CoreWeave's $8.5 billion investment-grade loan gave AI infrastructure a cheaper financing template, but the same story now carries a sharper question: whether contracted GPU revenue can support the debt load being built around it.
CoreWeave did not just raise another large check. It helped persuade a broader set of lenders that AI compute can be financed more like infrastructure than a speculative technology bet, provided the cash flows are tied to real customer contracts. That matters because the next phase of the AI buildout will be decided as much by the cost of capital as by who can get the newest Nvidia chips first.
The company announced an $8.5 billion delayed-draw term loan facility at the end of March, with Moody's assigning an A3 rating and DBRS Morningstar rating it A (low). As MarketWatch noted after the deal, the facility was unusual because it was backed by compute hardware, including GPUs, and by revenue tied to a major customer contract. The floating-rate tranche priced at SOFR plus 2.25 percentage points, while the fixed-rate tranche came in at 5.9%.
That rate is the point. It is not the kind of return investors demand when they see an unproven asset class. It looks closer to project finance, where lenders are focused on whether contracted cash flows can reliably service debt over time. For CoreWeave, which rents AI infrastructure to companies building and running large models, a lower financing cost directly changes the economics of adding capacity.
The structure also reflects how investors are trying to make sense of GPU clusters as collateral. A chip-heavy data center is not a toll road or a regulated utility. Hardware depreciates, model demand can shift, and customer concentration remains a real concern. But when capacity is paired with long-term commitments from creditworthy customers, lenders can begin to model it in a familiar way: asset, contract, repayment stream.
The financing story now cuts both ways
The more recent numbers show why this is still a live issue for investors. CoreWeave's contracted revenue backlog reached $99.4 billion in the first quarter, helped by an expanded Meta agreement worth about $21 billion through December 2032. That backlog gives lenders visibility, and visibility is what makes cheaper capital possible.
At the same time, the balance sheet has become harder to ignore. Barron's reported last week that CoreWeave posted a $740 million first-quarter loss while carrying about $25 billion in debt and another $10 billion in lease liabilities. The company expects capital expenditures of roughly $31 billion to $35 billion this year, largely funded through debt. Those figures do not invalidate the financing template, but they do make the test clearer.
For CoreWeave, the market is no longer only asking whether demand for AI compute is real. That part has been answered by the scale of contracts from Meta, OpenAI, and other large buyers. The harder question is whether the company can keep turning those contracts into cash fast enough to justify the leverage needed to build ahead of demand.
Applied Digital shows the same pressure from another angle. The company signed long-term leases with CoreWeave for capacity at its Ellendale, North Dakota campus, with total anticipated lease revenue rising to about $11 billion after additional agreements. That gives Applied Digital a clearer path to financing its data centers, while giving CoreWeave more capacity to meet customer commitments. The risk is that each participant in the chain is leaning on the next one to keep growing.
A template other AI infrastructure companies will copy
Every GPU cloud operator with meaningful backlog is now studying this playbook. The recipe is straightforward, even if execution is not: long-term customer agreements, specific pools of hardware or data center capacity, and financing structures that isolate lender exposure from the broader corporate balance sheet. Companies that can assemble those pieces may reach pools of institutional capital that were effectively closed to them two years ago.
That could push down the cost of AI infrastructure across the sector. A provider borrowing near investment-grade rates can bid more aggressively for power, build faster, and price capacity more competitively for enterprise customers. The advantage compounds with scale, especially when the same company can point to major customers and a growing secondary market for its debt.
But this is not a clean victory lap. The durability of AI infrastructure finance will depend on whether contracted revenue holds its value if model training slows, inference pricing falls, or customers renegotiate capacity needs. Lenders are accepting that GPU-backed cash flows can be investment-grade in the right structure. The next test is whether those structures still look sturdy when the AI cycle becomes less forgiving.
For now, CoreWeave has set the reference price. The market will be watching who can match it, and who has to pay far more to keep building.
Also read: China Is Turning the Compound Semiconductor at the Heart of Every AI Data Center Into Its Most Potent Trade Weapon • DDN Is Hunting a Strategic Investor and Its New Valuation Will Be a Referendum on AI Storage • Federal agencies now have three days to patch the most dangerous vulnerabilities and AI acceleration is the explicit reason
