1. CoreWeave capex and the customers standing behind it
CoreWeave capex is now mostly underwritten by the strongest names on its client list. The AI cloud provider said on November 10 that investment-grade and enterprise customers account for more than 75% of its committed capital expenditure through 2027, a disclosure aimed at investors weighing roughly $35 billion in debt against 2026 spending guidance of $37 billion to $39 billion.
The statement addresses the question that follows every company building data centers at this pace: who ultimately pays for the equipment. CoreWeave's answer is that the bulk of what it has promised to spend is matched to customers with high credit quality or established corporate balance sheets, and that the spending is tied to agreements already signed.
That framing matters because the company's model depends on borrowing heavily today against income that arrives over several years. The more reliable the customers on the other side of those contracts, the easier it is to persuade lenders and shareholders that the money will come back.
The disclosure also speaks to a broader debate about the AI build-out. Data center operators across the industry are committing enormous sums to hardware on the expectation that demand for computing will keep growing. Investors have asked how much of that spending is supported by firm contracts with dependable buyers and how much rests on optimism. By putting a number on the share of its commitments tied to its most creditworthy customers, CoreWeave is offering a direct answer for its own program.
2. How the business model works
At its core, the company is a landlord for computing power. It assembles large clusters of graphics processing units, the specialized chips used to train artificial intelligence models and to run them once trained, and rents that capacity to customers that would otherwise have to build it themselves.
The economics are front-loaded. Chips have to be purchased, data center space fitted out and power connected before a customer can use any of it. Revenue then accrues over the term of a contract. To bridge the interval, the company raises debt, which is why the identity of its customers and the terms of their contracts are scrutinized as closely as its revenue.
As of the second quarter of 2026, CoreWeave reported a revenue backlog of about $104 billion. Backlog is the value of contracted business that has not yet been delivered and recognized as revenue. Most of that total comes from long-term agreements structured on a take-or-pay basis.
Under a take-or-pay contract, the customer agrees to pay for a set amount of capacity whether or not it ends up using all of it. The arrangement shifts utilization risk from the provider to the buyer. For CoreWeave, it means that income from a given cluster does not depend on how busy that cluster is in any particular month, which gives the company a clear view of future revenue and gives lenders a predictable stream of payments to assess.
The same feature makes the contracts useful as a foundation for borrowing. A lender financing a cluster of chips wants to know that payments will arrive on a fixed schedule for the life of the loan. An agreement that obliges the customer to pay irrespective of usage provides that assurance in a way that pay-as-you-go cloud arrangements do not, which is why long-term commitments of this kind sit at the center of the company's financing.
3. What does investment grade mean for CoreWeave's contracts?
Investment grade is a credit rating category assigned to borrowers judged to have a low risk of default. For CoreWeave, a contract with an investment-grade customer is a promise of payment from a company that rating agencies consider highly likely to honor its obligations.
The label matters more in this business than in most. A take-or-pay agreement is only as good as the party that signed it. If a customer runs into financial trouble, a contractual commitment to pay for unused capacity offers limited protection. A commitment from a large, well-capitalized corporation is a different proposition, and it is the kind of counterparty lenders prefer to see when they finance the equipment involved.
Enterprise customers form the second category in the company's figure. These are established corporations buying computing capacity for their own operations, as distinct from young AI developers whose ability to pay may depend on continued fundraising. Grouping the two together captures the part of the customer base whose commitments are least likely to be disrupted by a downturn in venture funding or a shift in sentiment toward AI start-ups.
CoreWeave's client list includes some of the largest companies in technology. Meta and OpenAI are among its customers, and Microsoft and Nvidia also appear on the roster. The company's latest figure groups investment-grade customers together with enterprise clients and puts the combined share above 75% of committed capital spending through 2027.
The flip side of that number is the remainder. Somewhat less than a quarter of committed spending is linked to customers outside those two groups, and that slice carries more risk than the rest.
4. The financing behind the build-out
Credit quality on the customer side feeds directly into how the company funds itself. CoreWeave secured an $8.5 billion delayed-draw term loan to support its infrastructure, a facility built around the equipment and contracts that the money pays for.
A delayed-draw term loan differs from a conventional loan in its timing. Instead of receiving the full sum at closing, the borrower can take the money in stages as it is needed. For a company installing hardware across many sites over many months, that structure avoids paying interest on cash that is not yet being put to work, and it lets borrowing track the actual pace of construction and delivery.
The scale of the spending explains the need. The company's capital expenditure guidance for 2026 stands at $37 billion to $39 billion, and nearly all of it is linked to signed agreements. In other words, the company describes its spending as a response to contracted demand, not a speculative bet that customers will appear once capacity exists.
The obligations are nonetheless large. Total debt stood at about $35 billion as of June 30, 2026. Debt of that size has to be serviced regardless of who the customers are, and the strength of the client base reduces the risk without removing it.
5. Power as the next constraint
Electricity, more than chips or capital, is emerging as the limit on how quickly contracted demand becomes revenue. CoreWeave reports active power of more than 1.7 gigawatts and contracted power of more than 4.6 gigawatts.
The gap between those two figures is the company's pipeline. Active power is capacity that is energized and able to serve customers. Contracted power is capacity the company has secured rights to but has not yet brought fully online. The difference represents a substantial volume of data center capacity that is committed and still to be delivered.
Closing that gap is what turns backlog into billings. A customer contract cannot generate revenue until the hardware behind it is installed and powered. Delays in connecting sites to the grid, or in completing the facilities themselves, would push revenue further out while the associated debt continues to accrue interest. That makes the timetable for bringing contracted power online one of the most consequential variables in the company's outlook.
Power has become a bottleneck for the sector as a whole because the newest AI hardware draws far more electricity per rack than conventional servers, and utility connections take time to arrange. A provider can have chips on order and customers under contract and still be unable to deliver if the electricity is not available at the site. For CoreWeave, the 4.6 gigawatts it has contracted is therefore both an asset and an execution challenge.
6. The risks that remain
A stronger customer mix does not eliminate the pressure points in the model. Three stand out: the debt load, the portion of spending not covered by top-tier clients, and reliance on a small number of very large buyers.
On debt, roughly $35 billion is a heavy obligation under any assumptions about customer quality. Interest and principal fall due on a fixed schedule, while revenue depends on capacity being delivered on time.
On the remainder, the share of committed capital spending that is not tied to investment-grade or enterprise customers is where a default or renegotiation would be most likely to occur. It is a minority of the total, and still a meaningful sum given the size of the overall program.
On concentration, the company's biggest customers remain very large relative to the whole. A backlog of about $104 billion that leans on a handful of major contracts leaves CoreWeave exposed to the decisions of those buyers, which is a dependency investors will continue to track even as the client base broadens.
None of these points contradicts the company's central claim. They describe the conditions under which a business with strong contracted demand could still come under strain: capacity arriving late, costs running above plan, or a major customer changing course. The 75% figure narrows those risks and leaves the task of managing them in place.
7. What to watch next
Three questions will determine whether the company's reassurance holds up. Each can be measured against figures CoreWeave has already put on the record.
The first is power delivery: whether the company can bring its 4.6 gigawatts of contracted capacity online on schedule. The second is spending discipline: whether 2026 capital expenditure lands within the $37 billion to $39 billion range it has guided to. The third is diversification: whether it can keep adding customers and reducing its reliance on its largest accounts without relaxing the credit standards that underpin the 75% figure.
For now, the company's message is that its expansion is anchored to contracts with buyers able to pay. The coming quarters will show how quickly those contracts convert into operating capacity and cash, and whether the balance between contracted income and borrowed money continues to move in its favor.