When you put down a deposit on a GPU cluster or a reserved compute contract, you’re betting on a company, not on silicon. The GPUs are a commodity. You can spec them down to the part number. The provider behind them is not a commodity.
I’ve consulted on GPU cluster deals and built infrastructure, and the hardware is rarely the part that worries me. The company is. You don’t need to read a rack diagram to check it. There are six questions that separate a provider who can deliver and stay solvent for the term from one who can’t. Every one has a public, checkable answer. A refusal to answer tells you something too.
Who owns the metal?
Simple question first. Is the company on your contract the same company that owns the GPUs? Or is it reselling capacity it leases or subleases from someone else?
A broker isn’t automatically bad. But it adds a party you can’t see, and a margin stacked on the underlying deal. Ask whose balance sheet the hardware sits on. Ask who holds title if the provider defaults. Ask who signs the power and colocation contracts under the compute you’re buying.
If the answer passes through more than one company before it reaches the actual data center operator, get every name. Get comfortable with all of them, not just the one on your invoice.
Is the power priced and dated?
Power is the real constraint on capacity right now, not GPU supply. Goldman Sachs Research, publishing May 20, 2026, forecasts US data center power demand more than doubling from 31 gigawatts in 2025 to 66 gigawatts in 2027, with 41 gigawatts expected in 2026.
That demand is hitting a grid that can’t add capacity fast. Lawrence Berkeley National Laboratory’s “Queued Up: 2026 Edition,” covering the interconnection queue as of the end of 2025, counted roughly 8,200 projects waiting to connect, representing about 1,312 gigawatts of generation and about 749 gigawatts of storage. The typical wait is around five years from first application to commercial operation, and longer in the busiest zones.
So a provider without a signed, dated power or interconnection agreement is asking you to sit in that queue for them, on your timeline.
Ask for the contracted power in megawatts, the cost per kilowatt-hour, and the facility’s PUE. All dated, all in writing. Not described on a call.
What’s a normal PUE? The Uptime Institute’s 2024 Global Data Center Survey put the industry average at 1.56, flat for a fifth year in a row. That’s down from above 2.5 in 2007, but the industry as a whole has mostly stopped improving. If a facility built after 2020 reports a number well above that, ask why.
How is the hardware financed?
GPUs lose value fast, and providers borrow against them. Put those two facts together and you have the main solvency risk in this business.
CoreWeave’s own August 3, 2023 announcement disclosed a $2.3 billion debt facility, led by Magnetar Capital and Blackstone, collateralized by Nvidia H100 chips. It was one of the first times GPUs themselves were used as loan collateral at that scale.
Then, in a press release dated March 31, 2026, CoreWeave announced it had closed an $8.5 billion delayed draw term loan facility. The company says it was rated A3 by Moody’s and A (low) by DBRS, its first investment-grade rating for a financing secured by HPC infrastructure and an associated customer contract. It priced a floating-rate tranche at SOFR plus 2.25 percent and a fixed-rate tranche at approximately 5.9 percent.
Same kind of collateral, both deals public, roughly two and a half years apart. The difference is what a lender would charge once the borrower had a longer track record and stronger contracted revenue behind it.
Ask your provider the same questions. How much of the fleet is debt-financed? What’s the loan-to-value against what the hardware is worth today, not what they paid for it? What happens to your capacity if a lender forecloses mid-term?
A provider heavily levered against fast-depreciating GPUs, with thin contracted revenue behind the loan, is a weaker credit than its marketing page suggests.
How long have they done this specific thing?
Ask how long the company has run GPU infrastructure. Not compute in general, GPUs. Then ask for two or three referenceable customers at or above your scale who will actually take a call.
A provider that pivoted into AI hosting from another business isn’t automatically a red flag. It just changes what you check.
Core Scientific is a public example. It’s a Bitcoin miner that started hosting infrastructure in 2019, filed for Chapter 11 bankruptcy protection on December 21, 2022, citing falling crypto prices and rising energy costs, emerged in January 2024, and pivoted toward AI data center hosting.
The pivot isn’t disqualifying. Plenty of legitimate operators have repurposed power and space this way. But a bankruptcy in the track record is a fact you price in. Don’t skip it because the pitch deck starts after the restructuring.
Who depends on whom?
Concentration risk runs both ways. Ask what share of the provider’s revenue comes from its largest customer. Then ask what share of your own compute this provider would be.
CoreWeave’s S-1, filed with the SEC on March 3, 2025, disclosed that Microsoft accounted for 35 percent of its revenue in 2023 and 62 percent in 2024. That’s not a rumor. The company put it in its own filing. It’s exactly the kind of number a serious provider should share, or let you estimate from public filings if it already reports.
If a provider won’t talk about customer concentration at all, that’s your answer.
Now flip it. If you’d be a big share of their book, ask what happens to your pricing and priority when a bigger customer shows up next quarter.
What happens if they go under?
Get this in writing before you sign, not after you need it.
Bankruptcy doesn’t automatically mean the lights go out. Cyxtera, a colocation and interconnection provider running more than 60 data centers across more than 30 markets, filed a pre-arranged Chapter 11 in June 2023 in the District of New Jersey. It was backed by roughly $200 million in debtor-in-possession financing from lenders holding more than two-thirds of its term loan debt. It kept every facility running through the process.
That wasn’t luck. It happened because the filing was pre-arranged and the lenders were lined up in advance.
So ask:
- Does your contract survive assignment to a buyer or a lender who takes over the business?
- Is there a data escrow or export right if the provider stops operating?
- Who has a lien on the GPUs you’re paying to use, and does your service continue if that lienholder forecloses?
A provider that has thought about this will have a paragraph ready. One that hasn’t is asking you to find out during the bankruptcy. That’s the worst time to be asking for the first time.
The checklist, in the order to ask
- Is the entity on my contract the entity that owns the hardware, and if not, who is above them and what do their contracts say.
- What is the contracted power figure in megawatts, the price per kilowatt-hour, and the facility’s PUE, all dated in writing.
- How much of the fleet is debt-financed, against what loan-to-value, and what happens to my capacity if a lender forecloses.
- How long has this company operated GPU infrastructure specifically, and can I call two or three referenceable customers at my scale.
- What share of their revenue comes from their largest customer, and what share of my needs would this provider represent for me.
- What does the contract say happens to my capacity and my data if the provider files for bankruptcy or is acquired.
None of this needs an engineer. It needs you to ask the sixth question before the deposit goes out, not after you need the answer.
Once the racks are delivered, here’s how to tell if a GPU cluster actually works. For what that cluster should cost to run once it passes, see what it actually costs to run a GPU cluster.