Companies

CoreWeave Downgraded: 4.6 GW of Power, $13.4B of Debt, and a 2030 Deadline

CoreWeave hit 4.6 GW of contracted power and raised GPU rental prices 37.5%, but $13.4B in debt and $81.2B in lease commitments triggered a rating downgrade.

CoreWeave: Walking The Leverage Tightrope (Rating Downgrade)
CoreWeave: Walking The Leverage Tightrope (Rating Downgrade)AI-generated
By Marcus Bennett5 min read

Why it matters

  • CoreWeave reached 4.6 GW of contracted power in Q3 FY26, targeting 8 GW by 2030.
  • Average GPU rental prices rose 37.5% in Q3 FY26; Ampere and Hopper chips priced 30–50% higher.
  • Short-term contracts fetch up to $40 million per MW after Vera Rubin NVL72 deployments.
  • CoreWeave carries $13.4B in debt and $81.2B in lease commitments through FY27.
  • Roughly 300,000 GPUs come off contract in FY27–28; international sites hold 28% of contracted power.

CoreWeave now controls 4.6 GW of contracted data center power as of Q3 FY26, yet a rating downgrade signals that analysts view the company's expansion as a leverage problem as much as a growth story. The GPU cloud (a service renting NVIDIA-accelerated computing capacity to AI developers) is targeting 8 GW of capacity by 2030, and it is financing that doubling with an increasingly inventive mix of customer prepayments, asset-backed loans, and OEM agreements against $13.4 billion in debt and $81.2 billion in lease commitments.

The numbers behind the expansion are large on both sides of the balance sheet. Contracted power grew to 4.6 GW in Q3 FY26. The 2030 target is 8 GW. International data centers already account for 28% of contracted power, a share the analyst notes gives CoreWeave access to cheaper energy and geographic diversification while supporting the 8 GW goal.

Why did the rating come down?

The downgrade rests on financing risk rather than demand risk. CoreWeave carries $13.4 billion in debt and $81.2 billion in lease commitments through FY27. To service that stack while continuing to build, the company is deliberately reducing its reliance on unsecured debt and turning to what the analyst describes as creative financing: customer prepayments, asset-backed loans, and OEM agreements (financing arranged with equipment manufacturers such as NVIDIA's hardware partners).

Each mechanism shifts risk in a different way. Prepayments bring cash forward but tie CoreWeave to delivery schedules. Asset-backed loans collateralize GPUs and data center infrastructure, meaning the hardware itself underwrites the debt. OEM financing spreads capital expenditure across vendors. The analyst frames this combination as the tightrope of the title: the structure works while cash flow grows, and it strains if the growth assumptions break.

What is driving the margin expansion?

Pricing power is the strongest pillar of the bull case. CoreWeave raised average GPU rental prices by 37.5% in Q3 FY26. Ampere and Hopper generation chips — the prior architecture to NVIDIA's current lineup — are priced 30% to 50% higher than before. The analyst attributes the uplift to recent deployments of the Vera Rubin NVL72, NVIDIA's next-generation rack-scale system, and to scarcity pricing on short-term contracts, which are fetching up to $40 million per MW.

That figure is the most striking in the analysis. At $40 million per MW on short-term deals, rented capacity converts to cash at a rate that supports further capital expenditure, which in turn feeds the 8 GW buildout. The higher Ampere and Hopper prices boost margins and cash flow available for future CapEx, according to the analyst's summary of the quarter.

The aging-chip question sits underneath the pricing story. If older-generation GPUs command 30% to 50% premiums today, that reflects current scarcity of accelerated computing. It does not guarantee that the same chips hold their pricing once newer architectures flood the market.

What are the three risks the analyst flags?

The downgrade analysis identifies a concentrated set of threats to the model:

  • Contract roll-off. Roughly 300,000 GPUs come off contract in FY27 through FY28. CoreWeave must re-lease or replace that capacity, and the terms it obtains will test whether current pricing is durable or cyclical.
  • Demand shift on older silicon. If customers migrate demand away from older chips toward newer architectures, the revenue attached to expiring contracts may not return at the same rate, pressuring the cash flow that services the $81.2 billion in lease obligations.
  • Post-leadership-change uncertainty at NVIDIA. Following NVIDIA's leadership change, CoreWeave's access to allocation, pricing, and OEM-financing terms — all central to its supply model — carries an additional layer of uncertainty. CoreWeave's business is structurally dependent on its chip supplier, and the analyst treats a change at the top of that supplier as a live variable rather than a settled one.

How does the international footprint change the math?

The 28% share of contracted power held by international data centers does double work in the analysis. First, it provides access to cheaper energy, which lowers operating cost per MW across a portfolio that must scale from 4.6 GW to 8 GW in under four years. Second, it diversifies growth across jurisdictions, reducing single-market exposure on both power pricing and customer demand. The analyst presents the international buildout as a structural support for the 2030 capacity target rather than a side initiative.

What does the balance sheet look like through FY27?

The core of the downgrade is the interaction between obligations and cash flow through FY27. The company faces:

  • $13.4 billion in debt.
  • $81.2 billion in lease commitments.
  • A financing strategy moving away from unsecured borrowing toward prepayments, asset-backed loans, and OEM agreements.

The analyst's implicit test is straightforward. Margin expansion from the 37.5% price increase and the $40 million per MW short-term contracts must generate enough cash to cover the obligation stack while the 300,000-GPU contract roll-off plays out. If Ampere and Hopper demand holds and Vera Rubin NVL72 deployments continue to command premium pricing, the model compounds in CoreWeave's favor. If demand shifts toward newer chips faster than expected, the leverage that funded 4.6 GW becomes the constraint on reaching 8 GW.

What happens between now and 2030?

The analyst's forward frame runs through FY27–28, when the GPU contracts expire and the refinancing strategy meets its first full test under whatever allocation regime emerges from NVIDIA after its leadership change. The path to 8 GW by 2030 depends on three variables the analyst quantifies or names directly: sustained rental pricing above the levels implied by the 30–50% increases on legacy chips, continued access to sub-$40 million-per-MW-competitive energy through the 28% international footprint, and OEM-backed financing that keeps capital expenditure off the unsecured balance sheet. The rating downgrade says the market has not yet been shown that all three will hold at once.

Source: seekingalpha.com

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

718 articles

Related articles

  1. CoreWeave's $104.2B Backlog Hides a Debt Bill Its Landlord Pays First
  2. CoreWeave Surges 16% After Q3 Beat-and-Raise Eases Debt Fears
  3. CoreWeave default swaps signal distress as AI debt anxiety spreads
  4. CoreWeave Doubled Revenue and Still Lost More Money: The Middleman Problem
  5. CoreWeave Stock Jumps 13% After Blowout Quarter and Raised Outlook

« Previous articleNext article »