12 Data Center Reports Later, the Bubble Is the Wrong Question
Goldman Sachs, CBRE, and Brookings headline twelve reports converging on a 50-to-80 GW US power shortfall by 2030. Inside the phantom pipeline, the 72% on-time rate, and the 3x jobs overstatement.
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.
TL;DR
US data center power demand is projected to more than double from 31 GW in 2025 to 66 GW by 2027, while US generating capacity falls 50 to 80 GW short by 2030. Goldman Sachs projects data centers reaching 8.5 percent of US peak summer demand; BCG identifies gas-fired generation with carbon capture, up to $80 billion in US capex, as the most viable near-term firm-power bridge.
The announced data center pipeline is materially inflated: historically only about 72 percent of facilities scheduled within a four-quarter window come online on time, and developers file across multiple regions while building only one site. Utilities estimate power delivery 1.5 to 2 years later than developers expect. CBRE reports Northern Virginia vacancy at an all-time low of 0.3 percent even as constraints stretch timelines.
Brookings finds data center job creation is overstated by roughly a factor of three in industry-sponsored studies, and no report resolves whether AI demand converts to revenue. Colocation facilities deliver almost no local information-sector gains, which makes many local tax incentives poorly targeted. The deciding question for investors is how much announced capacity comes online on time, powered, and profitable.
We reviewed the 12 most consequential reports, investor analyses, and market outlooks shaping the global data center sector this quarter.
Each was selected for its impact on how operators deploy infrastructure, how investors allocate capital, and how policymakers frame the next phase of the AI buildout.
Across the period, three signals defined the market:
AI is the new power load.
Power is the new bottleneck.
Capital is the new variable.
Together, they describe an industry running at full throttle yet increasingly constrained by grid capacity, execution risk, community opposition, and unresolved monetization.
This is your Data Center Intelligence Briefing: a field guide to what the latest research reveals, where consensus is forming, and where critical uncertainties remain.
Here is what is inside:
A synthesis of the 12 reports shaping global data center strategy and how they redefine infrastructure, capital formation, and regional competition
The key trends driving capital discipline, power strategy, and the shift toward AI-centric, inference-led infrastructure
The opportunities emerging from power-secured development, clean-energy sourcing, and the telco AI Grid
The issues still constraining the next wave, from grid access and permitting to labor, water, and chip obsolescence
A clear view of where the market is heading and what investors, operators, and policymakers still need to solve to sustain the AI infrastructure buildout.
The 12 Most Influential Reports
Below are the most impactful research reports shaping global data center and AI infrastructure strategy this quarter, ranked by strategic relevance, capital impact, and long-term market influence.
1. Tracking Trillions: The Assumptions Shaping the Scale of the AI Build-Out
Source: Goldman Sachs
This report establishes the financial frame for the sector, sizing the AI build-out in the trillions of dollars.
It argues the total scale of investment is not fixed but hinges on four supply-side assumptions: hardware lifespan, data center complexity, chip architecture, and logistical bottlenecks.
Depreciation and power density can shift capital estimates by hundreds of billions, independent of demand.
Its core insight: the durability of the boom depends as much on these technical realities as on consumer adoption.
2. US Data Center Power Demand Projected to Double by 2027
Source: Goldman Sachs
Goldman projects US data center power demand more than doubling from 31 GW in 2025 to 66 GW by 2027, reaching 8.5% of total US peak summer demand.
It warns that supply chain shortages and labor deficits may delay nearly half of all planned facility activations.
The surge threatens regional grid stability and higher electricity costs, particularly in the Mid-Atlantic and Northwest.
3. Global Data Center Trends 2026
Source: CBRE
CBRE shows global supply reaching 16 GW across the 16 largest markets, up 25% year over year, yet average vacancy fell to 6.7% from 8.3%.
In the top four US markets, 80% of space under construction is already preleased, with Northern Virginia vacancy at an all-time low of 0.3%.
Power and grid constraints, plus public opposition, are curbing development timelines even as demand outruns record supply.
4. Global Data Center Market Comparison
Source: Cushman & Wakefield
Cushman reframes 2026 as a shift from expansion to managed growth, driven by financial self-regulation and stricter government oversight.
Developers increasingly pursue nuclear and private generation to bypass long utility wait times, while secondary and tertiary markets gain relevance on land and lighter regulation.
The report explicitly dismisses bubble fears in favor of disciplined, long-term profitability.
5. Solving the US Data Center Power Crunch
Source: Boston Consulting Group
BCG sizes a looming 50-to-80 GW US energy shortfall by 2030 and identifies gas-fired generation with carbon capture (CCUS) as the most viable near-term firm-power solution.
The approach lets developers secure fast, reliable, round-the-clock power while preserving low-carbon credentials, provided they move quickly on federal incentives.
It offers a roadmap prioritizing repeatable implementation and stakeholder engagement over one-off megaprojects.
6. 2026 Data Center Power Report
Source: Bloom Energy
Bloom Energy shows limited grid infrastructure fundamentally reorganizing the industry around a widening utility-developer timeline mismatch.
This is driving a strategic shift toward permanent onsite generation, gigawatt-scale campuses, and geographic reallocation toward power-rich regions like Texas.
It flags direct current distribution and other next-generation architectures as accelerating responses to the power boundary.
7. 2026 Asia Pacific Data Centre Trends & Outlook
Source: CBRE
CBRE examines how surging AI demand is reshaping APAC, with power scarcity and escalating construction costs as the defining bottlenecks.
It documents a geographic shift toward high-growth markets like India and Malaysia and the emergence of neocloud providers as a key demand source.
Advanced cooling and hyperscale expansion are reframing the region’s investment routes and locational dynamics.
8. The Outlook for Data Centers in Asia
Source: Goldman Sachs
Goldman identifies India, Japan, and the Philippines as high-potential growth markets while naming reliable power supply as the single largest obstacle to expansion.
It advocates public-private partnerships and innovative infrastructure designs that balance utility needs with sustainable cooling.
The report frames evolving customer requirements and regional regulation as the forces reshaping Asian capacity.
9. From Megawatts to Meaning: Why Cost per Token Matters in AI Data Centres
Source: Cushman & Wakefield
Cushman argues token-level metering is becoming the primary benchmark for performance and billing, replacing raw power capacity as the measure of work performed.
This gives operators a new competitive lever: tokens processed per megawatt, achieved through tighter integration of compute and networking.
The transition is complex given varying hardware efficiency but signals a structural repricing of the asset class.
10. New Evidence on Data Center Employment Effects
Source: Brookings
Using a synthetic control method, Brookings finds data center job creation is real but overstated by industry-sponsored studies by roughly a factor of three.
It draws a sharp distinction between hyperscale facilities, which build local technology ecosystems, and colocation facilities, which bring almost no local information-sector gains.
Its conclusion: local tax incentives are often poorly targeted, subsidizing the facilities with the smallest long-term benefit.
11. The AI Grid: A 2026 Playbook for Telco AI
Source: Nscale
Nscale outlines how telcos can evolve from data carriers into distributed AI infrastructure providers by converting last-mile networks and edge sites into a national AI Grid.
This solves the latency and data-sovereignty problems of centralized cloud, enabling secure, localized inference for regulated sectors like healthcare, defense, and finance.
It is a call to monetize existing real estate and power contracts through GPU-as-a-service and modular AI infrastructure.
12. Global Data Center Report 2026
Source: IDCA
IDCA provides the global baseline: capacity expanded to 67.7 GW, with data centers now consuming roughly 2% of world electricity and the US holding a dominant 43% share.
It reframes physical and digital security as equally vital against asymmetric threats in a geopolitically contested landscape.
The report serves as a strategic guide to a new era defined by energy constraints and immense capital intensity.
Key Trends
1. From Growth at All Costs to Capital Discipline
AI remains the primary driver of investment, but the narrative has shifted from scale to profitability. Investors are scrutinizing whether $4 trillion to $8 trillion in projected capex converts into repeatable free cash flow and pricing power.
These estimates are not fixed. They are highly sensitive to assumptions about chip lifespan and next-generation architecture costs.
In response, developers are recycling capital, spinning operational assets into fund and partnership structures to protect balance sheets and maintain liquidity.
2. Power as the Gating Variable
Access to reliable electricity has overtaken capital and land as the binding constraint on growth. Time-to-power is now the defining underwriting metric.
US generating capacity is projected to fall short by 50 to 80 GW by 2030, and utilities consistently estimate delivery will take 1.5 to 2 years longer than developers expect.
Developers are turning to onsite generation, storage, and hybrid energy, with more than a third of data centers projected to run on 100% onsite power by 2030.
3. The Phantom Pipeline and Execution Risk
The staggering figures for planned capacity are partly inflated. Developers file applications across multiple regions, then build only the single most favorable site.
Markets are suffering from elongation, the widening gap between capital deployment and live operations, driven by shortages of labor, transformers, and switchgear.
Historically, only about 72% of data centers scheduled within a four-quarter window actually come online on time. The headline pipeline deserves heavy discounts.
Key Opportunities
1. Power-Secured, High-Yield Development
Ground-up development offers the strongest returns where power can be secured early through onsite generation and long-term PPAs.
AI-focused facilities serving inference and regional demand are achieving higher yields despite greater capital intensity.
The clearest path to profitability currently runs through built-to-spec wholesale colocation, and in regulated markets, expanding existing facilities beats greenfield.
2. Clean Energy and the CCUS Bridge
The industry’s vocabulary shift from renewable to clean energy is a strategic admission that variable renewables cannot support 24/7 AI baseload.
Gas-fired power with carbon capture represents up to $80 billion in US capex, a natural opening for oil and gas majors seeking fast, reliable, low-carbon capacity.
Nuclear is emerging as the preferred scalable long-term solution, alongside immersion cooling and AI-driven workload optimization.
3. The Telco AI Grid and Sovereign Inference
Telcos hold the least-crowded opportunity in the set: converting distributed last-mile networks and edge sites into a national AI Grid.
Because they already operate trusted, regulated national networks, they are uniquely positioned to offer sovereign AI environments where sensitive data stays within borders.
The monetization window runs through GPU-as-a-service, localized inference platforms, and modular AI infrastructure.
Key Issues
1. Grid Deficiencies and Community Backlash
The 50-to-80 GW US shortfall and the 1.5-to-2-year time-to-power gap turn even funded projects into stranded capital.
US residential electricity rates rose roughly 32% between July 2020 and July 2025, largely driven by data center loads.
Local governments are responding with zoning restrictions, longer permitting, and moratoriums. The era of unchecked expansion is over.
2. Chip Obsolescence and Capital Risk
The short economic life of AI silicon creates transitional data centers that risk obsolescence within a few years of commissioning.
Facilities designed for 5-to-10 kW racks cannot support the 40 kW and higher densities modern AI workloads require.
This is why investors are repricing execution and demanding a credible path to free cash flow before committing capital.
3. The Unresolved Monetization and Jobs Questions
Every report assumes demand converts to revenue. None resolves what happens if enterprise adoption lags or model efficiency improves faster than expected, weakening utilization.
Brookings’ finding that job creation is overstated by 3x directly conflicts with industry studies and with more favorable reads on tax incentives elsewhere in the set.
The bubble question is the wrong question. The right one is narrower: of the capacity announced, how much will actually come online, on time, powered, and profitable?