, then pulled it back when the pace of new chip releases picked up.

Here is the honest half of the story, and it cuts both ways.

Amazon’s reversal is the strongest public evidence that the longer schedules elsewhere are generous, and critics have made exactly that argument.

It is also true that older chips remain heavily used, particularly for running finished AI models rather than training new ones, so the case for a very short life can be overstated. No regulator has found any of these companies at fault.

There is no rule that names the correct answer, because the correct answer depends on how fast the technology moves next.

The misunderstanding worth clearing is that an accounting footnote is a small thing.

On spending this size, a footnote measured in years is one of the largest single levers on reported profit that a company controls.

The chip still works. A newer one does the job for less.

A chip that has been retired is almost always a chip that still works. That distinction is where most of the confusion lives.

A five-year-old phone still turns on. It gets replaced because the new one is better and because apps stop being built for the old one. AI chips retire for a version of the same reason.

NVIDIA, the company whose chips most of the AI boom runs on, has moved to releasing a major new generation roughly every year. Each generation does more work for each dollar of electricity.

The older chip keeps running. It simply costs more to get the same result out of it, and at some point, the electricity bill alone makes retirement the cheaper option.

That is why nobody can settle the question by testing the hardware.

The machine keeps running and gets out-competed anyway, on a schedule set by a supplier’s product calendar rather than by any physical limit.

It is also why the second-hand market gives no clean answer.

Prices for renting and buying previous-generation chips have swung hard in both directions over the past two years, moving on chip supply, memory shortages, and demand for running finished models.

There is no settled resale value to check the estimate against.

If you have assumed that somewhere inside these companies sits a measured number, what sits there instead is a market that has yet to decide.

Microsoft, Alphabet, and Oracle all stretched the number first

Amazon’s reversal stands out because it runs against the direction everyone else moved.

Microsoft told investors in 2022 that it was extending the depreciable life of server and network equipment in its cloud business from four years to six, with a disclosed benefit of roughly $3.7 billion in the following financial year.

Alphabet, the parent company of Google, announced a similar assessment in its January 2023 earnings release, moving servers from four years to six and certain network equipment from five to six.

The actual effect for the full year 2023 was a $3.9 billion reduction in depreciation expense and a $3.0 billion increase in net income, twenty-four cents per diluted share.

Oracle raised its own estimate for servers and networking equipment from five years to six, effective at the start of its 2025 financial year.

Four large companies stretched the number. One pulled it back. Every one of them buys from the same handful of suppliers and watches the same product releases.

The thing to carry out of this is simple. One of the biggest inputs into whether AI makes money is an open disagreement, and the people disagreeing are the ones closest to the machines.

Why this sits under the AI you use every day

Every AI answer you have ever received came off a machine sitting in a building, drawing power, wearing out on a clock nobody has agreed how to read.

The public argument about whether all this construction makes sense gets conducted in the language of power and enormous spending announcements.

Underneath all of it, holding up the profit figures those announcements are measured against, is a number written in years, chosen in advance, and revised whenever the companies change their minds about the future.

Meta wrote five and a half. Amazon wrote five. They wrote it on the same day, about the same machines, and the world will not know which of them was right for several more years.

The next time a headline tells you an AI business has turned profitable, part of that answer was set by an estimate of how long a computer lasts.

What is the oldest piece of technology you still use every day, and what finally makes you replace something rather than keep it running?

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Equipment lead times now run longer than permitting and interconnection timelines, which makes the factory queue the binding constraint on data center delivery. Large transformers, generator step-up units, and heavy-frame gas turbines have moved from months of lead time to multi-year delivery windows.

• The industry still orders electrical hardware after financial close, and that sequence no longer works. The equipment queue is now longer than the entire approval cycle it was built to follow.

• Disciplined capital reserves manufacturing slots with non-refundable deposits before permits are granted and prices the deposit as an option premium on access. Balance-sheet buyers absorb that forfeiture risk, so the factory allocates capacity by balance-sheet strength.

The factory sets your delivery date

The factory queue now decides when your data center opens.

Power equipment lead times have outpaced the permitting and interconnection process, so the item ordered last is the item that arrives last.

Project finance was built on the opposite assumption.

It treated electrical hardware as a purchase that follows financial close, sequenced behind site control, offtake, and the system impact study.

Apply that sequence today, and you will complete a building that cannot be energized for years.

The shell is finished. The debt is drawn. The revenue is zero.

Three buyers, one supply base

The constraint is intensifying because demand arrived from several directions at once.

AI campuses, utility grid renewal programs, and industrial electrification are all bidding for the same transformers, switchgear, and turbine slots.

No buyer can substitute a different product.

So, the question is not how much capital has been committed to data centers.

The question is where that capital converts into energized capacity, and who controls the hardware that makes the conversion possible.

We made a version of this argument when chip allocation, not leasing, set absorption timing.

The bottleneck is a furnace

The bottleneck sits upstream of the assembly line.

Large power transformers depend on grain-oriented electrical steel, made by a small number of specialized mills, and the United States satisfies only a minority of its own consumption internally.

Coil winding remains a manual trade. A custom utility-intertie unit is not a catalog item. The order enters a queue now measured in years.

Turbines are worse. The constraint sits in hot-section castings, the blades and vanes that survive firing temperatures above 1,600 degrees Celsius.

Those parts come from a narrow tier of specialized foundries.

Expanding that tier requires vacuum melting furnaces and metallurgical labor that left the industry during the thermal downturn of the 2010s.

AI-grade workloads make this sharper. Density raises the load per square foot, which pushes projects toward larger step-up units and higher-voltage interties.

Larger units carry the longest lead times in the category. The more advanced your compute design, the further back you sit in the queue.

Consider a composite that repeats across secondary markets.

A regional developer secures land, signs an anchor tenant, and holds the transformer order until the transmission utility finishes its system impact study.

The order is then placed. The delivery window falls two years past rent commencement, and the lease carries no relief clause for equipment delay.

Scarcity is not a pricing problem

The industry has long assumed that equipment scarcity is a cost problem.

That assumption held when factories carried spare capacity and a buyer willing to pay a premium could move up the queue.

It does not hold now, because the constraint is physical throughput.

A buyer willing to pay more does not move up the queue, because the queue is set by furnace hours.

Capital cannot conjure a casting furnace.

Adding manufacturing capacity takes years after the funding is committed, and every OEM expanding today is expanding into a book that is already full.

Policy closed the side doors

Trade policy has narrowed the escape routes at the same time. Section 301 tariffs on Chinese lithium-ion cells for storage stepped up at the start of 2026.

Content rules under the One Big Beautiful Bill Act require 2026 storage projects to source most components outside the foreign-entity-of-concern definition, with the threshold rising later.

Solar carries its own lower threshold. A Section 232 investigation into imported polysilicon sits over the wafer supply that Western module assembly still depends on.

In Europe, the Net-Zero Industry Act sets a binding domestic manufacturing benchmark for 2030 and adds non-price criteria that penalize concentrated sourcing.

The pattern is clear. Compliance is a qualification process, and qualification takes time you didn't budget for.

Regulators have closed one of the faster workarounds.

When a co-location arrangement was tested at a large nuclear site, the federal energy regulator rejected the amended interconnection agreement, finding that the structure pushed transmission costs onto other ratepayers.

Sponsors who underwrote behind-the-meter co-location to bypass the queue must now re-underwrite it under standard network tariffs.

What binds, by lens

For independent operators, the binding variable is qualification, not price. You cannot standardize what you have not specified, and you cannot order what you have not specified. Freeze the electrical package early, at a common collection and transmission voltage, and a bespoke engineering order becomes repeatable. Operators that hold the specification open until the tenant signs forfeit the only lever that moves them up a manufacturing queue.

For private equity and infrastructure investors, the binding variable is the gap between the delivery date and the model. A project underwritten to distribute in year two behaves very differently when first revenue lands in year four. Interest during construction accrues against a drawn balance for two extra years. Coverage tightens, and a return profile underwritten in the low teens compresses toward single digits with no change to the operating assumptions. The schedule is the return.

For public equity, the dispersion is already visible in where the economics sit. Value has migrated from the asset to the supplier. Order books at the equipment level now carry multi-year visibility, a rare quality in a cyclical industry. Developers without secured hardware struggle to attract construction partners. This is the same dispersion that opened around route diversity in fiber. The spread between a platform that holds delivery slots and one that holds only announcements is not a valuation opinion. It reflects what each company can actually build.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Brazil’s Senate approved PL 278/2026 on September 1, 2026, creating Redata, the special tax regime for data center services. The bill suspends PIS/Cofins, PIS/Cofins on imports, IPI and the Import Tax on ICT equipment entering a data center’s fixed assets. The government estimates the relief at R$5.2 billion in 2026, R$1.0 billion in 2027 and R$1.05 billion in 2028. A companion bill, PLP 74/2026, cleared the Senate 66 votes to none on September 3.

• Article 11-J of Redata ends three of the four federal waivers on December 31, 2026, leaving the Import Tax as the only levy waived across the regime’s five-year term. PIS/Cofins and IPI are extinguished on January 1, 2027, under Constitutional Amendment 132 of 2023 and replaced by CBS and IBS. Redata is worth R$1.0 billion a year from 2027, one fifth of its first-year value.

• Brazil’s Senate widened Redata’s energy standard from clean or renewable sources to renewable or low-emission sources through a drafting amendment, admitting gas-fired generation. The amendment class was chosen to avoid returning the bill to the Chamber of Deputies. The statute leaves the definition of low emission to executive decree.

Part One: The Record

Brazil’s Senate approved PL 278/2026 on September 1, 2026, creating Redata, the Regime Especial de Tributação para Serviços de Datacenter.

The bill suspends four federal taxes on information and communications technology equipment bought or imported for data center fixed assets.

The government estimates the relief at R$5.2 billion in 2026, R$1.0 billion in 2027 and R$1.05 billion in 2028.

The Senate passed the text by symbolic vote without changes to its merits and sent it to President Luiz Inácio Lula da Silva for sanction.

The Four Taxes And The Qualifying Goods

Article 11-C suspends PIS/Pasep and Cofins on revenue, PIS/Cofins-Importação, IPI, and the Import Tax.

The suspension covers electronic components and other ICT products destined for the fixed assets of an approved company.

It converts to a zero rate once the company meets its commitments and puts the asset into service.

The covered products are set by executive decree and once published that list may only be amended to add goods.

The IPI suspension excludes goods industrialized in the Manaus Free Trade Zone.

The Import Tax suspension includes them and is otherwise limited to goods with no equivalent national production, a test the Senate amended from no similar national product.

Suppliers under contract to an approved operator may seek co-authorization and take the same suspensions on their inputs.

Counterpart Obligations And Regional Terms

Approval carries five commitments.

The company must make at least 10 percent of installed processing, storage and data handling capacity available to the Brazilian domestic market.

It must meet sustainability indicators set by decree. It must cover its full electricity demand through supply contracts or self-generation from renewable or low-emission sources, a standard the Senate widened from clean or renewable sources.

It must hold a Water Usage Effectiveness index at or below 0.05 liters per kilowatt-hour, measured annually. It must invest 2 percent of the value of benefited goods in Brazilian research and development.

North, Northeast and Center-West projects receive a 20 percent reduction on the capacity floor and the research commitment, to 8 percent and 1.6 percent.

At least 40 percent of research funds the regime generates must go to those regions.

The capacity floor may be replaced by an additional 10 percent research investment.

Receita Federal grants authorization. Companies under the Simples Nacional regime are barred.

Missing the capacity floor suspends benefits on new purchases and, after 180 days uncured, cancels authorization and bars the group for two years.

The Companion Bill And Presidential Sanction

The 2026 budget had earmarked R$5.2 billion for the provisional measure that first created Redata, and that authorization lapsed when the measure expired.

A companion bill, PLP 74/2026, placed Redata among the exceptions to fiscal rules limiting new tax benefits in 2026.

Redata was struck from the rapporteur’s report on September 3 and restored before the vote.

The Chamber of Deputies approved the bill that day and the Senate approved it hours later by 66 votes to none.

Both texts sit with the president. Neither is in force.

Part Two: The Read

R$1.0 Billion Is What Redata Is Worth From 2027

Redata is worth R$1.0 billion a year from 2027, one fifth of its first-year value.

Coverage has carried the five-year term and the R$5.2 billion headline without reading Article 11-J, whose second paragraph ends the PIS/Cofins and IPI waivers on December 31, 2026.

An operator underwriting a 2028 energization on the strength of a five-year regime is underwriting one tax, not four.

Three Of Four Waivers Die With The Taxes They Waive

Article 11-J gives the Article 11-C benefits a five-year term.

Its second paragraph then limits the benefits attached to the first three items on that list to December 31, 2026, subject to Constitutional Amendment 132 of 2023 and Complementary Law 214 of 2025.

Those three items are PIS/Cofins on revenue, PIS/Cofins on imports, and IPI. The fourth is the Import Tax.

The reason is the consumption tax reform. PIS/Cofins and IPI are themselves extinguished on January 1, 2027, and replaced by CBS and IBS.

A waiver cannot outlive the tax it waives. Nothing in the statute carries equivalent relief into CBS.

The official fiscal series confirms the reading. R$5.2 billion in 2026 covers four taxes.

R$1.0 billion in 2027 and R$1.05 billion in 2028 cover the Import Tax alone, which is the only levy the reform leaves standing.

Published commentary has attributed the 81 percent drop to an initial import wave normalizing. The statute supplies a simpler explanation.

The consequence over the next four quarters is a procurement race into 2026.

Every server, switchgear set and power distribution unit that clears customs before December 31 carries relief on four taxes.

Everything after it carries relief on one and recovers the rest through CBS input credits on whatever timetable the transition rules set.

That converts an absolute cost saving into a working capital position, and Brazilian input credits have never been quick to monetize.

A Baseload Standard Passed As A Drafting Correction

The energy commitment moved from clean or renewable sources to renewable or low-emission sources.

That change was made as an emenda de redação, the amendment class reserved for wording, chosen precisely because a merit amendment would have sent the bill back to the Chamber of Deputies and past the legislative window. Hugo Motta objected to the maneuver.

The wording class does not match what the wording does. Low-emission admits gas-fired self-generation, which clean or renewable did not.

The statute leaves the definition to executive decree.

Brazil’s principal pitch to hyperscale tenants is a grid that runs overwhelmingly on hydro, wind and solar, and the sovereign has now made the eligibility standard behind that pitch conditional on a decree nobody has read.

Watch the decree, not the law. It sets the emissions threshold, and with it whether a gas turbine behind the meter in São Paulo qualifies on the same terms as a wind contract in Rio Grande do Norte.

The Domestic Capacity Floor Has A Published Price

Redata’s sovereignty commitment is priced.

An operator must reserve 10 percent of installed capacity for the Brazilian market or invest an additional 10 percent of the value of its benefited hardware in domestic research.

The second option is a fee, and the statute publishes it.

Capacity ceded free to research institutions or public bodies counts toward the floor with a multiplier set by decree.

The multiplier and the fee together decide what Redata produces. Set the multiplier high and the floor is cheap to satisfy with donated capacity.

Leave the fee attractive and export-oriented operators buy their way out, and the regime becomes an equipment subsidy with a research levy attached rather than a domestic compute program.

Neither number is in the statute. Both are in the regulation.

What Each Segment Prices Before December 31

Private Capital. Underwrite Brazilian data center capex on two tax cases, not one. Model the 2026 procurement window at four-tax relief, with later periods using Import Tax relief plus CBS credit recovery. Keep the monetization period as an explicit variable, not working capital. Funds that underwrite a single five-year case will carry a cost base that is right for four months and wrong for fifty-six.

Public Markets. The listed exposure runs through operators with Brazilian platforms, Digital Realty’s Ascenty among them, and through the equipment vendors selling into the 2026 window. Brasscom projects US$92 billions of cumulative investment between 2025 and 2031, US$69 billions of its equipment. That projection is not conditioned on which taxes survive 2026. Holders should expect a 2026 revenue pull-forward and should ask whether guidance separates it from run rate.

Operators. Land the equipment order before December 31 or accept a different cost base. That is a procurement decision available now and unavailable in January. Operators also carry the regulatory risk that the other two segments only price: the emissions threshold, the capacity multiplier, and the product list, all set by decree, all capable of moving a project’s eligibility after the capital is committed.

The Decree Outranks The Statute Here

Brazil has spent a year fighting over Redata’s headline and delegated its economics.

The emissions threshold, the capacity multiplier, the covered product list and the compliance timetable all sit in a regulation that does not yet exist, and three of the four tax waivers expire before most projects that qualify for them will energize.

The regime that matters is the one the executive writes, and it will be written under the same deadline pressure that produced a baseload energy standard passed as a spelling correction.

The question for the next two quarters is whether the CBS transition carries equivalent relief for data center hardware.

If it does not, Redata was a four-month import window with a five-year name, and every capital plan built on the longer number reprices on January 1.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Anthropic has reportedly signed a $35 billion, six-year cloud computing agreement with Lambda covering roughly 350 megawatts at Hut 8’s Beacon Point campus in Nueces County, Texas. The Wall Street Journal reported the agreement, and Reuters confirmed it with a single source familiar with the matter. No party has confirmed the terms, and Anthropic, NVIDIA, Lambda and Hut 8 have all declined to comment or not responded.

• Hut 8’s disclosed leases price Beacon Point at $1.86 million per megawatt a year, fixed for fifteen years with a 3.0 percent annual escalator. That figure comes from Hut 8’s own disclosure of $19.6 billion of base-term contract value across 704 megawatts. It is the only price in this transaction that appears in a company filing.

• The reported compute contract prices at roughly nine times the rent underneath it, and only the rent sits inside the security package. Hut 8 closed $4.25 billions of senior secured notes against fifteen years of triple net rent from a high-investment-grade tenant in June 2026, two months before any Anthropic agreement was reported.

Part One: The Record

Anthropic has signed a cloud computing agreement worth $35 billion with Lambda, according to a source familiar with the matter cited by Reuters on August 31, 2026.

The Wall Street Journal reported the agreement first. No party has confirmed the terms.

Anthropic declined to comment, and NVIDIA, Lambda and Hut 8 did not respond to requests for comment.

Bloomberg reported that the arrangement covers roughly 350 megawatts at a data center in Nueces County, Texas, developed by Hut 8.

A source familiar with the deal told AFP that the term runs six years.

Reported Structure And Counterparty Roles

The Wall Street Journal reported that NVIDIA holds the lease on the data center and reached its own agreement with Hut 8 some weeks before the Anthropic agreement.

Under the structure the Journal described, Lambda installs NVIDIA chips at the facility and sells the resulting compute capacity to Anthropic.

The Journal reported that Lambda pays NVIDIA an undisclosed amount for access to the space, and that NVIDIA is an investor in both Anthropic and Lambda.

The Financial Times identified NVIDIA as the previously unnamed Beacon Point tenant on July 28, 2026, citing five people familiar with the arrangement.

Reuters reported that it could not independently verify the identification.

NVIDIA has not confirmed it, and Hut 8 has not named the tenant.

Beacon Point Leases And Disclosed Terms

Hut 8 disclosed the underlying leases directly. On July 20, 2026, the company announced a second fifteen-year lease for 352 megawatts of IT capacity at Beacon Point, executed with the same high-investment-grade tenant that signed the Phase 1 lease.

The second lease carries a base-term contract value of $9.8 billion inclusive of a 3.0 percent annual base rent escalator, and it doubles that tenant’s contracted capacity at the campus to 704 megawatts.

Hut 8 stated that base-term contract value for the full campus rises to $19.6 billion.

Both leases are triple net. Hut 8 expects average annual net operating income of $1.31 billion across the campus once stabilized.

Three five-year renewal options per lease would increase potential campus-level contract value to $50.2 billion if all are exercised.

Power Supply And Delivery Timeline

Beacon Point sits on 525 acres in Nueces County and is secured by an interconnection agreement with AEP Texas for 1,000 megawatts of utility capacity.

Hut 8 stated that it redesigned the first data hall around NVIDIA’s architecture, raising capacity 57 percent within the same land and utility footprint.

The second phase is designed to the NVIDIA DSX reference architecture. Initial energization remains on schedule for the first quarter of 2027, and Hut 8 expects initial Phase 2 data hall delivery in the second quarter of 2028.

Lambda priced a $926 million senior secured term loan B on August 27, 2026, with Morgan Stanley as lead and MUFG as joint bookrunner, to fund GPU infrastructure for a committed customer deployment.

Neither Anthropic nor Lambda has disclosed a start date for capacity under the reported agreement.

Part Two: The Read

The Signal Is $1.86 Million Per Megawatt

Hut 8’s disclosed leases price Beacon Point at $1.86 million per megawatt a year, fixed across a fifteen-year base term and rising 3.0 percent annually.

That figure is Hut 8’s own arithmetic: $19.6 billion of base-term contract value across 704 megawatts and fifteen years.

Coverage of the Anthropic agreement has carried Hut 8’s rent and the reported compute contract in the same articles without ever dividing one into the other.

Everything that determines who is exposed to what lives in the distance between those two numbers.

The Financeable Document Is The One Hut 8 Filed

Hut 8 closed $4.25 billions of senior secured notes for Beacon Point in June 2026, two months before any Anthropic agreement was reported.

That paper is secured against fifteen years of triple net rent from a high-investment-grade tenant.

Lenders sized it on lease cash flow with a fixed escalator and a named credit tier attached.

Anthropic’s compute contract has no public existence beyond reporting.

Hut 8’s leases sit in company disclosure with terms, escalators and renewal options attached.

When a project needs capital, the document that carries a rating carries the deal.

The practical consequence over the next four quarters is that any lender pricing exposure to this campus is pricing a real estate lease with a chip vendor’s name reportedly on it, and the AI offtake below that lease sits outside the security package entirely.

The same gap between disclosed dates and announced dollars runs through an entire market in Saudi Arabia's LEAP 2026: $15 Billion, Two Hyperscaler Regions, One Sovereign Operator.

Nine Dollars Of Compute Sit On One Dollar Of Rent

Take the reported figures at face value for one calculation. $35 billion across roughly 350 megawatts over six years works out to $16.7 million per megawatt a year, derived entirely from figures no party has confirmed.

Hut 8’s disclosed rent on the same class of space is $1.86 million per megawatt a year.

The compute contract prices at roughly nine times the rent beneath it.

That spread buys everything the shell does not contain. GPUs, networking, power, cooling, staffing, and the operator’s return all come out of it.

Most of it buys silicon. Silicon depreciates on a two-to-three-year cadence, and the building depreciates over decades.

A structure that places fifteen years of fixed rent underneath a six-year compute contract has put the durable claim on the shell and the perishable claim on the hardware.

Anyone financing GPUs into this campus over the next twelve to twenty-four months is underwriting what that hardware is worth when the contract is halfway through and the chips are two generations old.

The Escalator Runs Longer Than Any Chip Generation

Hut 8’s leases run fifteen years with a 3.0 percent annual escalator and three five-year renewal options that could carry campus contract value to $50.2 billion.

Phase 1 energizes in the first quarter of 2027.

Phase 2 data halls arrive in the second quarter of 2028. By the time Phase 2 delivers, the equipment installed in Phase 1 is a generation behind.

The landlord holds a fifteen-year claim with a known escalator and no exposure to which chips fill the halls. The renewal options belong to the tenant.

Whoever installs the silicon absorbs all of the obsolescence.

That asymmetry explains why former bitcoin miners keep winning these contracts.

Hut 8 took a greenfield site in Nueces County to full commercialization against 1,000 megawatts of AEP Texas interconnection in a matter of months.

The scarce asset was the interconnection agreement, and the party holding it wrote fifteen-year triple net leases against it.

Underwrite The Lease Before The Offtake

Private Capital. Beacon Point paper is investment-grade real estate cash flow carrying an AI label.

Firms bidding on the $4.25 billion notes or on comparable special purpose vehicle issuance should price the lease term, the escalator, the tenant’s credit tier and the renewal optionality, and treat the compute contract above as information about demand.

Funds that underwrite the AI story pay an AI premium for rent. Funds that wait for the tenant to be confirmed in a filing will bid against a repriced curve, because that confirmation is the event that compresses the spread.

Public Markets. Hut 8 reports 949 megawatts of contracted capacity, $26.6 billion of aggregate base-term contract value and average annual net operating income above $1.75 billion.

Beacon Point accounts for 704 of those 949 megawatts under two leases with a single tenant.

Holders should track tenant concentration disclosure and the delivery dates in the first quarter of 2027 and the second quarter of 2028.

The equity trades on compute demand while the cash flows are rent, and the cost of missing that distinction is owning a landlord at a neocloud multiple.

Operators. Hut 8 secured 1,000 megawatts from AEP Texas before it had a tenant, then redesigned Phase 1 around NVIDIA’s architecture and lifted capacity 57 percent inside the same footprint.

Operators holding live interconnection agreements in ERCOT and comparable markets can write leases on these terms now.

Operators still in queue will lease at whatever price the first movers have already set, and Beacon Point has now set one in public.

The Next Lease Prices Against Beacon Point

Beacon Point has established a public benchmark. $1.86 million per megawatt a year, fifteen years, 3.0 percent annual escalation, triple net, with three five-year renewals behind it.

Every developer negotiating a hyperscale lease over the next four quarters will have that number put in front of them by a counterparty, and every lender will use it to size the next special purpose vehicle.

The reported $35 billion above it will be remembered as a demand signal long after the headline fades.

The open question is which number gets confirmed first. If the tenant identification appears in a filing before Phase 1 energizes in the first quarter of 2027, the lease becomes a rated benchmark and the spread on the next gigawatt campus compresses against it.

If it does not, the market spends another year pricing a gigawatt of Texas real estate on five anonymous sources and a press release that names nobody.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Saudi Arabia’s LEAP 2026 closed with nearly $15 billion in announcements, investments and agreements. The fifth edition of the conference ran from August 31 to September 3, 2026 at the Riyadh Exhibition and Convention Center in Malham. The total spans AI infrastructure, data centers, cloud, manufacturing, venture capital and skills programs.

• HUMAIN, the artificial intelligence company owned by Saudi Arabia’s Public Investment Fund, anchored the largest infrastructure commitments at LEAP 2026. These include a 250 megawatt phase with AMD and Cisco beginning in 2027, a planned 250 megawatt facility with Together AI, and 100 megawatts inside DataVolt’s campus at Oxagon in NEOM. A separate vehicle managed by BSF Capital is targeting an initial $2.5 billion to finance 250 megawatts being built with Al Moammar Information Systems.

• Microsoft opens its Saudi Arabia East cloud region in November 2026 and AWS opens its Saudi region in December 2026, placing two hyperscaler regions in the Kingdom within roughly thirty days. AWS reports more than $5.3 billion of planned investment in the region and has committed with HUMAIN to deliver up to 50 megawatts inside a dedicated AI zone by 2028.
  

What LEAP Is

LEAP is Saudi Arabia’s national technology conference. It is held once a year in Riyadh, and the 2026 edition was the fifth. It ran for four days, from August 31 to September 3, 2026, at the Riyadh Exhibition and Convention Center in Malham. The theme was “Into New Worlds.”

Four bodies organize it. The Ministry of Communications and Information Technology leads. It works with the Saudi Federation for Cybersecurity, Programming and Drones, an events company called Tahaluf, and the Events Investment Fund.

LEAP is not a trade show that happens to be in Saudi Arabia. It is a government event, and the government uses it as a release window. Saudi ministries, state-backed companies and their foreign partners hold their news until LEAP week and then publish it together. Global vendors do the same. A launch announced at LEAP reaches Saudi buyers and the Saudi state at the same moment.

So one week in Riyadh sets much of the year’s Saudi technology news. The 2026 edition produced nearly $15 billion in announcements, investments and agreements. That money covers AI infrastructure, data centers and cloud. It also covers manufacturing, venture capital and skills.

Why LEAP Matters to the Global Data Center Sector

Most readers of this publication will never attend LEAP. It still shapes the market they invest in, for four reasons.

It is one of the few places where power arrives with the project. In most large markets a developer buys land first. Then the developer joins a waiting list for a grid connection. The wait sets the schedule. Saudi Arabia is trying to run those steps at the same time.

At LEAP 2026, the national grid operator and the state electricity developer signed agreements directly with data center companies. Those agreements cover electricity supply, substation design and the regulatory work behind both. None of them is binding. None named a connection date. But the power talks happened in public, in the same week as the buildings.

It sits between three continents. Saudi Arabia is close to Europe, Asia and Africa at once. The subsea cables that carry internet traffic between them already pass nearby. One deal at LEAP 2026 made this concrete. Saudi infrastructure investor Vision Invest and the Africa Finance Corporation put $300 million into WIOCC Group. WIOCC builds data centers, land fiber and subsea cable across Africa. Saudi capital is not only building at home. It is buying the routes.

It is a rare multi-vendor market. Saudi platforms now run AMD accelerators, NVIDIA accelerators, Amazon Trainium chips, Groq inference hardware and xAI systems. Few markets run more than one or two at scale. So the Kingdom is a live test. Can buyers really move work from one chip vendor to another? Or does the software hold them in place? The answer matters far beyond Saudi Arabia.

Two hyperscaler cloud regions open there within about a month of each other. Microsoft in November 2026 and AWS in December 2026. For companies that must keep data inside the country, the choice goes from none to two in a single quarter.

There is a difference between money announced and capacity delivered. It is the same difference as between a mortgage pre-approval and a set of keys. A pre-approval tells you a lender is willing. It does not tell you the sale closed, the house exists, or anyone has moved in. That distinction runs through everything below.

Key Takeaways and Observations

The customers named at LEAP 2026 were not only American. NAVER Cloud and BytePlus were named as the first anchor tenants at Khuzam Digital Valley. BytePlus is also placing its first Middle East and Africa public cloud region inside Mobily’s estate. Together AI brings open-model demand from outside the region.

So Saudi Arabia is not simply hosting American hyperscalers behind a local label. It is selling the same shells to Asian platforms as well. That widens the pool of tenants available to fill the megawatts, and it matters more than any single capital figure announced that week.

Announced capacity is far ahead of operating capacity, and the gap is now measurable. HUMAIN launched its NVIDIA cloud with 1.1 megawatts running. Utilization was above 90 percent. The target is 35.1 megawatts by 2027. In the same week the same company put its name to a 250 megawatt phase with AMD and Cisco, a 250 megawatt initial build with center3, a planned 250 megawatt facility with

Together AI, and 100 megawatts at Oxagon. Two separate frameworks carry one gigawatt targets. The live number and the target numbers are three orders of magnitude apart. That is not a criticism. It is the right place to start. The categories that matter are operational load, energized but empty, contracted, under construction, land and power pipeline, and aspiration.

One company now sits in the middle of nearly everything. HUMAIN signed multiple agreements at LEAP 2026 (AMD and Cisco venture, NVIDIA cloud, Together AI partnership, center3 capacity, Oxagon phase, AWS AI zone, proposed BSF Capital fund, national grid agreement. xAI deployment). HUMAIN buys silicon, develops sites, gathers demand, distributes cloud and arranges power. The upside is speed. One party can move land, chips, customers and electricity together. The risk is concentration. A single state-backed orchestrator carries the execution risk for the whole program.

Power moved from subtext to signed paper, but not yet to firm delivery. Three agreements addressed electricity directly. National Grid SA agreed with HUMAIN on supply for a Riyadh project. The Saudi Electricity Project Development Company agreed with center3 on feasibility studies and substation design. The same developer signed a memorandum with Huawei on energy solutions.

Financing is starting to move off sponsor balance sheets. Two items point the same way. The proposed BSF Capital vehicle would bring outside debt and equity into 250 megawatts of Saudi capacity. It could scale toward one gigawatt. It still needs regulatory approval. The WIOCC subscription is a signed equity investment by two institutional investors into a working African platform. Neither is large by global standards. Both suggest the market is preparing for outside capital rather than state funding alone.

The open question is whether Saudi AI capacity can pass the tests project finance normally applies. Those tests are contracted revenue, bankable power, fixed-price construction, and some protection against accelerators losing their value.

The most valuable disclosures at LEAP 2026 were dates, not dollars. A launch month for Microsoft. A launch month for AWS. A construction start at Oxagon with first capacity in 2028. AMD capacity from the second half of 2027. Utilization above 90 percent on a live cluster. Groq reported racking 19,000 inference chips in under seven days. These are checkable claims with deadlines attached. That makes them more useful than any capital figure announced the same week.

What did not appear is also information. Google and Oracle disclosed no new Saudi data center build, megawatt commitment or infrastructure investment at LEAP 2026. Qualcomm and HUMAIN launched an AI personal computer, which is an edge device rather than infrastructure. Absence of an announcement is not absence of activity. But for a market promoted this heavily, silence from two major cloud providers is worth noting.

The Data Center and Cloud Developments Announced at LEAP 2026

Below are the infrastructure and cloud developments disclosed at the event. They are ordered by how much they change the delivery picture, not by headline value. Software, device and skills announcements are excluded.

1. AMD, Cisco and HUMAIN put production compute in service and set a 250 megawatt next phase

AMD Instinct MI355X accelerators, AMD EPYC processors and Cisco Silicon One networking are live in Saudi Arabia and serving customers. The companies call it AMD’s largest active inference cluster outside the United States. A further 250 megawatts starts in 2027, using MI400-series accelerators. Capacity is expected from the second half of that year. The joint venture target remains one gigawatt by 2030. No new capital figure was disclosed. [Read here]

2. AWS confirms a December 2026 Saudi region and a 50 megawatt AI zone

AWS said its first Saudi infrastructure region remains on track for December 2026. The region represents more than $5.3 billion, or SAR19.88 billion, of planned investment. AWS and HUMAIN also committed to deliver up to 50 megawatts in the first Saudi AI zone by 2028, combining AWS Trainium chips with NVIDIA hardware. HUMAIN’s ALLAM model is slated for Amazon Bedrock, and HUMAIN Fabric for AWS Marketplace. The launch month and the 50 megawatt target are the new information. [Read here]

3. Microsoft dates its Saudi Arabia East region to November 2026

Microsoft confirmed that Saudi Arabia East becomes available in November 2026. The region sits in the Eastern Province and holds three Azure availability zones. It offers in-country data residency, high availability and lower latency. Construction across all three sites finished in 2024. This is a launch date, not a new build. No megawatt or capital figure was disclosed. [Read here]

4. HUMAIN launches an NVIDIA-powered cloud at 1.1 megawatts and above 90 percent utilization

HUMAIN brought a locally hosted AI cloud into service on NVIDIA HGX B300 Blackwell Ultra systems. It reported 1.1 megawatts of operational infrastructure and utilization above 90 percent. The target is 35.1 megawatts by 2027. The Saudi Ministry of Commerce was named as a customer. This is the clearest operating data point from the event. It also sets the honest baseline against every gigawatt target announced beside it. [Read here]

5. HUMAIN and DataVolt start construction at Oxagon in NEOM

HUMAIN will develop 100 megawatts inside DataVolt’s 360 megawatt first phase at Oxagon. Construction is underway and the first 100 megawatts is anticipated in 2028. The phase sits within a planned 1.5 gigawatt campus designed around renewable power, advanced cooling, high-density racks and subsea connectivity to Europe and Africa. The first phase was previously described as 300 megawatts, so the scope has grown. [Read here]

6. HUMAIN and Together AI plan a 250 megawatt facility tied to global demand

Together AI and HUMAIN announced a partnership. It includes a new 250 megawatt Saudi data center for training and inference. It also includes 55 megawatts of dedicated capacity. The companies forecast more than $5 billion of gross annualized revenue in the first year. That figure is projected revenue. It is not capital spend, equity, debt or contracted backlog. No site, power source, delivery schedule or financing structure was disclosed. [Read here]

7. HUMAIN, stc group and center3 set an initial 250 megawatts with a path to one gigawatt

HUMAIN and stc group deepened cooperation across data centers, connectivity and sales. At the center of it is an initial 250 megawatts of AI-ready capacity, with a longer path toward one gigawatt. center3 brings Saudi colocation and subsea reach. stc brings enterprise distribution. HUMAIN brings demand and compute relationships. The underlying joint venture predates the event. This is a framework and route-to-market update. [Read here]

8. Al Moammar Information Systems commits $1.2 billion toward 192 megawatts

Al Moammar Information Systems announced $1.2 billion to expand its Saudi data centers to 192 megawatts of total operational capacity. This is one of the few genuinely new capital commitments from the event. The official summary gave no locations and no commissioning schedule. It did not split existing megawatts from incremental ones. It did not disclose the mix of debt and equity. [Read here]

9. HUMAIN targets $2.5 billion for a BSF Capital data center fund

HUMAIN was reported to be seeking an initial $2.5 billion from domestic and international investors. The fund would be managed by BSF Capital. It would finance 250 megawatts being built with Al Moammar Information Systems, and could later scale toward one gigawatt, using both debt and equity. It needs regulatory approval first. This is a fundraising plan, not closed capital, and it may cover the same assets as the $1.2 billion program above. [Read here]

10. NHC Innovation launches Khuzam Digital Valley with NAVER Cloud and BytePlus as anchors

NHC Innovation announced a phased data center development at Khuzam Digital Valley. The investment is SAR3.3 billion, or roughly $880 million, and the site scales to 65 megawatts by 2033. NAVER Cloud and BytePlus were named as initial anchor customers. Both are East Asian platforms rather than American hyperscalers, which is the notable part. [Read here]

11. Mobily and BytePlus fund the first BytePlus public cloud region in the Middle East and Africa

Mobily and BytePlus announced a joint investment of $120 million to $150 million. It funds the first BytePlus public cloud region in the Middle East and Africa. The platform combines Mobily’s Saudi data center and network estate with BytePlus cloud, AI and analytics technology. No megawatt figure or ownership split was disclosed. [Read here]

12. WIOCC secures $300 million from Africa Finance Corporation and Vision Invest

WIOCC Group signed a shareholder subscription agreement. The Africa Finance Corporation and Saudi infrastructure investor Vision Invest will provide a combined $300 million. The money is meant to speed up African data center deployment. It also funds terrestrial fiber expansion and subsea network development. The investor split, the valuation and the resulting ownership were not disclosed. Vision Invest is a private Saudi investor. It is not a Public Investment Fund vehicle. [Read here]

13. Saudi power entities sign three enabling agreements with data center developers

Three agreements addressed electricity. National Grid SA is the national grid operator. It agreed with HUMAIN on supply for a Riyadh AI data center project and for future needs. The Saudi Electricity Project Development Company agreed with center3 on early power feasibility studies. That work also covers substation design and regulatory preparation. The same developer signed a separate memorandum with Huawei. That one covers data center projects, digital infrastructure and energy solutions. No value or megawatt figure was disclosed for any of the three. [Read here]

14. alfanar commits $150 million to manufacture data center components locally

alfanar announced $150 million to manufacture data center components inside Saudi Arabia. HPE, Intel and the Ministry of Communications and Information Technology launched a local production initiative alongside it. HPE added storage systems made in the Kingdom. It also formalized alfanar’s role in assembling ready-to-run infrastructure. HPE and Intel announced plans for a Saudi innovation center. This addresses supply chain lead times. It does not add operating capacity. [Read here]

15. Zain KSA and DataVolt sign a nonbinding capacity memorandum

Zain KSA and DataVolt signed a nonbinding memorandum of understanding. The two will explore Saudi data center capacity together. Target users are Zain itself, government bodies, enterprises and cloud providers. The sites would carry high-density AI workloads and use energy-efficient cooling. No capital figure, megawatt figure, site or date was disclosed. The deal widens DataVolt’s pipeline beyond Oxagon. It is not yet an investable commitment. [Read here]

What LEAP 2026 leaves behind is a set of dated commitments that can be checked. Two cloud regions arrive in November and December 2026. AMD capacity follows from the second half of 2027, along with 35.1 megawatts of NVIDIA capacity. AWS delivers up to 50 megawatts and Oxagon delivers its first 100 megawatts by 2028. Khuzam reaches 65 megawatts by 2033.

For the Gulf, the signal is a change in what the market competes on. The contest is no longer announced ambition. It is connected power, named anchor customers, and financing that does not sit on a sponsor balance sheet. Grid planning now surfaces in the same week as the buildings. That is the sequence every power-constrained market has been trying to reach, and this one reached it in public.

Three things travel further than the region. Saudi Arabia runs AMD, NVIDIA, Trainium, Groq and xAI hardware side by side, which makes it a live test of whether AI workloads move between chip vendors or stay put. Its anchor tenants are American and Asian at once, which widens the pool of customers available to any market selling data residency. And Saudi capital is buying African fiber and subsea routes, tying Gulf compute to the traffic that would use it.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

In this week’s issue:

First, SB Energy filed to go public. The company holds 8.8GW of contracted capacity, 803MW under construction, and no operating data center. It issued about 4.0 million warrants to OpenAI at one cent and carries a $105 billion Nvidia residual-value guarantee. So, what is an investor actually buying when the tenant and the supplier both sit in the equity?

Then, the Nordics closed the cleanest deal of the week. CPP Investments and Equinix completed the $4 billion acquisition of atNorth, backed by a $4.1 billion financing package underwritten by European and Canadian lenders.

Finally, Tokyo repriced. Keppel DC REIT is paying JPY 168.4 billion for 88.62% of two Inzai City data centers, at a 2.1% discount to valuation.

Let us get into it.

When the Tenant Owns Part of the Landlord

The anchor tenant is now paid in equity, and that payment reaches the accounts before the rent does. SB Energy proved it on 1 September.

The company filed a Form S-1 to list on the Nasdaq Global Select Market and Nasdaq Texas under the ticker SBE.

It holds 8.8GW of contracted capacity and 803MW under construction. Not one data center is operational.

First data center revenue is expected in the fourth quarter of this year. To secure OpenAI as anchor tenant, it issued about 4.0 million warrants exercisable at one cent, with a board designation right for as long as OpenAI holds above 5%.

That warrant liability was reportedly worth $3.6 billion when issued in January and roughly $5.5 billion on 30 June, per draft documents reviewed by the Wall Street Journal.

First-half net loss came in at approximately $3.2 billion against about $140 million of renewables revenue. The tenant incentive is the largest single line in the loss.

So here is the discipline. Before backing an AI infrastructure landlord, ask what it paid its anchor tenant and whether that payment sits in rent or equity. If it sits in equity, the tenant’s upside and reported earnings move in opposite directions.

What the warrants cost, and what atNorth did instead, are below.

THIS WEEK BY REGION

The week’s biggest moves — what happened and what it signals.

North America

The supplier moved inside the capital structure twice this week. Anthropic signed a six-year, $35 billion cloud deal with Lambda, backed by Nvidia’s lease of the Hut 8 site and investment in Lambda. Lambda won the contract without sourcing or leasing the underlying facility.

SB Energy filed to list on Nasdaq under SBE with 8.8GW of contracted capacity but no operating data center. Vertiv agreed to acquire UtilityInnovation Group for $1.45 billion, plus up to $1.15 billion tied to earnings. Crusoe reportedly raised $3 billion at a $30 billion valuation.

Public markets equity investors get the first clean look at this structure when SB Energy prices, because the prospectus fixes numbers that reporting has only estimated.

Europe

The Nordics ran the week’s only conventional capital structure. CPP Investments and Equinix acquired atNorth for $4 billion, taking 51% and 34%, while Partners Group retained about 10%. A $4.1 billion lender-backed financing package supports the deal.

Cerebras began construction of a 165MW Finland campus, with each phase contracted before buildout. Meanwhile, UK planning faced opposition, including petitions against projects in Reading, Wales, and Scotland.

For infrastructure funds, the Nordic bid clears while the UK queue lengthens, and that gap is now a pricing input rather than a footnote.

Asia-Pacific

Japan repriced on rent, not scarcity. Keppel DC REIT is paying JPY 168.4 billion for an 88.62% interest, with JPY 190 billion total consideration at a 2.1% discount. Rents are at least 30% below market, making reversion the key return driver.

Bitdeer AI sold out its 9.5MW Malaysia facility, with expected AI cloud revenue above $800 million. Firmus signed a trans-Pacific deal with SUBCO and proposed a third Tasmania project, both without disclosed terms.

Real estate investors should read the Keppel entry price against the rent gap, because a 2.1% discount is not what makes that deal work.

South America

Brazil set the terms before the capital arrived. Brazil’s special data center regime was approved and now goes to the president. It suspends federal taxes on equipment for five years, against conditions on energy, water and domestic capacity.

Developers are moving on that. Yamna reserved land at the Port of Açu for a 250MW first phase, with no offtaker named and no capital disclosed.

Project finance institutions should watch the sequence, because a tax regime landing before the offtake decides which projects reach financial close.

PUBLISHED THIS WEEK

• NVIDIA Did Not Buy Powered Land. It Bought The Design Standard. — The design standard lands during civil engineering, so a site built to one accelerator carries a redesign cost rather than a procurement cost.

• KKR Is Not Behind MTN’s Africa Data Hub. Tarek Al Ashram Is. — Al Ashram funds the venture through his own investment firm, so the KKR-backed Gulf Data Hub balance sheet does not stand behind it.

• What $725 Billion Cannot Buy (And Why the AI Buildout Now Waits on Electricians) — The schedule constraint is the electrical trade, and no volume of hyperscaler capex shortens an apprenticeship.

• What a Data Center Actually Costs: CapEx Breakdown and the Drivers That Move It — Power and cooling capex shares, redundancy tier multipliers, air versus liquid economics, geographic cost spread, equipment lead times, cost per megawatt.

Sponsored By: Global Data Center Hub

Your trusted source for global AI infrastructure analysis. Every week, we break down the trends shaping the future of data centers, cloud, and AI, from emerging markets to hyperscale battlegrounds.

📰 Subscribe to get expert insights delivered to your inbox.

What the Warrants Cost

SB Energy paid for its anchor tenant in equity, and the filing shows the price. OpenAI holds about 4.0 million warrants exercisable at one cent, plus the right to designate a board member while it holds above 5%.

The vesting schedule says what the tenant is being paid for. One tranche of 408,942 vests on the conditions precedent to initial project financing for at least 753 MW-IT at Milam County. Five further tranches of 369,000 each vest only at equity values of $80 billion, $100 billion, $125 billion, $150 billion and $200 billion. Five of six tranches turn on the share price rather than on the space.

That liability was reportedly worth $3.6 billion at issuance in January and roughly $5.5 billion at 30 June, per draft documents reviewed by the Wall Street Journal. First-half net loss came in at approximately $3.2 billion against about $140 million of renewables revenue. The company holds 8.8GW of contracted capacity and 803MW under construction. Nothing is operational, and first data center revenue is expected in the fourth quarter.

Nvidia occupies the other side of the same structure. It committed $1.5 billion of non-voting Class N shares at the offering price, on top of a $1.5 billion prepaid forward entered on 17 August. It also guaranteed residual value across roughly 4.25GW of initial PORTS-Pike capacity, at an aggregate $105 billion, conditioned on Nvidia compute being the only compute deployed at that site.

Run the two questions and the second one fails. The first answers cleanly, because the payment to the tenant is disclosed, marked to market, and sitting in the loss. How long the comfort has to hold does not answer. The exclusivity condition on the residual-value guarantee carries no disclosed term, so a reader cannot tell how long the company is locked out of switching silicon. A missing disclosure is a finding.

So the document I am waiting for is the pricing prospectus on EDGAR. It fixes the per-share target behind every warrant tranche, and until it lands the largest liability on this balance sheet has no strike anyone outside the company can test.

NOTABLE TRANSACTIONS

Key structures and capital moves from this week’s deal tape.

Vertiv: approximately $1.45 billion cash for UtilityInnovation Group, with up to $1.15 billion contingent

Vertiv agreed to acquire Utility Innovation Holdings for $1.45 billion, with up to $1.15 billion in earnouts tied to EBITDA targets, shifting execution risk to the seller while securing behind-the-meter capabilities.

If you operate at scale, the mechanism is that speed to power is now bought as a product line, and the queue you were waiting in is what the acquirer is pricing.

Keppel DC REIT: JPY 168.4 billion for 88.62% of two Greater Tokyo data centers

Keppel DC REIT and Keppel are acquiring a 90% effective interest in Tokyo Data Centres 4 and 5 for JPY 190 billion, a 2.1% discount to valuation, funded largely by a S$625 million private placement.

For a listed landlord, the mechanism is rent reversion rather than entry discount, and in-place rents at least 30% below market are where that return actually lives.

atNorth: $4 billion acquisition backed by a $4.1 billion underwritten financing package

CPP Investments acquired 51% for $1.3 billion and Equinix 34% for $895 million, while Partners Group retained 10% for $260 million. European and Canadian lenders underwrote $4.1 billion for the transaction and expansion capital.

If you allocate to infrastructure, the mechanism is ordinary bank leverage against operating assets, and it is worth noting how unusual that has become in a week like this one.

If you are new to Global Data Center Hub, visit our Start Here page for more information about the newsletter.

If you found this newsletter useful, share it with a colleague.

Have a great week.

— Obinna

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

This article is the 12th article in the series: Ground to Grid: A Free 21-Lesson Guide to Mastering Data Center Development

Join Global Data Center Hub

Most people assume the servers are the cost. For the developer, the lender, and the infrastructure investor, they are not. The servers sit on the tenant’s balance sheet.

What gets financed as a data center project is power, cooling, and the shell that protects them. The drivers behind those three decide whether the project pencils.

The distinction is structural. A facility amortizes over 15 to 30 years. The GPU payload inside it depreciates over three to five.

In a 20-megawatt AI hall, the hardware can cost $500 million to $1 billion, compared with $200 million to $300 million for the building. The hardware is the larger number. It is not the number in the project finance package.

This is the stage of development where a design becomes a budget someone will lend against. The headline figure matters less than the drivers underneath it.

Where the capital actually goes

A standard air-cooled Tier III facility delivers at $8 million to $14 million per critical IT megawatt in 2026. A fully fitted direct-to-chip liquid facility delivers at $17 million to $25 million.

Global shell-and-core averaged $11.3 million per megawatt this year, up from $7.7 million in 2020.

Power infrastructure accounts for 40 to 45 percent of hard construction costs.

Substation, medium-voltage switchgear, generators, UPS arrays, distribution units, and busway.

This is the category that scales directly with capacity.

Interconnection often sits outside the headline benchmark and rarely stays small.

Core substation equipment for a large campus costs $5 million to $12 million, but switchyards, transmission extensions, protection relays, and transformer bays push the total interconnection cost to $25 million to $60 million.

In PJM, ERCOT, and CAISO, developers also fund off-site network upgrades and post queue deposits reaching $4,000 per megawatt before construction starts.

Cooling takes 15 to 20 percent, and the share is rising.

Chillers, cooling towers, pumping skids, air handlers, and coolant distribution units.

As racks shift to direct-to-chip liquid, mechanical systems' share of the build climbs while electrical and shell both fall.

Shell and site take 12 to 15 percent, with another 4 to 7 percent for site preparation.

Foundations, tilt-up walls, and floor slabs rated for 2,500 to 5,000 pounds per square foot. Land is the variable inside this line.

An acre in Loudoun County or Santa Clara now costs $2.5 million to $4.5 million, compared with $100,000 to $300,000 in Ohio, Indiana, or the Carolinas.

Fit-out, soft costs, and commissioning absorb the remainder.

Design and permitting take 8 to 12 percent. White space containment takes 5 to 8 percent. Fire suppression and security take 2 to 3 percent.

Commissioning takes 1 to 3 percent and moves through five stages, ending with a black-start test that drops utility power and confirms the whole plant carries a simulated load.

The drivers that move the number

Redundancy tier. Tier III delivers concurrent maintainability and 99.982 percent availability, which caps unplanned downtime at 1.6 hours a year.

Tier IV adds full fault tolerance and costs 25 to 40 percent more for identical IT capacity because every path doubles to physically separated rooms.

Hyperscalers do not buy it. They build Tier III and handle software failures across availability zones.

Power density. A GB200 NVL72 rack draws 120 to 140 kilowatts. Next-generation silicon is projected at 300 to 600 kilowatts.

Density sets the cooling architecture, and cooling architecture sets the electrical scale behind it.

Cooling choice. Turner & Townsend puts the construction premium for liquid-cooled capacity at 7-10 percent.

The rack-level hardware carries more: coolant distribution units, manifolds, and cold plates add $30,000 to $75,000 per rack.

Liquid buys some of that back by cutting white space 70 to 85 percent per megawatt, but busway, stainless piping, leak detection, and slab reinforcement consume the saving.

Location. Tokyo builds at $14.30 per watt. Mumbai builds at $6.60. Union trade labor runs $140 to over $200 an hour in New York, Chicago, and San Francisco against $65 to $95 in right-to-work states.

Seismic zones add 8 to 15 percent to structural and mechanical budgets.

Market timing. Large power transformer lead times have grown from 30–45 weeks before 2021 to 120–160 weeks today, with a 30% structural supply deficit and costs up 45–95%.

Reservations now require 15–30% upfront, while schedules have stretched to 24–36 months or more.

On a $1.2 billion campus, a 12-month delay can add roughly $96 million in capitalized interest, pushing contingencies from 5–8% toward 12–20%.

Why cost is measured differently now

Cost per square foot is a real estate metric. It describes floor area, and floor area no longer describes what the asset produces.

Revenue, lease commitments, and valuation all track energized capacity. Cost per critical IT megawatt is the appropriate unit of comparison.

As density rises, cost per rack and cost per unit of compute become the finer measures, because a 120-kilowatt rack and a 10-kilowatt rack occupy similar floor space but represent entirely different capital.

The same nominal facility can carry a 3x cost range depending on tier and density alone.

Which direction each driver pushes

What this means for each party

For developers, tier and density are choices; the equipment market isn’t. Forward procurement is the only step you can take before signing a lease, but it requires 15–30% of equipment value upfront. Commit too early, and you risk leaving gear unused; wait, and you may miss your delivery date.

For investors, ask about the tier, density, market, and whether transformer and switchgear orders have been placed. Two builds at the same cost per megawatt can carry very different risk, often revealed by the completion date. A 5–8% contingency may signal the sponsor hasn’t repriced for today’s equipment market.

For operators, these choices shape costs for the asset’s life. Industry-average PUE is 1.54, while leading sites reach 1.10–1.20. For a 50 MW facility, that gap can mean roughly $10 million more in annual power costs. Electricity accounts for 40–60% of opex, potentially matching or exceeding the original build cost within five years.

For policymakers, equipment supply is a constraint that incentives cannot bypass. Lagos costs $12.00 per watt, compared with $11.20 in London, largely due to limited domestic manufacturing. Tax breaks reduce costs at the margin; local manufacturing lowers them structurally.

Which cost driver do you think is most underestimated in early budgets, and which one has surprised you most on a real project?

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Amazon, Microsoft, Alphabet and Meta plan to spend roughly $725 billion building AI infrastructure in 2026, and the shortage now slowing them down is electricians. Microsoft president Brad Smith described the electrician shortage in March 2026 as the number one problem slowing the company’s data center expansion in the United States. The Financial Times compiled the $725 billion figure from the four companies’ first-quarter earnings guidance, up about 77 percent from roughly $410 billion in 2025.

• Electrical work accounts for 45 to 70 percent of what it costs to build a data center, according to the International Brotherhood of Electrical Workers. A data center is largely an electrical system with a roof over it, since the path power takes from the grid to the computers is the expensive part of the building. A shortage in that one trade therefore sets the schedule for everything else on the site.

• A licensed electrician takes four to five years to train, which is longer than most data centers take to plan and build. The standard apprenticeship run by the electrical training ALLIANCE, the training body created by the IBEW and the National Electrical Contractors Association, requires a minimum of 8,000 hours of supervised work on real job sites alongside classroom instruction. Wages can rise overnight, and that clock runs at its own pace.

Somebody wired the room you are sitting in. They pulled cable through the walls, landed it in a panel, and left.

The largest construction program in modern American history now runs on that same trade, and it has run short of people.

This is the part of the AI story that gets almost no coverage, because it does not sound like technology.

Everything you read about the buildout is about chips, models, or money.

The thing actually holding up job sites is a person with a license, a van, and four to five years of training behind them, and there are not enough of them to go around.

That matters because the pace of AI, electricity prices, and career opportunities for the next generation increasingly depend on how many electricians America can produce. None of that is decided by a chip.

Here is how the money ran into the workforce.

Three quarters of a trillion dollars is already committed

Amazon, Alphabet, Meta and Microsoft together plan to spend roughly $725 billion on capital projects in 2026, according to the Financial Times tally of their first-quarter earnings guidance.

That is up about 77 percent from roughly $410 billion in 2025, and several of the four have raised their guidance again since.

Capital expenditure, or capex, is simply money spent on physical things a company will own for years.

Buildings. Machines. In this case, data centers and the equipment inside them.

Most coverage treats that number as a schedule.

It reads as though $725 billion buys a certain quantity of finished buildings on a certain date, the way a grocery bill buys groceries.

Read it that way and you have mistaken a budget for a delivery plan.

Microsoft is driving electricians in from 75 miles away

In March 2026, Microsoft president Brad Smith said the shortage of electrical talent was the biggest constraint on U.S. data center expansion, ahead of chips and land.

He also described what the company does about it.

Microsoft has been bringing electricians to job sites from more than 75 miles away, and in some cases relocating them temporarily, to keep work moving.

Google’s own policy work says the same thing in flatter language:

“A shortage of electrical workers may constrain America’s ability to build the infrastructure needed to support AI”.

When the two largest builders of data center infrastructure both name the same trade as the thing in their way, the shortage has stopped being a human resources problem.

If you have been reading construction delays as a sign that demand is softening, this is the alternative explanation, and it is the one the builders themselves give.

Nearly half the cost of a data center is electrical work

Electrical systems account for somewhere between 45 and 70 percent of the total cost of building a data center, according to the IBEW.

That figure surprises people, and it is the key to the whole story.

A data center is best understood as an enormous electrical machine. The concrete is there mostly to keep the weather off it.

Think of the electrical panel in your hallway, the gray box with the breakers in it.

A large data center campus is that idea at a scale most people never see. Switchgear the size of shipping containers.

Cable too thick for a person to bend by hand. Backup generators the size of a bus.

Thousands of connections that each have to be made correctly by someone licensed to make them.

All of it is done by hand. The machine that pulls cable and lands terminations has yet to be invented.

So when the trade that does it runs short, the shortage sets the pace of the whole building.

A workforce problem then shows up on a schedule exactly the way a supply problem does.

Wages can rise tonight. An apprenticeship takes five years.

This is the part that money cannot solve quickly, and it is worth being precise about why.

The main path is a registered apprenticeship through the electrical training ALLIANCE, jointly run by IBEW and NECA.

The inside wireman program for commercial, industrial, and data center work requires 8,000 hours of fieldwork plus classroom training, typically over four to five years.

Apprentices are paid, and tuition is generally free.

The pipeline is real, and it is small relative to what has been asked of it.

The Bureau of Labor Statistics projects around 81,000 electrician openings a year over the coming decade, against a workforce of roughly 819,000, with median pay of $62,350 as of May 2024.

Now set the demand beside it. Associated Builders and Contractors estimates the construction industry needs about 349,000 net new workers in 2026 alone.

The Information Technology and Innovation Foundation has put the shortfall specific to data center construction at roughly 439,000 people.

A company can raise a wage on a Tuesday afternoon. The first cohort hired into an apprenticeship this month becomes licensed around 2031.

If you have ever wondered why an industry with unlimited money still misses its dates, this is the mechanism, and it is the one part of the chain that no cheque shortens.

Google is now paying to train electricians in Indiana

There is a straightforward test for whether a company believes a constraint is real: watch whether it spends money on something far outside its own business.

Google.org committed $15 million to the electrical training ALLIANCE to reach 100,000 electrical workers and 30,000 new apprentices.

In September 2025, $400,000 was awarded to three Indiana training centers, aimed at adding 524 apprentices by 2030 a 42% increase in the state’s projected electrician pipeline.

Consider the two numbers together. A company built on search and cloud software is funding trade apprenticeships in the Midwest, yet the state-level result is just 524 people over five years against a national gap of hundreds of thousands.

Both facts are true at once, and the second one does not cancel the first.

The scale of the response tells you the problem is genuine. The scale of the gap tells you it is not going to be closed by philanthropy.

Read the training announcements as a fix and you will be surprised in 2028. Read them as a measurement of the problem and they make sense.

The crew wiring the data center was going to wire something else

This is where the shortage leaves the sector and arrives in your town.

Turner & Townsend surveyed 112 markets across 44 countries for its Global Construction Market Intelligence 2026 report.

It found data center construction to be the most capacity-constrained building sector in the world.

More than 70 percent of those markets reported contractor capacity as tightening or overstretched.

Around 87 percent reported shortages in the mechanical, electrical and plumbing trades, and roughly 71 percent of markets named labor shortages as the leading driver of rising costs.

The report describes a two-speed market.

Contractors and suppliers are moving toward data center work because the margins are better, while housing, hospitality and ordinary commercial developers reassess whether their projects still make financial sense.

That is the honest cost of this buildout, and it is a real one. There is one pool of licensed electricians in any given region.

When a hyperscale campus arrives and can pay above the local rate, the crews that would have wired apartments, schools and hospitals go to the campus instead.

The projects they leave behind get slower and more expensive, and the people who notice are not reading data center coverage.

The AI buildout is happening inside the same economy you live in, competing for the same tradespeople.

Oracle’s completion dates moved a year to the right

Schedules are already moving, and the reporting on why is worth reading carefully.

Oracle is building data center capacity for OpenAI.

Bloomberg reported that the company moved the construction completion date from 2027 to 2028 and cited labor shortages as one reason.

Oracle disputed that account. It told Fortune that its projects remain on schedule and on plan, and that it intends to invest in local workforce training so nearby residents can do the work.

What is not in dispute is the underlying arithmetic, which every builder in the sector faces regardless of whose schedule slips first.

A single large campus can need on the order of a thousand or more workers at peak construction.

Several of them are being built at once, in the same states, drawing from the same halls.

The temptation is to read a delayed data center as evidence that AI demand was overstated.

Sometimes that is what a delay means. In this case there is a simpler explanation available first, and it is that the crew has not finished the last one.

What this actually means

The most heavily funded construction program in the country has run into a limit made of people.

That limit is different from the sector’s usual constraints.

Transformers arrive and grid connections clear, but electricians must be recruited, trained for years, and supervised by qualified workers, meaning each new electrician requires time from an existing one.

There is another point that complicates the good news: construction jobs are large in number but temporary.

A campus may employ over 1,000 workers during construction but only a few dozen once operational. As Louisiana union officials note, the work arrives in a wave and eventually moves on.

Communities should recognize that construction booms have an end date.

What does not have an end date is the reversal underneath all of this. For thirty years the advice was to get a degree and a desk.

The companies that automated the desk are now unable to build their own future fast enough because there are not enough people who can pull wire.

The AI you use answers in a second. Behind it is a building somebody had to wire by hand, and the pace of the whole thing is set by how many hands there are.

Did anyone you know go into the trades in the last few years? What made them pick it?

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Meta expanded its Richland Parish campus in Louisiana to 5GW of compute capacity, the largest capacity figure among July’s five regional anchor deals. Meta is recorded as its own anchor customer, and the campus is under construction rather than proposed. Committed investment exceeds $50 billion, with a further sum for roads, water and wastewater held outside that figure.

• SK Telecom committed $91.5 billion, the equivalent of KRW140 trillion, to more than 2GW of AI data center capacity across Yeongnam in South Korea. SK Telecom describes the programme as the first major stage of a national platform targeting 15GW by 2035. No anchor customer is disclosed, and investment size, timing and financing structure remain subject to refinement.

• Nscale and Nordkraft formed Nordscale Operations AS to run the 230MW Narvik campus, the only one of July’s five regional anchor deals that names a generation source. Nscale Norway holds 51 percent of the operating company and Nordkraft holds 49 percent. The underlying Kvandal facility is designed on hydropower, with OpenAI recorded as the initial offtaker.

Top 5 Deals — Deep Dives

Five transactions anchor the month, with Entergy’s power arrangement among the most significant. Its supply portfolio combines new natural gas generation, nuclear uprates, batteries and purchased electricity to serve large, steady loads without relying on a single technology. The nuclear up rates are particularly important, adding firm capacity without new plant siting. Taken together they describe five different answers to the same question, which is how you get compute built where the power actually is.

Meta Richland Parish 5GW Expansion (North America)

Meta expanded the scale and the investment commitment of its Louisiana campus this month, taking Richland Parish to 5GW of compute capacity with more than $50 billion committed.

The status matters as much as the number. This is recorded as under construction, not proposed, which separates it from most of what sits above 1GW in the July log.

The power arrangement is the key feature.

Entergy’s supply portfolio combines new natural gas generation, nuclear uprates, batteries and purchased electricity to serve a large, steady load without relying on a single technology.

That is why the tracker records the source as grid rather than a specific fuel.

The nuclear uprates are particularly important because they add firm capacity without requiring new plant siting, a meaningful advantage for a 5GW load.

On capital structure, Meta is its own anchor customer.

There is no third-party tenant, no lease, no offtake contract to underwrite.

The company is funding the campus against its own demand, and the more than $1 billion of roads, water and wastewater work sits outside the headline figure.

That combination, self-anchored demand and a regulated utility building the supply portfolio around it, is available to a very short list of companies.

What Richland Parish reveals about North America is that the region’s largest credible capacity now moves through utility planning rather than through developer speculation.

The campus is not competing for grid headroom that already exists. Entergy is building the headroom to fit it.

Nscale–Nordkraft Nordscale Operations, Narvik (Europe)

The Narvik transaction is the smallest of the five and the most instructive.

Nscale and Nordkraft created Nordscale Operations AS, a jointly owned company that will manage Nscale’s AI data center facilities in the Narvik region, with Nscale Norway holding 51 percent and Nordkraft 49 percent.

The company was announced in early July and is expected to begin operations during 2026. No deal value was disclosed.

The underlying asset is a 230MW hydropower-backed campus at Kvandal, with OpenAI as the initial offtaker. A further 290MW of potential expansion is excluded.

The key feature is the structure: Nscale gave regional utility Nordkraft 49 percent of the operating company, aligning the power provider directly with the facility’s performance.

In a European market where interconnection queues and local opposition decide which projects survive, that is a structural move rather than an administrative one.

Nscale did not buy power. It bought alignment with the entity that supplies it, and it paid in ownership.

What Narvik reveals about Europe is that the projects clearing the bar are the ones that solved power before they solved anything else.

It is also the only anchor deal this month that names its fuel, which says something uncomfortable about the disclosure standard everywhere else.

SK Telecom $91.5 Billion Yeongnam Buildout (APAC)

SK Telecom outlined a plan to build more than 2GW of AI data center capacity across Yeongnam, with $91.5 billion attached, the equivalent of KRW140 trillion.

The programme starts with a 100MW facility in Ulsan, adds a 900MW expansion and then a further 1GW elsewhere in the region.

SK Telecom frames it as the first major stage of a national platform targeting 15GW by 2035.

The $91.5 billion capital figure requires context. Against 2GW of disclosed capacity, it implies roughly $45.75 million per MW, well above typical hyperscale construction costs.

The gap reflects both understated capacity and substantial spending on GPUs, high-bandwidth memory and other AI infrastructure beyond the facility shell.

The 5GW and 15GW targets are longer-term portfolio goals and are not directly tied to this capital figure.

No anchor customer is disclosed, and no power source is disclosed. For a programme of this size in a grid-constrained country, the second omission is the more significant one.

What Yeongnam reveals about APAC is that the region’s largest commitments are now national in framing and corporate in execution.

A telecoms operator is carrying a state-scale compute target on its own capital plan, and the announcement arrived before the power arrangement did.

Trinidad and Tobago–EY 300MW Development MOU (Latin America)

Trinidad and Tobago signed a development memorandum with EY and third-party developers for a 300MW data center project as part of a broader digital infrastructure push.

The 300MW represents peak power capacity rather than critical IT load.

A reported investment of more than $5 billion spans three national initiatives and cannot be attributed to this project, while its power source remains undisclosed.

The government also signed a separate memorandum with Hummingbird AI Holdings LLC on the same date for an initial 150MW with potential expansion to 500MW.

Two agreements, same week, same counterparty on the state side, different partners and different capacities.

An MOU with an advisory firm and unnamed developers is an early instrument. It commits nobody to spend anything.

What it does is signal that the state intends to package land, gas and policy for this use, and in a country with a long-established gas industry that is not an idle signal.

What Trinidad and Tobago reveals about Latin America is how thin the region’s disclosed pipeline is.

Three transactions across the entire month, and the region’s strongest entry carries no capital figure and no power source.

Any read on Latin America from this dataset is a read on three rows, and the report treats it at that weight rather than inflating it to match the other regions.

Crusoe $10 Billion Israel Lease Expansion to 150MW (Middle East & Africa)

Crusoe was reported to be expanding its leased footprint in Israel to approximately 150MW, with $10 billions of investment over roughly ten to fifteen years.

The expansion combines an additional 40MW on an existing Anan Group lease with a new 67.6MW agreement with MegaDC, across Afula, the Idan HaNegev Industrial Park and Haifa.

The status is Reported rather than announced, and the reason is specific.

Globes identified Crusoe as the customer, but neither MegaDC nor Crusoe had publicly confirmed the full programme, underscoring the distinction between reported and formally disclosed leased capacity.

The capital structure is unusual for a data center entry of this size, because Crusoe is not building.

It is leasing from third-party operators, and a substantial portion of the $10 billion is expected to fund GPU procurement rather than physical facilities.

That makes the figure incomparable with single-site construction capex and it makes the commitment far more reversible than an owned campus would be.

Power source is not disclosed for any of the sites.

What Israel reveals about the Middle East and Africa is that capacity in this region can now be assembled by contract rather than by construction.

Crusoe is acquiring one of its largest international positions without pouring concrete, which compresses the timeline and leaves the power question with the landlords.

Regional Scorecards

North America

Sixty-one transactions made this the deepest pipeline, led by long-term AI leases and power-backed projects.

TeraWulf, CleanSpark and Hut 8 secured 928MW of commitments, while major projects in Texas and Georgia paired data centers with dedicated power strategies.

Many Virginia and Pennsylvania proposals, however, lacked capacity, capital or power details.

The month was defined by hyperscalers and AI labs converting demand into firm commitments.

Europe

Fifty transactions showed a market dividing between projects with secured power and those still waiting for it.

Pure DC, Pantheon AI and Apto advanced large campuses through power and grid agreements, while others pursued alternatives such as offshore gas and wind.

Prologis’ $18.8 billion SEGRO proposal added a major consolidation signal. Europe’s defining constraint remains power access.

APAC

Thirty-three transactions showed the strongest concentration of state-adjacent ambition, led by major AI infrastructure programs in South Korea.

Australia delivered the region’s most complete power arrangement, while Southeast Asia advanced through large campus partnerships and financing.

India remained sovereign-led but modest in disclosed scale, while China surfaced mainly through early-stage initiatives.

Latin America

Just three transactions made July too thin for a confident regional read.

Trinidad and Tobago produced two government-backed memoranda totaling 450MW, while a nuclear-powered initiative emerged in Argentina and Brazil.

None disclosed capital commitments, and established markets such as Brazil, Chile and Mexico were largely absent. The month was defined by intent rather than execution.

Middle East & Africa

Seven transactions showed a region dominated by sovereign-led development.

The UAE, Saudi Arabia and Côte d’Ivoire advanced national AI and data center initiatives, while Jordan added a gas-backed campus proposal.

Commercial activity was limited, led by Crusoe’s reported Israel expansion and projects in Egypt and South Africa. Capital and power disclosures remained scarce.

Forward Indicators

The next 500MW to 1GW increments will land where secured land, power and demand converge. Few markets currently clear all three.

Texas is the strongest signal, with six independent developers assembling power-first positions, including MARA, Liberty Energy and PowerBridge, QTS, Crusoe, Nexus and Hut 8.

In Europe, Finland, Croatia and Poland stand out.

Seinäjoki has the full stack of permits, power, leasing and financing; Topusko has secured technical approval for 800MW; and Poland’s Gryfin Project offers a credible phased path toward 790MW.

Spain is also building a corridor of pre-secured power positions.

Australia leads APAC because large-scale power is already contracted, while South Korea’s major projects still lack disclosed power arrangements.

The UAE and Saudi Arabia are building sovereign delivery platforms ahead of capacity.

Latin America remains speculative, with Trinidad and Tobago’s memoranda still requiring developers and power contracts to convert.

Investor Takeaways

For institutional investors, July reinforced the separation of ownership from operations.

Meta, Nscale, Orange, TeraWulf and others brought outside capital into infrastructure while retaining operating or customer relationships.

These structures create new entry points, but should be priced on contracted demand, not announced capacity.

For sovereign investors, national ambition is not enough. The strongest projects pair policy support with a named utility or power counterparty from the outset.

For operators, contracted demand is now the entry ticket.

CleanSpark, TeraWulf and Hut 8 converted secured demand into long-term leases and financing, while speculative filings without tenants or power increasingly register as noise.

For utilities and policymakers, Richland Parish offers the clearest model: build new generation around future load rather than ration existing capacity.

Where utilities cannot do this, developers are increasingly pursuing off-grid and behind-the-meter alternatives.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• MTN Digital Infrastructure entered a strategic partnership with a UAE-based data center investment platform on August 27, 2026, operating through a new vehicle called Africa Data Hub Holding Limited. MTN disclosed no capital commitment, no equity split, and no financial close date. Ralph Mupita said phase one targets 150MW in South Africa and Nigeria and that MTN will hold a minority position.

• The party funding Africa Data Hub Holding Limited is Tarek Al Ashram’s own investment firm, not the KKR-backed Gulf Data Hub he founded. Bloomberg reported the distinction on August 27, 2026. Gulf Data Hub operated around 40 megawatts when KKR agreed to acquire a stake in January 2025, and Gulf Data Hub is not a party to the MTN transaction.

• Africa Data Hub Holding Limited has no disclosed jurisdiction, equity split, or committed capital, which leaves a co-investor or lender with nothing to underwrite. MTN contributes land, grid access, licenses, and Bayobab fiber, and takes the minority position. The equity split and the holding company domicile are the disclosures that convert the announcement into a transaction.

Part One: The Record

MTN Group announced on August 27, 2026, that MTN Digital Infrastructure has entered a strategic partnership with a UAE-based data center investment platform to accelerate the development of AI-ready digital infrastructure platforms across Africa.

The announcement disclosed no capital commitment, no equity split, and no financial close date.

The partners will operate through Africa Data Hub Holding Limited, a new vehicle that MTN says will serve as the platform through which future digital infrastructure opportunities are developed and scaled across key African markets, with an initial focus on South Africa and Nigeria.

Bayobab, MTN’s digital connectivity business, is a shareholder in the partnership and will provide open access connectivity and go-to-market support through its pan-African network.

The Partner And His Platform

MTN’s release names the founder of the UAE-based investment platform as Tarek Al Ashram and describes him as an experienced data center developer, operator and investor with an established track record across the Middle East.

The release does not name the platform itself. Al Ashram founded Gulf Data Hub, a data center company established in 2012 and headquartered in Dubai.

KKR announced in January 2025 that funds affiliated with it would acquire a stake in Gulf Data Hub through its Global Infrastructure strategy, and the two parties committed to support more than $5 billion of investment in Gulf data center capacity.

Gulf Data Hub operated around 40 megawatts across the United Arab Emirates and Saudi Arabia at that point.

Bloomberg reported on August 27, 2026, that Al Ashram will back Africa Data Hub Holding Limited through his own investment firm.

Scope, Markets And Capacity

Ralph Mupita, MTN Group chief executive, set out the scope of the first phase at the group’s interim results presentation.

Capacity reported him saying the partnership is looking at 150MW as phase one, in South Africa and Nigeria only, and that the partners will build out as demand requires.

Mupita also said MTN will hold a minority position in the venture and described the group’s approach to AI infrastructure as partner led. No percentage has been published.

Mazen Mroué, chief executive of MTN Group Digital Infrastructure, said the partnership brings investment, infrastructure expertise, connectivity and market access into a single platform capable of supporting hyperscalers, cloud providers, enterprises and technology innovators across Africa.

Part Two: The Read

The Credibility In This Deal Sits On Another Balance Sheet

Gulf Data Hub operated roughly 40 megawatts when KKR agreed to buy into it in January 2025. Africa Data Hub Holding Limited is targeting 150 megawatts in phase one.

Coverage of the announcement has repeatedly set the phrase KKR-backed Gulf Data Hub next to the new venture, while the Bloomberg report those same articles cite says something different: Al Ashram is funding Africa Data Hub through his own investment firm.

The track record carrying this announcement belongs to a company that is not a party to the transaction, and the balance sheet that is a party has not been described at all.

Gulf Data Hub Is Not A Party Here

The MTN release names Al Ashram. It does not name Gulf Data Hub, his investment firm, or KKR.

MTN’s own language describes a person with a track record, not an institution with committed capital.

That is an accurate description of what has been agreed, and it is a much narrower thing than the market is reading. The distinction is not pedantic.

Gulf Data Hub carries an institutional shareholder, a stated multi-billion-dollar investment programme, and an operating platform with revenue and lender relationships.

A founder’s personal investment vehicle carries whatever its founder chooses to put into it, disclosed to nobody.

Those two counterparties price differently in a credit committee, and only one of them has been confirmed here.

Watch for one event over the next four quarters.

If Gulf Data Hub or KKR is brought into Africa Data Hub Holding Limited in any contractual role, the venture becomes financeable on institutional terms almost immediately.

If neither appears, the vehicle has to raise third-party capital on a sponsor whose commitment nobody outside the deal has seen, into two markets where sponsor quality is the first question a lender asks.

MTN Contributes What Capital Cannot Buy And Takes The Minority

MTN brings land, grid interconnection, licenses, an existing data center footprint in both target markets, and through Bayobab the fiber that connects them.

The partner brings money and construction experience. MTN takes the minority position.

In South Africa and Nigeria, the binding constraint on data center development is not capital, it is power and the queue to get it.

MTN is contributing the scarce input and accepting the smaller share of the equity.

It gets paid twice for doing so: once on the minority stake, and once through Bayobab, which is a separate shareholder and sells connectivity into the facilities the venture builds.

The result is a structure that keeps 150 megawatts of construction risk off MTN’s balance sheet in the same month the group reported record margin and started buying back its own shares.

That is a coherent capital allocation policy.

It is also a template every other pan-African operator with land and fiber will now examine, and the ones that move first will trade their grid positions into ownership while grid positions still command a premium.

What Each Segment Prices Before The First Site

Private Capital should underwrite the sponsor rather than the track record.

Before indicative pricing, get four things in writing: the equity split, holding company jurisdiction, partner’s committed capital, and whether Gulf Data Hub has any contractual role.

Firms that price this off the Gulf Data Hub name are paying for a credit that has not been offered to them.

Public Markets holders of MTN are buying a minority interest in an unfunded vehicle, attached to a connectivity contract that is real and already earning. That is a reasonable thing to own.

It is not AI infrastructure exposure in the sense the share price will be asked to reflect, and the gap between those two readings widens every quarter the equity split stays unpublished.

Operators in South Africa and Nigeria now face a competitor with 150 megawatts of stated intent, existing sites in both markets, and its own fiber.

The scarce resource is the grid connection, not the capital.

Operators that have not secured their next interconnection agreement within the next two to three quarters will be negotiating for it against a party that already holds a queue position and a telecom licence in the same market.

The Equity Split Is The Next Real Disclosure

The megawatt number is the least informative thing MTN has said about this partnership.

A 150-megawatt phase one target from a sponsor whose only public operating record runs to roughly 40 megawatts is an ambition, and ambitions are free to state.

The disclosures that will actually move this from announcement to transaction are the equity split, the jurisdiction, and the first named site with a signed power agreement behind it.

Expect at least one of those inside two quarters, because a vehicle that intends to raise outside money cannot stay this quiet and still close.

The question to hold until then is simple.

When the terms arrive, whose balance sheet is standing behind the first 150 megawatts, and does that balance sheet have a name the market already trusts, or one it is being asked to trust for the first time.

Join Global Data Center Hub

elcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Anthropic has reportedly agreed to pay Nscale roughly $45 billion over six years for 460 megawatts of computing capacity at the Monarch Compute Campus in Mason County, West Virginia. Bloomberg first reported the terms on August 26, 2026, citing people familiar with the matter. Neither Anthropic nor Nscale has confirmed the agreement, and no contract terms have been published.

• The reported terms imply $16.3 million per megawatt per year, or roughly 31 percent of allocated build cost returned annually. Reported Phase 1 economics of approximately $71 billion across 1.35 gigawatts put Anthropic’s allocated capital cost near $24 billion, against $45 billions of payments. A contract returning build cost 1.86 times in six years amortizes the NVIDIA Vera Rubin silicon inside the term.

• Nscale retains the residual value of a 460-megawatt accelerator fleet at the end of the term in roughly late 2033, and no public disclosure prices it. Anthropic pays for the hardware across the six years and holds nothing at expiry. The residual and renewal terms are most likely to surface in Nscale’s listing document rather than in any announcement.

Part One: The Record

Anthropic has reportedly agreed to pay Nscale roughly $45 billion over six years for artificial intelligence computing capacity in West Virginia.

Bloomberg first reported the agreement on August 26, 2026, citing people familiar with the matter, and CNBC and Reuters followed with the same terms from the same class of sources.

Neither Anthropic nor Nscale has confirmed the deal.

Nscale has not commented, no contract terms have been published, and every figure below is reported rather than announced.

The reported commitment covers approximately 460 megawatts at the Monarch Compute Campus in Mason County, West Virginia, according to Bloomberg.

Capacity is scheduled to begin coming online in late 2027. The reported term averages roughly $7.5 billion a year.

Capacity, Silicon, And Delivery Timeline

Nscale will deploy NVIDIA Vera Rubin systems at the site, per Bloomberg and CNBC.

The 460 megawatts cover the first of three planned buildings at Monarch, which Blockspace calculates at about 34 percent of the campus first phase.

Nscale has designed Phase 1 to hold 1.35 gigawatts of computing capacity and targets delivery by early 2028.

Reporting carried by TradingKey puts Phase 1 development cost at approximately $71 billion, including around $47 billion for chip purchases.

Nscale separately identifies more than $20 billion in planned capital investment.

Campus Scale And Onsite Generation

The Monarch Compute Campus occupies more than 2,250 acres, with a long-term expansion path exceeding 8 gigawatts.

The campus generates its own electricity rather than drawing from local utilities.

Caterpillar is supplying G3500-series natural gas generator sets, and Nscale targets 2 gigawatts of onsite generation by early 2028.

Nscale has said the development will use closed-loop cooling without municipal drinking water.

Planning documents cited by Bloomberg project 4,375 construction workers at peak, 645 direct operating jobs, and roughly 3,600 indirect jobs across the state.

Site History And Prior Tenants

Nscale acquired the Monarch site in March 2026 through its purchase of American Intelligence & Power Corporation.

Fidelis New Energy and 8090 Industries previously sponsored the development.

Nscale announced a letter of intent with Microsoft for up to 1.35 gigawatts at the campus.

Semafor reported that Google and Microsoft were both in talks for capacity at Monarch before the lease went to Anthropic.

Governor Patrick Morrisey issued a statement calling the reported agreement an extraordinary vote of confidence in West Virginia, and credited State Senator Glenn Jeffries for the state marketing effort.

Anthropic has signed other large compute agreements, including a $50 billion partnership with Fluidstack and capacity arrangements with Amazon Web Services, CoreWeave, and Akamai, according to Data Center Dynamics.

Nscale is preparing a public listing, with reports putting the target raise at around $3 billion.

Part Two: The Read

The Signal Is $16.3 Million Per Megawatt

The number that describes this contract is $16.3 million per megawatt per year.

That is what $45 billion over six years buys across 460 megawatts, and it is a rate of return on build cost that no coverage of the deal has calculated.

Every figure needed to run it has been public since August 26.

The consequence of leaving it unrun is that the market is reading a real estate lease when the payment profile describes something else entirely.

The Payment Profile Amortizes Chips Inside The Term

Take the reported Phase 1 economics at face value.

Approximately $71 billions of development cost across 1.35 gigawatts works out to roughly $52.6 million per megawatt, of which the reported $47 billion chip line accounts for about $34.8 million.

Anthropic’s 460 megawatts therefore carries an allocated build cost near $24 billion. Against that, six years of payments total $45 billion.

The ratio is 1.86 times. Annual rent runs at roughly 31 percent of allocated capital cost.

A stabilized wholesale data center lease does not return its build cost inside the term, because the building outlives the tenant and the landlord underwrites the reversion.

This contract returns the build cost nearly twice over in six years.

The silicon is being paid off inside the lease, and the shell and the power plant are riding along.

That distinction determines who carries technology risk.

Over the next four to six quarters, expect lenders to demand the split be visible in the documents, with the silicon component and the power-and-shell component priced separately.

The contracts that do not break it out will trade wider.

Nscale Owns The Residual And Nobody Has Priced It

Vera Rubin capacity comes online in late 2027 on the reported timeline.

A six-year term carries the contract to roughly late 2033.

Anthropic pays for the accelerators across that window and holds nothing at the end of it.

Nscale keeps a 460-megawatt fleet of six-year-old silicon, a building, and a gas plant.

What that fleet is worth in 2033 is the single largest open variable in the transaction, and it sits entirely on the Nscale side of the table.

If the residual holds value, Nscale has financed a campus on one contract and kept the asset. If it does not, Nscale has run a six-year amortization at a decent yield and Anthropic has bought compute at a fair clip. Both outcomes are live.

Neither is priced, because no one outside the two counterparties has read a renewal clause or a residual term.

The disclosure event is not a press release. It is Nscale’s listing document.

Watch whether the residual and renewal terms appear there, because the first neocloud contract whose back end becomes public sets the template every lender after it inherits.

What a 2033 buyer can do with that hardware was decided in civil engineering, back when the substations were sized and the floor loads were set. Those choices narrow the exit buyer pool years before anyone opens the contract to renew it.

Nscale’s Project Debt Underwrites Anthropic, Not Monarch

Forbes reported that the Anthropic contract functions as the collateral that lets Nscale finance a facility that does not yet exist.

Follow what that means for the lender.

The security is a six-year receivable from a private, unrated company with no published financials and major compute obligations, including the reported $50 billion Fluidstack partnership and deals with AWS, CoreWeave, and Akamai.

The tenant history sharpens the point.

Microsoft held a letter of intent for 1.35 gigawatts at Monarch. Semafor reported Google was in talks.

The anchor tenancy landed with the counterparty that has the largest forward obligations and the least public disclosure of the three.

Over the next twelve to twenty-four months, the pricing input that moves these facilities is not West Virginia gas or NVIDIA delivery schedules.

It is the first comprehensive disclosure of Anthropic’s aggregate compute commitments.

That disclosure reprices every project financing built on an Anthropic covenant on the same day.

What Each Segment Prices Before The First Filing

Private Capital underwriting Nscale project debt or pre-listing equity should treat the $71 billion Phase 1 figure as unverified reporting and rebuild the per-megawatt stack from the chip line up.

Demand the residual definition, the renewal mechanism, and the termination waterfall before pricing.

Firms that skip this step price a real estate risk and hold a technology residual, and they find out which one they bought in 2033.

Public Markets investors have no security to trade yet, which is the point.

The event to prepare for is Nscale’s listing document.

The question to bring to it is what share of contracted backlog this single contract represents.

A backlog concentrated in one unrated private counterparty is a different security from a diversified one at the same headline number, and it will not be presented that way.

Operators competing for the same tenants should read the microgrid, not the acreage.

Monarch’s differentiator is 2 gigawatts of behind-the-meter gas generation targeted for early 2028, which removes the interconnection queue from the delivery schedule.

Sites that depend on a utility connection are selling a later date to tenants who are paying for an earlier one.

Operators who have not secured generation by the end of 2027 compete for the tenants these campuses decline.

The Contract Nobody Has Read Sets The Benchmark

The reported $45 billion is not the durable output of this transaction.

The durable output is a per-megawatt clearing rate for chip-inclusive compute capacity, established by two counterparties who have said nothing on the record, and inherited by every neocloud contract negotiated after it.

Anthropic and Nscale have set a price without publishing one.

The question worth holding for the next two quarters is narrow.

When Nscale’s listing document arrives, does it disclose the residual and renewal terms, or does it report $45 billion as backlog and leave the back end unread?

The first version gives the market a benchmark. The second gives it a number, and a number without a residual clause behind it is the thing that gets mispriced.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• NVIDIA made a minority equity investment in Cloverleaf Infrastructure, a Houston data center site developer, under a strategic partnership announced on August 21, 2026. Neither company disclosed the investment size or the equity percentage. The announcement attaches no offtake, no chip supply agreement, and no power purchase agreement. J.P. Morgan Securities served as exclusive financial advisor to Cloverleaf Infrastructure and Kirkland & Ellis served as legal counsel.

• NVIDIA declined to acquire Cloverleaf Infrastructure outright and took a minority position instead. The Wall Street Journal reported that Cloverleaf Infrastructure was working with JPMorgan Chase on strategic alternatives that included a full sale, and that the company raised $300 million at founding in 2024. Control of a site developer carries the entitlement, permitting, and interconnection liabilities that a minority stake avoids.

• The NVIDIA DSX Platform enters Cloverleaf Infrastructure sites at the design phase, before utility agreements are signed. Substation sizing, floor loading, and coolant capacity are fixed during civil engineering, which moves accelerator switching cost out of procurement. Lenders underwriting powered land will need to price architecture neutrality as a separate variable.

Part One: The Record

NVIDIA made a minority equity investment in Cloverleaf Infrastructure, a privately held data center site developer, as part of a strategic partnership announced on August 21, 2026.

Neither company disclosed the size of the investment or the equity percentage acquired.

The stated purpose is to support the buildout of AI factories and to help Cloverleaf Infrastructure serve customers and utility partners meeting rising demand for accelerated computing.

Cloverleaf Infrastructure is headquartered in Houston.

Disclosed Terms And Transaction Advisors

The announcement records no price for the stake and no equity percentage.

It names no chip supply agreement and no power purchase agreement attached to the investment. It sets out no minimum customer commitment from either party.

J.P. Morgan Securities served as exclusive financial advisor to Cloverleaf Infrastructure, and Kirkland & Ellis served as legal counsel.

The Wall Street Journal reported before the announcement that NVIDIA was expected to invest several hundred million dollars.

The same report stated that Cloverleaf Infrastructure had been working with JPMorgan Chase in recent months on strategic alternatives that included a possible sale of the entire business.

Neither the figure nor the sale process appears in the announcement, and Reuters confirmed only that financial terms were undisclosed.

Founding Backers And Site Portfolio

Cloverleaf Infrastructure was formed in 2024 with an initial investment from Sandbrook Capital and NGP Energy Capital.

The company describes itself as a real estate developer working with investors, energy innovators, and utilities to deliver clean-powered, shovel-ready sites to technology companies that build and operate data centers.

Since founding it has delivered multiple gigawatt-scale projects to customers across North America.

The Wall Street Journal reported that the founding raise totaled $300 million, that Cloverleaf Infrastructure has sold projects amounting to more than 7 gigawatts of powered land, and that its pipeline exceeds 10 gigawatts.

The same report identified Wisconsin sites intended for Oracle and OpenAI. None of those figures appears in the announcement.

David Berry, Co-Founder and CEO of Cloverleaf Infrastructure, said the partnership strengthens the company’s ability to identify and develop high-quality sites.

Application Of The NVIDIA DSX Platform

Cloverleaf Infrastructure will apply the NVIDIA DSX Platform to bring site, power, cooling, computing, and facility decisions together earlier in the design phase.

NVIDIA states the platform helps customers weigh infrastructure tradeoffs and build AI factories that deliver more useful output within available power, water, and grid constraints.

Once sites are operational, NVIDIA DSX infrastructure and software are intended to optimize energy use and computing capacity.

Cloverleaf Infrastructure customers will be able to engage NVIDIA across accelerated computing, high-performance networking, infrastructure and platform software, and NVIDIA DSX.

Nico Caprez, Vice President of Global AI Infrastructure Growth at NVIDIA, said that land, power, and shell form the foundation of AI factories.

Neither company named a first joint site, a capacity target, or a delivery date.

Part Two: The Read

Three Hundred Million Is The Number That Prices This Deal

$300 million is the amount Cloverleaf Infrastructure raised at founding in 2024, according to The Wall Street Journal, and it sets the scale of what NVIDIA turned down.

The sale process and the minority structure have both been reported, and no coverage has put them together: NVIDIA had the whole company available and took a slice.

Every read treating this as NVIDIA moving upstream into land has the direction of travel backwards.

A Declined Acquisition Prices What NVIDIA Actually Wanted

NVIDIA has been writing far larger checks into the power layer.

Reporting places roughly $2 billion into Lancium and $1.5 billion into SB Energy, both ahead of this transaction, and both figures are secondary-sourced.

Against those, a company capitalized at $300 million two years ago was affordable at any structure NVIDIA preferred. It chose the smallest one available.

Control of a site developer means control of its liabilities.

Zoning, interconnection, transformer delays, water permits, and opposition become consolidated risks when a buyer takes the whole entity and a semiconductor balance sheet doesn’t make them disappear.

A minority position with a design integration attached delivers the outcome NVIDIA wants without the drag.

Watch the next site-layer investment. If it is also minority with platform integration, NVIDIA has a repeatable template, the acquisition path is closed, and site development stays fragmented for another cycle.

The Design Standard Lands Before The Interconnection Agreement

The NVIDIA DSX Platform enters at the design phase, which places NVIDIA’s density, cooling, and electrical assumptions inside the site before the utility agreement is signed.

Substation sizing, floor loading, coolant distribution capacity, and water strategy are fixed during civil engineering, and civil engineering is the least reversible layer of a data center.

That timing changes where switching costs live.

A site built around one accelerator’s power and thermal profile can require costly redesign, requalification, and utility renegotiation to switch. Procurement decisions cost little by comparison.

The choice is made years before anyone selects a chip.

Over the next twelve to twenty-four months, lenders underwriting powered land will need to price architecture neutrality as a distinct variable.

A site’s residual value depends on how many buyers can use it as entitled, and a portfolio designed to one standard has fewer of them.

Channel Access Without A Purchase Obligation

The announcement gives Cloverleaf Infrastructure customers access to NVIDIA across the full AI factory stack.

It attaches no offtake, no chip supply agreement, and no minimum commitment in either direction.

NVIDIA reaches gigawatt-scale projects years before any procurement process opens, and it owes nothing for the privilege.

Demand risk sits where it sat before, with the developers and hyperscalers who lease the land and buy the hardware.

Cloverleaf Infrastructure gains capital and a design partner, and gains no contracted revenue.

That distinction matters for anyone marking a site developer with a strategic investor on the register.

Nothing disclosed here guarantees Cloverleaf Infrastructure a dollar of NVIDIA-driven revenue.

Underwrite The Specification, Not The Stake

Private Capital. Infrastructure funds and family offices evaluating site developers should ask: when does a hardware vendor’s design standard enter the site, and what does removing it cost. Price re-specification as a line item.

Funds that skip this will buy powered land portfolios whose optionality was spent during entitlement, and they will find out at exit, when the buyer pool is narrower than the model assumed.

Public Markets. NVIDIA holders now have equity exposure to Lancium, SB Energy, and Cloverleaf Infrastructure, alongside a financing platform targeting over $500 billion in third-party capital. Each stake is individually immaterial, and none carries a disclosed size.

The aggregate is a portfolio of illiquid infrastructure exposures accreting inside a semiconductor income statement with no line item to track it. The disclosure that makes this measurable may take several quarters to arrive.

Operators. Developers and hyperscalers leasing Cloverleaf Infrastructure land inherit whatever specification the entitlement carries. Negotiate design neutrality into the site agreement now, while the pipeline is still being entitled and the terms are open.

Operators who take delivery of a finished site and then discover its constraints will pay for flexibility they could have written in for nothing.

The Standard Outlives The Stake

NVIDIA bought a position in how American AI sites get designed, and it paid a price that will look small against what the standard is worth.

The stake can be sold. The substations, the floor loads, and the coolant capacity built to one specification stay in the ground for thirty years.

The open question is whether any competing accelerator vendor can now enter the design phase at all.

Cloverleaf Infrastructure is one developer among several, and the model here is cheap to copy.

If the answer arrives as silence over the next four quarters, the design layer consolidated while the market was still counting gigawatts.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

In this week’s issue:

First, a utility gave up the one thing utilities never give up. Georgia Power won approval to serve OpenAI’s 3.2GW Effingham County campus and agreed to recover no shortfall from other customers if OpenAI leaves early. Nobody outside the room has read the contract, so who carries that cost instead?

Then, Nvidia kept buying the layer beneath its own customers. It took a minority stake in Cloverleaf Infrastructure, a powered-land developer founded in 2024 with a pipeline above 10GW. That is the third land and power position in a month, after Lancium and SB Energy.

Finally, Spain made cheap renewables the barrier to entry. A fast-tracked draft decree would make new data centers above 1MW match 80% of every hour’s consumption with new renewable generation. The penalty is loss of grid access.

Let us get into it.

What Georgia Sold to Get 3.2 Gigawatts

A regulated utility gave up guaranteed cost recovery and nobody can read the price. Georgia Power won staff approval on 26 August to serve OpenAI’s Project Camellia in Effingham County. The contract covers 3,200 megawatts over 25 years. OpenAI puts the campus investment at $20 billion.

The concession is specific. Georgia Power will not recover any revenue shortfall from other customers if OpenAI terminates early. It must tell the commission within fifteen days of any large-load cancellation. OpenAI pays the full cost of the infrastructure built to serve it.

Commission staff cleared the contract with no commissioner vote, under a rule the commission wrote in April 2025. The public version is entirely redacted. So are the six other data center contracts Georgia Power has filed under that rule.

So here is the discipline. Georgia Power’s protection is a promise, and the promise sits inside a document you cannot read. Before you treat a utility’s ratepayer pledge as protection, ask what the filing would have to disclose, then go and check whether it does.

What Georgia Power bought with it, and the filing that would show whether the trade holds, is below.

THIS WEEK BY REGION

The week’s biggest moves — what happened and what it signals.

North America

Termination risk became the term being negotiated. Georgia Power won approval for its OpenAI contract with no recovery from other customers if OpenAI leaves. Nvidia took a minority stake in Cloverleaf Infrastructure, its third land and power position this month.

nVent agreed to buy Maverick Power for $1.75 billion, with up to $550 million tied to 2027–28 performance. Anthropic reportedly committed $45 billion over six years for 460MW at Nscale’s West Virginia campus. Neither company confirmed the terms.

If you invest in public equities, the vendors are now buying their own demand, and that changes what a backlog is worth.

Europe

Europe is writing the entry rules before capital arrives. Spain agreed to fast-track a draft decree requiring data centers above 1MW to match 80% of hourly consumption with new renewables.

The penalty escalates to loss of grid access. Projects already in the permitting queue get six months to comply or forfeit their connection rights without compensation.

If you develop assets, a queue position in Spain is now conditional on generation you have not built.

Asia-Pacific

Capital in the region is moving toward power at source. Zerra DC lodged a development application for the Western Downs Digital Park, proposing four 360MW buildings for 1.44GW of IT capacity near the Braemar substation.

Cushman & Wakefield puts regional data center asset values above $950 billion by 2030, requiring over $280 billion in new capital. Japan accounts for nearly a quarter of that.

For infrastructure funds, the binding constraint is deliverable capacity, and Queensland is the test case for clearing it.

South America

Colombia’s expansion is carrying its own credit. Equinix raised Bogotá investment to $73 million, completing BG2 at 1,100 cabinets across 6MW to 7MW.

If you sit at a project finance institution, Colombia’s credit quality is settled, and the open question is whether anchor demand ever arrives.

PUBLISHED THIS WEEK

• NVIDIA Did Not Guarantee The Rent. It Guaranteed The Resale. — NVIDIA committed up to $105 billions of residual value guarantees on SB Energy’s Ohio AI data center campus on August 17. Inside the recovery waterfall and the undisclosed minimum value.

• The Texas Pause Is Not The Constraint. Docket 59220 Is. — Governor Greg Abbott froze 474 gigawatts of ERCOT data center interconnection requests on August 3. Inside the 525.5 MW curtailment order, the 30-minute shutdown rule, and the December 10 report.

• The Four-Year Wait for a Transformer (And Why It Decides How Fast AI Grows) — Every AI answer runs through a steel box that takes more than two years to order. That wait sets the pace of the AI buildout.

• Procurement Strategy: Equipment, EPC, and the Contracts That Set Your Timeline — Long-lead equipment, vendor concentration, EPC versus design-build, fixed-price and GMP risk allocation, liquidated damages.

Sponsored By: Global Data Center Hub

Your trusted source for global AI infrastructure analysis. Every week, we break down the trends shaping the future of data centers, cloud, and AI, from emerging markets to hyperscale battlegrounds.

📰 Subscribe to get expert insights delivered to your inbox.

What Georgia Power Bought With It

The utility took a curtailment right in exchange. OpenAI committed up to 1,000 megawatts of flexible demand response. That is close to a third of the load, released back to the grid on request. Georgia Power can serve peak demand without building generation for the full 3,200 megawatts.

The economics run on that trade. Georgia Power projects about $950 million a year of incremental large-load revenue. It projects $2.8 billion of customer savings between 2029 and 2031, and roughly $180 a year off residential bills from 2029.

Those are utility projections filed with the commission. They are not commitments. The bill reduction also depends on enough large-load contracts clearing the commission, and one contract does not deliver it.

What the trade cost is not disclosed. The tariff rate OpenAI pays, the demand-response compensation and the termination terms are redacted. Without those three numbers, nobody outside the commission can tell whether the curtailment right covers the recovery Georgia Power surrendered. A missing disclosure is a finding.

The marker to here watch here is Georgia Power’s 2028 rate adjustment. That filing has to show the large-load revenue actually collected against the $950 million projection. If the number lands short and the bill reduction does not arrive, the commission will have to defend a contract it never voted on.

NOTABLE TRANSACTIONS

Key structures and capital moves from this week’s deal tape.

nVent: $1.75 billion definitive agreement to acquire Maverick Power

nVent is acquiring McKinney switchgear maker Maverick Power for $1.75 billion in cash and new debt, with Bank of America providing the bridge. The deal values Maverick at about 11.5× 2026 adjusted EBITDA, plus a $550 million earnout.

If you underwrite industrials, the mechanism is the earnout, because it prices the buyer’s own doubt about how long the switchgear shortage lasts.

Cloverleaf Infrastructure: undisclosed minority investment from Nvidia

Nvidia took a minority stake in Cloverleaf and agreed to use its DSX design across a 10GW pipeline. The specification lock effectively turns the supplier relationship into a demand guarantee without balance-sheet consolidation.

For operators, the mechanism is standard setting bought with equity, and it decides whose racks fit the sites coming to market in 2028.

Anthropic: reported ~$45 billion, six-year capacity commitment at Nscale

Bloomberg reported a six-year, roughly 460MW commitment at Nscale’s West Virginia campus, later followed by CNBC and Reuters, though unconfirmed. The offtake contract effectively provides project credit to the developer.

If you buy neocloud paper, the mechanism is offtake standing in for balance sheet, and the term length decides whether it holds.

If you are new to Global Data Center Hub, visit our Start Here page for more information about the newsletter.

If you found this newsletter useful, share it with a colleague.

Have a great week.

— Obi

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

This article is the 11th article in the series: Ground to Grid: A Free 21-Lesson Guide to Mastering Data Center Development

Join Global Data Center Hub

TL;DR

• Large power transformers above 50 MVA now run 128 to 160 weeks, which makes procurement the binding constraint on data center delivery. Tier-1 manufacturers ABB, Siemens Energy, Hitachi Energy, and GE Vernova quote 48 to 60 months and hold backlogs into 2030. Tier-2 producers Efacec, SGB-SMIT, and CG Power quote 24 to 36 months, and medium-voltage switchgear runs 44 to 84 weeks.

• The manufacturing clock on data center equipment starts at Release to Manufacture, which can sit a full quarter behind the purchase order. A purchase order issued in January with Release to Manufacture held until April loses twelve weeks before fabrication begins. Hyperscale developers reserve factory slots with deposits of 10 to 30 percent of equipment cost at land close, six to twelve months before groundbreaking, using ratings frozen at roughly 60 percent single-line design completion.

• Guaranteed maximum price contracts signed against incomplete design have produced change orders of 8 to 14 percent on data center projects. Contracts where design is frozen at around 60 percent completion produce 3 to 5 percent. Delay liquidated damages in data center agreements run 0.1 to 0.5 percent of contract value per day with caps of 5 to 10 percent, against caps of 10 to 20 percent in the wider capital projects market.

Procurement sits between design and construction in the data center development sequence.

It is the stage where engineering specifications become purchase orders, vendor contracts, and delivery dates.

On most large projects it is also the stage that determines when the facility energizes.

The reason is straightforward. Construction speed has not changed much. Equipment lead times have.

A team can hold every civil and structural milestone and still wait a year for a transformer, and no amount of site productivity recovers that time.

The order date sets the delivery date before the first crew mobilizes.

The second half of this stage is contractual. The equipment schedule creates risk, and the contract decides who carries it.

Delivery models, pricing structures, and damages provisions allocate the consequences of a slip before anyone knows whether one will occur.

The four procurement decisions that set the timeline

Long-lead equipment. Transformers, switchgear, generators, and cooling systems often drive data center schedules.

Large power transformers above 50 MVA can take 128–160 weeks, while Tier-1 manufacturers quote 48–60 months, with backlogs into 2030. Tier-2 producers typically quote 24–36 months.

Medium-voltage switchgear takes 44–84 weeks, diesel generators 50–110 weeks, and behind-the-meter gas turbines 36–84 months.

Vendor strategy. The choice is between sole-source and multi-vendor procurement, and between bespoke and standardized designs across a portfolio.

Sole-sourcing improves commercial terms and removes the substitution option if that manufacturer slips.

Multi-vendor procurement requires performance-based specifications, written by voltage class, fault rating, and footprint, so several approved manufacturers can bid the scope.

The EPC contract. Design-build and EPC together account for roughly 70 percent of hyperscale project starts. EPC concentrates schedule and performance responsibility with one contractor.

Design-build keeps more design oversight with the owner and often leaves long-lead equipment for the owner to buy directly.

EPCM, more common internationally, makes the owner the direct counterparty to every vendor.

Contract structure and risk allocation. Fixed-price, cost-plus, and guaranteed maximum price contracts distribute cost and schedule exposure differently, and the choice usually follows design maturity.

Guaranteed maximum price contracts signed against incomplete design have produced change orders of 8 to 14 percent, against 3 to 5 percent where design is frozen at around 60 percent completion.

Why equipment lead time is the binding constraint

Three mechanics explain why the equipment schedule governs the project schedule.

The first is that the manufacturing clock frequently does not start when the purchase order is issued.

On many packages it starts at approved submittal or at Release to Manufacture.

A purchase order placed in January with Release to Manufacture held until April has lost twelve weeks before fabrication begins. A single blended lead-time figure hides this.

Purchase order date, submittal approval, Release to Manufacture, factory slot confirmation, factory acceptance test, ship date, and site-need date each need separate monitoring.

The second is that ordering can begin before design is complete.

High-voltage transformers and switchgear typically need ratings frozen only at roughly 60 percent single-line design completion.

Reserving a factory slot with a deposit of 10 to 30 percent of equipment cost at land close is now standard on hyperscale campuses, with the order date preceding groundbreaking by six to twelve months.

Without a reserved slot, nominal lead times extend by another eight to sixteen weeks.

The third is supply concentration. A small group of manufacturers dominates the categories that gate delivery.

Two to three global producers control the high-voltage transformer market, constrained by grain-oriented electrical steel supply, and there is no fast substitute.

Vertiv, Eaton, and Schneider Electric sit across power distribution, uninterruptible supply, switchgear, and cooling.

Qualifying a second manufacturer in every critical category, and writing specifications by configuration, preserves the option to move when a primary vendor slips.

Contract structures and who carries the risk

Two provisions do most of the work inside these structures.

Delay liquidated damages set a daily rate for late completion, commonly 0.1 to 0.5 percent of contract value per day in data center agreements, capped at 5 to 10 percent.

The wider capital projects market runs caps of 10 to 20 percent.

Delay damages and performance damages belong in separate provisions with separate caps, triggered by defined milestones such as mechanical completion.

Force majeure and change-in-law provisions allocate the delays nobody controls.

Current drafting practice enumerates the foreseeable categories explicitly, including equipment lead-time delays, interconnection delays, permitting delays, and tariffs.

These provisions need coordinating across the supply agreements, the construction contract, and the financing documents, so a declaration in one does not create a default in another.

Where the owner orders equipment early and novates the purchase order to the contractor, two mismatches appear.

The vendor damages cap is set against equipment value while the prime contract cap is set against project value.

Vendor warranties often begin at shipment and can expire before the equipment is energized.

Both are addressed by drafting the original purchase order for eventual novation.

What one-month costs

A one-month delay on a 60 MW AI data center has been costed at roughly $14.2 million dollars, covering lost lease revenue, idle labour, and penalty exposure.

That figure explains the behaviour on both sides of the table.

Owners press for wide liquidated damages coverage. Contractors press equally hard for caps.

It also explains why deposits placed years ahead of delivery are now treated as ordinary project cost.

Against a monthly delay cost in the tens of millions, a slot reservation deposit is cheap.

Strategic implications

For developers. Order the long-lead equipment first. The order goes out at land close, six to twelve months before groundbreaking, using ratings frozen at roughly 60 percent single-line design completion.

Specify by configuration so several manufacturers can bid the package and get the factory slot confirmed in writing.

For investors. Ask when the long-lead equipment was ordered and what the Release to Manufacture date is. A purchase order date alone proves little, because fabrication often begins a full quarter later.

A late order is a slipped delivery date the pro forma has not caught up to yet.

For operators. The vendors selected here define the maintenance regime, the spares inventory, and the reliability profile for the life of the asset.

Standardization cuts spares holding and simplifies training.

Multi-sourcing adds resilience at procurement and complexity at operation.

Key takeaway

Construction gets the attention, but procurement sets the date.

The developer who orders early controls the timeline. The one who orders late discovers it is already lost.

Equipment scarcity has moved the first meaningful milestone of a data center programme from groundbreaking to the purchase order.

Everything downstream, including the financing schedule, the lease commencement date, and the tenant fit-out plan, now depends on a manufacturing slot that was allocated years earlier.

The projects delivering in 2029 were ordered in 2025 and 2026.

Capacity beyond that is being allocated now.

On your projects, what is the long-lead item that most often controls the schedule and how early do you now commit to it?

Part 12 examines what a data center actually costs and the drivers that move it: What a Data Center Actually Costs: CapEx Breakdown and the Drivers That Move It

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Large power transformers, the equipment that converts the very high voltage carried by long-distance power lines into a voltage a building can actually use, now average 128 weeks from order to delivery, according to a Wood Mackenzie survey of utility and developer orders. The largest units, which connect power plants to the grid, average 144 weeks. A data center building takes roughly two years to construct.

• Developers now order power equipment years before a project has a customer or a final go-ahead. Chevron ordered seven large gas turbines from GE Vernova, the power equipment company spun out of General Electric in 2024, roughly two years before signing the twenty-year electricity agreement with Microsoft that those turbines will serve.

• The published order books of power equipment makers show whether an announced AI buildout is real. GE Vernova reported $2.4 billion of data center electrical equipment orders in the first quarter of 2026, more than the company booked in all of 2025, and those figures are filed with regulators every quarter while project announcements carry no such requirement.

Somewhere near your house there is a grey drum on a wooden pole, or a green metal cabinet sitting on a concrete pad at the end of the street.

That is a transformer. It takes the very high voltage that travels down long-distance power lines and turns it into the much lower voltage your kitchen can handle.

Every building on earth has one of these standing between it and the grid.

So does every data center, except theirs can be the size of a small house and weigh several hundred tons.

That box is currently the thing the AI boom is waiting for.

Chips are shipping in volume. Money is arriving faster than anyone forecast. Land has been bought across half a dozen states.

The queue that all of it now stands behind runs through a handful of heavy-equipment factories, and it is measured in years.

This is worth a newcomer’s attention for a reason beyond engineering.

AI infrastructure stories often focus on technology companies or massive investments, but neither explains the pace.

The real constraint is less glamorous: factory schedules booked years in advance.

Once you see that layer, the delays and unusual announcements make sense.

Chevron ordered the turbines two years before it had a customer

In June 2026, Chevron announced a deal with Microsoft.

Chevron will build a power plant near Pecos, in West Texas, and sell all of its electricity directly to a Microsoft data center under a twenty-year agreement.

The plant is designed for roughly 2.67 gigawatts, which is about the electricity used by two million American homes. First power is expected in 2028.

Read as a headline, that is a story about an oil company walking into the AI business.

The detail that actually explains it sat in an interview.

Jeff Gustavson, who runs Chevron’s New Energies business, said the company began thinking about the project about three years earlier and committed to it about two years earlier.

What that commitment consisted of was placing the equipment order.

A gas turbine is essentially a truck-sized jet engine that burns natural gas to spin a generator.

Chevron ordered seven large ones from GE Vernova, the power equipment company spun out of General Electric in 2024.

It placed that order before it had a signed customer, before its own board had given the project a final approval, and before the design was settled.

That order was the deal.

By the time Microsoft was deciding where to put a very large data center, land was available in several places, natural gas was available in several places, and money was available almost everywhere.

A delivery slot for seven large turbines was available from almost nowhere. Chevron was holding one.

The project has a public cost attached to it.

The plant burns natural gas, and Microsoft has a standing pledge to remove more carbon from the atmosphere than it emits by 2030, a target this makes harder to reach.

Chevron has said the plant will run separately from the regional grid so that local households are not competing with it for power.

If you read that announcement as an oil company discovering technology, you missed the part that decided it, which is that the winner was whoever had joined the queue earliest.

The building takes two years. The equipment takes longer.

Here is the number that reorganizes everything.

Wood Mackenzie, an energy research firm, surveyed actual utility and developer orders and found that standard large power transformers average 128 weeks between order and delivery.

That is close to two and a half years. The largest units, the ones that connect a power plant to the grid, average 144 weeks.

Analysts at PwC have put the wait for the highest-capacity units as long as four years.

Now hold that against the building itself, which takes roughly two years to put up.

The reason is unglamorous and hard to fix.

A large power transformer needs a specialized electrical steel that only a few mills in the world produce, plus copper, plus custom engineering, plus months of testing before it ships.

Adding factory capacity takes years, because a transformer plant is itself a heavy industrial building.

And three separate waves of demand arrived at those same factories at once.

Data centers need them. Older grid equipment installed decades ago needs replacing.

Factories and vehicle fleets switching from fuel to electricity need them too. None of those three has any reason to slow down for the others.

So when you read that an AI project has been delayed, the cause is usually a delivery date agreed years ago rather than anything happening on the construction site.

You can now reserve a place in a factory line for power you have not designed yet

Because the wait got long enough, the industry invented a way to buy a spot in it.

GE Vernova calls these slot reservation agreements.

A customer pays to hold a place in the manufacturing schedule before knowing exactly what it will need, in the same way a person books a table months ahead and works out later who is coming.

The scale of this is easy to miss. In the first quarter of 2026 alone, GE Vernova signed 21 gigawatts of new gas equipment agreements.

Only 2 gigawatts of that were firm orders. The other 19 gigawatts were reservations.

The company’s reserved slots rose from 43 to 56 gigawatts in a single quarter, and it told investors it expects to reach at least 110 gigawatts of combined orders and reservations by the end of 2026.

For context, 110 gigawatts is roughly a tenth of all the electricity generating capacity in the United States, held on one manufacturer’s books as future work.

This changes how a headline should be read.

When a company announces gigawatts of new AI capacity, a portion of that number may be a place in a line rather than a machine that exists.

That is why announced capacity and delivered capacity have drifted so far apart.

The factory built to fix the shortage opens in 2028

The obvious response to a shortage is to build more factories, and that is happening.

In South Boston, Virginia, Hitachi Energy has made transformers since 1968. In 2026, it began a $457 million expansion to make the site the largest power transformer factory in the U.S.

The project adds around 825 jobs to a workforce of about 850, close to doubling the place.

It is part of a wider investment of more than a billion dollars in American grid manufacturing.

The plant is expected to begin operating in 2028.

Sit with that timeline. A factory announced to ease today’s shortage may take two years to start production, with transformers requiring another two years to build and deliver.

Relief may not arrive until the end of the decade.

Virginia is a fitting place for it. Northern Virginia holds the largest concentration of data centers on earth.

Data centers already account for roughly a quarter of the electricity sold by the state’s main utility.

The useful correction here is about time.

This shortage is shaped like a decade-long industrial rebuild, and if you read it as a news story that a strong quarter settles, almost every forecast you meet will look more optimistic than it should.

$2.4 billion in one quarter, more than all of last year

There is a practical skill buried in all of this, and it costs nothing to use.

Equipment makers are public companies, so they report what customers actually bought, every three months, in filings that carry legal consequences if they are wrong.

Announcements carry no such requirement.

In the first quarter of 2026, GE Vernova’s electrification division, which makes transformers, switchgear, and the substations that connect large buildings to the grid, booked $2.4 billion of orders tied to data centers.

That was more than the division booked from data centers in the whole of 2025.

Total orders across the company reached $18.3 billion for the quarter, up 71 percent on the year before, and its order backlog, meaning work sold but not yet delivered, stood at $163 billion.

Those numbers say something that a press release cannot. Somebody signed for the equipment, and somebody is paying for it.

So, the next time a very large AI project is announced, the fastest way to know whether it is real is to wait a quarter and look at what the equipment makers say they sold.

The slowest part of the fastest thing

The AI boom is described almost everywhere as a story about speed.

Models improve in months. Companies are built in a year. Announcements arrive weekly.

Underneath that sits an industry moving at the speed of heavy manufacturing. A decision made in 2026 shows up as electricity in 2030.

The scarcest thing in the whole chain is a place in a factory schedule, and somebody else may already be holding yours.

Every answer you get from an AI tool travelled through one of those grey boxes on its way to you.

The chain that ends in a chip begins in a steel mill, and the whole of it moves at the pace of the slowest link.

Before reading this, what did you assume was the slowest part of building a data center?

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Fiber has become a gating constraint on data center development. The markets absorbing power-driven spillover carry land and grid headroom without dense, diverse route inventory. Route construction runs on permitting timelines, so it cannot be compressed after land capital is committed.

• The market still underwrites connectivity as abundant and fast to provision. That assumption formed when facilities carried north-south user traffic across existing metro rings. AI-grade clusters generate east-west traffic between processors and require an order-of-magnitude increase in strand count per rack, which those rings were never engineered to carry.

• Disciplined capital screens for fiber before it options land. Route diversity is verified by physical survey, secured through long-dated dark fiber rights, and modeled as a base-case cost and schedule item on every secondary-market site. Sponsors who apply that sequence acquire at a discount the undisciplined bidder creates.

Connectivity now decides whether a power-available site is developable.

The old underwriting model treated fiber as a short-lead utility, ordered during construction and energized before commissioning.

Sponsors who still underwrite that way commit land capital against a route that does not exist and will not exist inside the development window.

That model worked for twenty years. It worked because facilities were built where fiber already was.

Power scarcity moved the sites. Fiber did not follow.

Two constraints are now moving in opposite directions.

Interconnection queues and transformer lead times have closed the primary hubs to new mega-scale development, so capital is dispersing into secondary and tertiary metros with available land and available power.

Route density is not dispersing with it, because building fiber is a right-of-way problem before it is a capital problem.

The question worth asking is not where capacity is being announced.

It is where announced capacity can actually be absorbed, and the interval between substantial completion and rent commencement is where levered returns are decided.

The strand count is the mechanic

Legacy colocation was engineered for north-south traffic.

A user made a request, a server answered it, and the fiber count per rack was measured in tens. Metro rings were sized for exactly that load.

AI training changed the direction of the traffic.

GPU clusters synchronize across thousands of parallel processors, so the dominant flow runs east-west, between machines inside the hall and between halls across a campus.

Strand demand per rack rises by an order of magnitude. A route that carried a legacy colocation campus cannot carry an AI campus without a high-count overlay, and that overlay is a construction project measured in quarters.

Diversity is the second half of the problem.

AI-grade tenants require two physically separate paths, and two routes that look diverse on a carrier diagram frequently share one duct or one bridge crossing.

Verification requires an engineering survey. It does not come from a network map.

Consider a regional operator in a power-rich secondary metro. It optioned land, secured a grid allocation, and signed a letter of intent with an AI tenant on a twenty-six-month delivery.

Connectivity diligence came last in the sequence.

The two diverse routes serving the parcel converged at a single river crossing four miles out, and building a genuine second path required a new easement across rail land.

The tenant walked before the easement cleared. The land is still owned. The power allocation is still held. Neither one earns anything.

The market is wrong about provisioning speed

The industry has long assumed that connectivity is abundant, cheap, and fast to provision.

That assumption held when facilities were sited inside dense metro rings built for enterprise traffic and when strand requirements per rack were measured in tens.

It does not hold now, because power scarcity has relocated development into markets that were never carriered for AI density.

AI-grade tenants require a combination of strand count and physical diversity that thin markets cannot supply within any timeline a development pro forma can tolerate.

Fiber is no longer the last item on the checklist. It is the first filter.

The cost is time

Supply-side friction is real and it is measurable.

Cable lead times stretched from a few months to more than a year at the peak of the cycle.

Manufacturing lines reallocated toward higher-margin AI-grade fiber, which created a secondary shortage in conventional telecom grades.

Certified splicing and civil labor is short by tens of thousands of technicians globally, and a large share of that labor is committed to subsidized rural broadband programs.

Permitting is worse than the supply chain.

Right-of-way approvals cross municipal, rail, utility, and private easements, and each one runs on its own calendar.

Make-ready costs on utility poles have multiplied in several US programs.

A middle-mile lateral of only a few miles routinely takes twelve to twenty-four months to clear.

Policy is not stabilizing the market. NTIA’s 2025 BEAD overhaul removed the fiber preference and wholesale access requirement, weakening subsidies for the corridors now targeted by data center capital.

Regulatory gates are tightening on the other end at the same time.

Virginia moved to withhold final site plan approval until interconnection conditions are certified.

New York froze environmental permits on facilities above 20 megawatts for a year under the Responsible Data Center Development Act.

EirGrid held Dublin connections closed from 2021 and reopened them only against an on-site dispatchable generation mandate.

Every gate pushes another cohort of projects toward markets with thinner routes.

Hyperscalers are moving faster than everyone else on the response.

Multi-billion-dollar commitments to cable manufacturers and to dark fiber inventory are removing allocation from the merchant market before a mid-market sponsor reaches the table.

Three lenses on the same constraint

For independent operators, the binding variable is qualification. A parcel that clears on acreage, power, and tax abatement can still fail on route diversity, and the failure surfaces after the option payment.

Site selection has to be re-ordered so that connectivity screening runs first and eliminates parcels before any other diligence dollar is spent.

For private equity and infrastructure investors, the binding variable is the yield date.

Asset value depends on when cash flow starts, so a twelve-month slip in connectivity delivery degrades the equity IRR before a single rack is energized.

Contingency has to carry the fiber build as a base case. Treating it as an exception is what broke the last vintage of secondary-market underwriting.

For public equity, the key variable is dispersion. Carrier-dense assets with diverse metro rings, subsea landings, and direct cloud on-ramps generate compounding cross-connect revenue as capacity grows.

Generic wholesale halls in fiber-thin markets compete on power price and rack rate, which are commoditized inputs that compress over time.

The multiple gaps between those two asset types is widening, and nothing in the supply picture closes it.

The disciplines that hold

Screen fiber before land. Apply a hard distance rule to verified dense route inventory and kill the parcel before spending on power studies or incentive negotiation.

Verify diversity by survey. Accept engineering-grade route confirmation only and treat any carrier diagram as a marketing document until a surveyor contradicts it.

Model the route as the base case, including full lateral costs, a 12–24-month permitting window, and a contingency floor for infrastructure builds at every secondary-market site.

Own the access. Secure long-dated dark fiber rights across multiple underlying providers so that pricing and schedule control sit with you and not with a carrier’s allocation committee.

Reset the commercial terms. Tie rent commencement to connectivity milestones, not shell completion, and require anchor offtake to cover baseline capex before long-haul extensions begin.

Aggregate the route. A single lateral to one facility rarely clears on a standalone basis, but a route engineered to serve a corridor captures every developer who arrives after you.

Run that sequence and connectivity stops being a risk you manage. It becomes the reason you win parcels the next bidder cannot underwrite.

Join Global Data Center Hub

Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Join Global Data Center Hub

TL;DR

• Governor Greg Abbott directed the Public Utility Commission of Texas and the Electric Reliability Council of Texas on August 3, 2026, to audit every data center in the ERCOT interconnection queue before any additional project advances. ERCOT is processing more than 1,800 large-load requests totaling approximately 474 gigawatts, roughly 90 percent of it data centers. The audit covers 250 to 300 Batch Zero projects and reports on December 10, 2026.

• PUCT Docket No. 59220, decided July 24, 2026, imposed a cumulative 525.5-megawatt emergency curtailment obligation on two co-located data centers sitting behind a single 265.5-megawatt wind facility. The commission held that curtailment on a co-located load is not capped by the paired generator’s capacity. The conditions are uncompensated and attach to the physical asset.

• The Batch Zero financial security schedule, not the audit, is what clears the 474-gigawatt queue. Developers post a default 50,000 dollars per megawatt and forfeit 80 percent of it on withdrawal or a six-month milestone miss. ERCOT will not meet the April 9, 2027, study deadline and has proposed no replacement date.

Part One: The Record

Governor Greg Abbott directed the Public Utility Commission of Texas and the Electric Reliability Council of Texas to audit every data center advancing through the ERCOT interconnection process on August 3, 2026.

The instruction came by letter to PUCT Chairman Thomas Gleeson and ERCOT Chief Executive Pablo Vegas, and it stated that any project failing the review must be denied connection to the Texas grid.

ERCOT is processing more than 1,800 large-load interconnection requests totaling approximately 474 gigawatts, and roughly 90 percent of that volume is data centers.

ERCOT suspended its Batch Zero classification process the same day and filed for regulatory relief seven days later.

Queue Volume And Disclosure Requirements

The 474 gigawatts is more than five times the all-time hourly peak of 91,089 megawatts that ERCOT recorded on July 22, 2026.

ERCOT delivered the 90 percent data center share in testimony to the Texas Senate on July 29.

Abbott also cited the failure of some data centers to comply with the PUCT survey measuring water and power usage under the General Appropriations Act.

The directive requires every project to disclose five categories of information.

These cover public financial assistance, projected annual and peak electricity use, on-site generation, water use and cooling technology, community impact mitigation, and ownership and control.

Batch Zero Suspension And Good Cause Exceptions

ERCOT’s August 3, 2026, Market Notice confirmed it would not deliver Batch Zero large-load classification notices to transmission and distribution providers by the August 7 deadline.

Batch Zero is the system-wide interconnection study the PUCT approved on June 18, 2026, for loads of 75 megawatts and above.

ERCOT filed for three good cause exceptions on August 10, 2026, under PUCT Project No. 59142.

It sought relief from the August 7 classification deadline, permission to include unclassified base-load candidates in the August and November Quarterly Stability Assessments, and a 24-day window to cure dynamic-data deficiencies.

The PUCT granted all three on August 20, 2026, and approved conditional classification by August 31.

ERCOT also told the PUCT it has paused approvals to energize large computational loads, the category covering data centers and cryptocurrency mining facilities.

Seventeen large loads representing approximately 6.6 gigawatts of peak demand ramping over five years have cleared every ERCOT gate except the approval to energize.

Audit Scope And Reporting Timeline

The audit covers approximately 250 to 300 projects in the Batch Zero cohort rather than the full queue.

Most are data centers. ERCOT General Counsel Chad Seely described the effect of the letter as moving verification to the front of the line, ahead of the interconnection study.

ERCOT issues requests for information through interconnecting utilities in August and September 2026, with information gathering and cure periods running through October and November.

ERCOT files the Batch Zero Eligibility Verification Report and the Community Impact Review Report on December 10, 2026, and presents both at the PUCT open meeting on December 17.

ERCOT stated it will not meet the April 9, 2027, deadline to deliver Batch Zero study results and has not proposed a replacement date.

Part Two: The Read

The Binding Term Was Set Ten Days Before The Freeze

The number that governs Texas data center underwriting is 525.5 megawatts, and the PUCT fixed it on July 24, 2026, ten days before Abbott wrote his letter.

In Docket No. 59220 the commission imposed a cumulative 525.5-megawatt emergency curtailment obligation on two co-located data centers sitting behind a single 265.5-megawatt wind facility.

Coverage of the August 3 pause has treated behind-the-meter co-location as the route around the interconnection queue, and Docket 59220 was decided, published, and analyzed without that implication being drawn.

A developer who reads the pause as a timing problem and co-location as the answer is acquiring an asset whose shutdown obligation is sized by the grid rather than by the generator it paired with.

Curtailment Is Sized Against The Grid, Not The Generator

FGE Goodnight I and Crusoe Energy Systems applied on January 9, 2026, to net meter Goodnight Wind, an existing 265.5-megawatt facility in Armstrong County, with Crusoe Load Two, a 260-megawatt AI data center developed with Google.

Crusoe transferred the load to Ensign Infrastructure, a special purpose vehicle, in March.

Docket No. 58881 had already placed a 265.5 megawatt 30-minute curtailment obligation on the adjacent Crusoe Load One, and ERCOT recommended stacking a second full obligation on Crusoe Load Two.

Ensign, Crusoe, and Goodnight Wind called this double counting. The PUCT disagreed.

Both loads net behind the same point of interconnection, so curtailing one and not the other lets the other absorb the wind output and keeps the generator off the grid.

The obligation is uncompensated. Co-located loads are barred from paid demand response, ancillary services, Emergency Response Service, and utility load management programs.

The conditions attach to the physical assets and bind successors in interest. A joint review falls due between 36 and 60 months.

The commission dismissed Crusoe’s argument that repeated 30-minute shutdowns damage server hardware, calling it a commercial trade-off accepted for faster interconnection.

Over the next four to eight quarters co-location stops being a queue workaround and becomes a curtailment liability that travels with the title.

Full islanding remains outside the ERCOT interconnection process entirely.

The 933 megawatts of off-grid gas turbines already under construction at that same Armstrong County site is the structure the market moves toward.

Posted Security Turns A Queue Slot Into A Priced Position

The second mechanism sits inside Batch Zero.

Where upgrade costs cannot be determined, developers post financial security at a default 50,000 dollars per megawatt.

All direct interconnection costs are paid in cash through contribution in aid of construction, with no offset, and the interconnecting utility cannot recover them through PUCT-regulated rates.

A developer who withdraws, or who misses a phased energization milestone by six months, forfeits 80 percent of posted security to the utility’s rate base.

The remaining 20 percent returns on energization and after five years of operation.

A materially false eligibility attestation makes a project ineligible outright.

That regime, not the audit, clears the 474 gigawatts. An audit of 250 to 300 projects tests disclosure.

The security schedule tests conviction, and it prices a queue slot that used to be free.

When the December 10 report lands, the number worth watching is not how many projects passed.

It is the gap between requested megawatts and posted megawatts, the first honest measure of how much of the Texas pipeline was ever real.

December 10 Is Reversible. April 2027 Is Already Gone.

Read the calendar and the audit is the reversible part.

Seely told the commission the verification work was already scheduled under Batch Zero and the letter moved it to the front of the line.

Gleeson framed the goal as more confidence in the process. The PUCT granted every exception ERCOT requested.

What has actually slipped is the study. ERCOT will not deliver Batch Zero study results by April 9, 2027, and has proposed no replacement date.

That pushes the Long-Term Load Forecast and the Reliability Assessment, and it puts the December 2026 Capacity, Demand and Reserves Report at risk.

BloombergNEF put the delay cost at just over 8 billion dollars by the first quarter of 2027 at a 60 percent AI-compute mix, and near 15 billion at a full AI mix.

Those are revenue estimates on slipped capacity, not write-downs.

Megawatt allocations, not audit clearance, set build dates. A project can pass in December and still not know what it was allocated.

What Each Segment Prices Before December 17

Private Capital should reprice every co-located Texas position against an uncapped curtailment obligation rather than a proportional one and treat the conditions as attached to the asset in any change of control.

Diligence the 36-to-60-month review clock.

Firms underwriting co-location on the pre-59220 assumption are paying an islanded-asset price for a curtailable one.

Public Markets holders own the disclosure gap. Seventeen large loads representing 6.6 gigawatts sit at the final gate, and no operator has said which of them are theirs.

Ask on the next call whether guided 2027 energization dates assume an April 2027 study result ERCOT has already withdrawn.

Waiting for December 17 means repricing after everyone else hears the same date.

Operators face the most expensive choice. Post the security, answer the request for information inside the cure window, and price a fully islanded configuration in parallel rather than sequentially.

Developers who treat the pause as a reason to slow down concede allocation priority to those who use the same four months to lock generation and cure their filing.

The Doctrine Outlives The Audit

The audit ends. The curtailment doctrine does not.

December 10 closes a disclosure exercise Texas regulators designed to close, and the market will read that report as the end of the pause.

The durable change came July 24, when the PUCT ruled that emergency curtailment is measured against grid needs, not paired generation, and applied the condition for the asset’s economic life.

Every behind-the-meter structure in Texas now carries a shutdown obligation its financial model was probably not built to hold.

The question for the next two quarters is whether the first co-located project to trade does so at a discount to its islanded neighbor, because that is the moment the doctrine gets priced rather than discussed.

Join Global Data Center Hub