Quarterly Earnings: CoreWeave Rents the Balance Sheet, IREN Owns It
CoreWeave borrowed $8.5 billion at investment grade on Meta's credit and Blackwell collateral in Q1 2026. IREN borrowed against a projected $3.7 billion ARR. Two answers to who holds AI capital risk.
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
CoreWeave closed an $8.5 billion GPU-backed term loan in Q1 2026, the first such facility to receive investment-grade ratings (A3/A-low). The debt sits in a bankruptcy-remote entity backed by Blackwell GPUs and Meta’s contracted cash flows, letting CoreWeave borrow at investment-grade cost without being investment grade itself. Revenue more than doubled to $2.08 billion while active capacity passed 1 GW.
IREN retained full balance-sheet risk, owning land, power, data centers, and GPUs while converting Bitcoin mining capacity to AI. AI Cloud Services revenue grew 94% sequentially, and a $3.4 billion Nvidia agreement plus an existing Microsoft partnership back a projected $3.7 billion ARR by fiscal 2027. The tradeoff is higher long-term margins in exchange for carrying the debt hyperscalers avoid.
Below the hyperscaler tier, capital structure is the business model, not a financing choice, and neither model is proven through a full GPU depreciation cycle. CoreWeave’s structure depends on Blackwell GPUs holding collateral value; IREN’s depends on energizing capacity before debt matures. If GPU residual values disappoint, both models reprice at once.
The AI infrastructure cycle has largely been viewed through hyperscaler balance sheets, where demand was never the constraint.
Capital determined the pace of expansion, but financing remained a strategic choice rather than a necessity.
Alphabet redirected surplus cash into capex, Microsoft invested against a massive backlog, and Meta funded gigawatt-scale campuses from advertising cash flow.
Each could optimize capital allocation because strong cash generation supported growth.
That framing changes one tier down.
CoreWeave and IREN face the same constraints power, silicon, cooling, and grid interconnection but lack mature cash engines to fund expansion internally.
For hyperscalers, capital structure is a lever; for neoclouds, it is the business model.
Q1 2026 highlights this divide: CoreWeave shifted risk onto a tenant’s credit, while IREN retained it on its own balance sheet.
Together, they illustrate the full range of financing strategies once the hyperscaler safety net disappears.
CoreWeave: Renting the Balance Sheet
CoreWeave’s defining Q1 event was not its backlog or earnings, but its financing.
On March 31, it closed an $8.5 billion delayed-draw term loan (DDTL 4.0), the first GPU-backed facility to receive investment-grade ratings (A3/A-low), priced far below the ~15% rates of similar financings just three years earlier.
The breakthrough was structural: the debt sits in a bankruptcy-remote entity backed by Blackwell GPUs and Meta’s contracted cash flows, allowing investment-grade funding without making CoreWeave itself investment grade.
Its infrastructure strategy complements this model by converting powered shells instead of building greenfield campuses, helping it surpasses 1 GW of active capacity while expanding contracted power beyond 3.5 GW.
However, litigation tied to Core Scientific projects highlights execution risk.
Financially, the company reflects the cost of rapid expansion.
Revenue more than doubled to $2.08 billion, but heavy interest expense, high leverage, and negative free cash flow underscore that financing not operations is the central challenge.
Unlike hyperscalers that fund growth from surplus cash, CoreWeave depends on capital markets to sustain expansion.
Its lasting advantage is therefore not GPU access but the cost of capital.
Investment-grade financing enables pricing competitors with higher borrowing costs cannot easily match.
Yet the same relationship that underpins this model Meta’s long-term commitment also creates its greatest risk if the customer increasingly shifts workloads to its own AI infrastructure.
Read the full CoreWeave analysis: The $8.5B Quarter That Made GPU Debt Investment-Grade
IREN: Carrying the Stack
IREN represents the opposite financing model from CoreWeave.
Rather than relying on a tenant’s balance sheet, it owns the entire AI infrastructure stack land, power, data centers, and GPUs.
This full-ownership approach is capital-intensive but captures the full economic value of each megawatt instead of earning only a lease spread.
Its quarter appeared weak on the surface, with revenue below expectations and a wider net loss, but the decline reflected a strategic transition rather than operational weakness.
IREN retired Bitcoin mining hardware to free capacity for AI, resulting in impairment charges and lower mining revenue while AI Cloud Services revenue grew 94% sequentially.
Positive adjusted EBITDA and operating cash flow underscored the strength of the underlying business.
The company’s value lies in its contracted infrastructure, not its current earnings.
A $3.4 billion Nvidia AI Cloud agreement expanded equity alignment with Nvidia, and an existing Microsoft partnership validate its strategy of converting previously secured power assets into AI capacity.
Like CoreWeave, IREN’s balance sheet is its business model.
It is funding heavy capital investment against future contracted revenue rather than existing cash flow.
The tradeoff is clear: higher long-term margins in exchange for assuming the balance-sheet risk that traditional infrastructure operators avoid, with its key competitive advantage remaining scarce, grid-connected power secured years before today’s AI demand.
Read the full IREN analysis: The $3.7B ARR Against a $3.7B Debt Stack Inversion
The Same Constraint, Opposite Structures
Read together, CoreWeave and IREN tell one story from opposite ends of the same financing spectrum.
Both face the same constraints power, infrastructure, and strong AI demand but diverge on a single question: who holds the capital risk.
CoreWeave transfers it through a bankruptcy-remote structure backed by Meta’s credit and Blackwell GPU collateral, enabling investment-grade financing.
IREN retains the risk, financing owned infrastructure with convertibles and GPU-backed debt against future revenue.
The lesson for operators below the hyperscaler tier is clear.
Without the cash flow that lets hyperscalers absorb volatility, capital risk must be engineered through financing.
CoreWeave outsources balance-sheet risk to its tenant, while IREN keeps it in-house.
Both are responses to the same infrastructure scarcity; they differ only in who bears the risk if demand weakens.
What Resolves Next
Three signals will determine whether either model endures.
First is the durability of CoreWeave’s investment-grade financing.
Its A3 GPU-backed facility set a benchmark, but its later DDTL 5.0 financing priced below investment grade, leaving open whether the earlier rating reflects a repeatable structure or favorable timing.
Second is IREN’s commissioning pace, as bringing capacity online on schedule is essential to converting its projected $3.7 billion ARR into fiscal 2027 revenue.
Third is GPU depreciation.
No hyperscale deployment has completed a full asset cycle, so the residual values underpinning both companies remain unproven.
For investors, valuation has shifted from operating results to financing durability and free cash flow.
For operators, investment-grade tenants and secured power are now prerequisites for low-cost capital.
For policymakers, concentration risk is deepening as hyperscalers increasingly drive demand, support financing, and compete within the same capital stack.
Hyperscalers proved abundant cash makes financing strategy a choice.
CoreWeave and IREN show that without it, capital structure becomes the business itself.
The Bottom Line
The AI infrastructure cycle has moved beyond hyperscaler spending.
Demand is contracted, capital is committed, and power and components remain the bottlenecks.
Once the mature cash engine is removed, the key question becomes who bears the capital risk and how it is financed.
CoreWeave and IREN offer two contrasting answers.
CoreWeave transfers risk into a bankruptcy-remote structure to access investment-grade financing, while IREN keeps the risk on its own balance sheet, borrowing against revenue that will not fully materialize until fiscal 2027.
One is a counterparty bet; the other is a timing bet.
Neither model has been tested through a full depreciation cycle.
CoreWeave’s structure depends on Blackwell GPUs retaining collateral value, while IREN’s depends on its energized fleet generating returns before debt matures.
Both ultimately rely on the same unproven assumption: that GPU-backed assets will preserve their value across a full investment cycle.