On 28 March 2025 CoreWeave listed on Nasdaq as "the AI Hyperscaler." Eleven months earlier its S-1 had said the vast majority of revenue came from multi-year take-or-pay contracts, and that Microsoft was 62% of 2024 revenue. A hyperscaler that loses one tenant does not have a platform. It has a lease.
Sovereign AI debt is about who funds the hall. National compute programmes are about what governments count. This piece is about the private company in the middle: the GPU landlord that borrowed against a chip and a name.
Tap to enlargeWhat the word is for
SemiAnalysis put the label on the page in 2024: a cloud that rents GPU compute. No CRM. No object-storage empire. GPUs, a fast network, an invoice. The four names it treated as giants were CoreWeave, Lambda, Crusoe, and Nebius.
AWS is hundreds of products, including P5 GPU instances that sit inside a much larger catalogue. A neocloud is a rack and a contract. If the contract is take-or-pay, the customer pays whether the cluster is busy. If the loan is secured by that contract, the tenant is the credit.
Cloud implies many customers, many products, and a platform that survives a lost account. The S-1, the 10-Q concentration table, and the delayed-draw loan say the opposite: one tenant class, one SKU family, and a special-purpose borrower. The credit agreement is the document. The homepage is marketing.
The word then spread to anything with a GPU and a press release. Three products get mixed under it.
GPU hours. CoreWeave, Lambda, Nebius, Crusoe Cloud, IREN's Microsoft cloud deal. The product is committed capacity on named NVIDIA SKUs. CoreWeave's weighted contract term moved from about four years to about five.
Hall megawatts. Applied Digital and Hut 8 sign 15-year take-or-pay leases of IT load to unnamed investment-grade hyperscalers. The tenant brings the chips. Equinix and Digital Realty have sold that product for years. An AI paint job does not make it a GPU cloud.
Inference APIs. Together, Fireworks, Groq, Cerebras. You buy tokens, not a rack. I already put those names in the agent stack. Mixing them into "neocloud" hides the credit.
What CoreWeave's filings show
CoreWeave is a US-incorporated GPU landlord with an S-1, a 10-K, and a 10-Q trail. Nebius is also on Nasdaq. Use CoreWeave as the example with the longest US filing set, not as the whole market.
Revenue: $16 million in 2022, $229 million in 2023, $1.915 billion in 2024, then $5.131 billion in 2025. Net loss: $863 million in 2024, $1.167 billion in 2025. Interest expense in 2025 was $1.229 billion. Adjusted EBITDA that year was $3.093 billion. GAAP net loss and adjusted EBITDA are not the same number. Interest is why they diverge.
The S-1 was blunt about concentration. Largest customer: 16% of 2022, 35% of 2023, 62% of 2024. The 2023 and 2024 name was Microsoft. Top two customers: 77% of 2024. CRN's S-1 read-through is still a clean secondary walk through those concentration rows. The Q1 2026 10-Q still will not print the names in the concentration table. Customer A was 72% of Q1 2025 revenue and 45% of Q1 2026. Customer B was 20% in Q1 2026. Committed contracts, including capacity delivered before the start date, were 98% of revenue in both of those quarters. On-demand is the leftover.
Tap to enlargeThe S-1 also said current customers had "contractually specified our use of NVIDIA GPUs." The contract assumes NVIDIA. NVIDIA is the chip vendor, a shareholder, and, as of the IPO prospectus, a customer. CNBC and Reuters reported that NVIDIA anchored the downsized IPO with a $250 million order at $40 a share. That is press, from people familiar with the order. The public facts are the $40 price, $1.5 billion raised, a muted close at $40, and a fully diluted valuation near $23 billion.
Reuters also recorded the origin: Atlantic Crypto, an Ethereum miner, founded in 2017. The Merge in 2022 cut mining rewards. The GPUs stayed. The invoice changed.
The loan is against the tenant
In August 2023 Blackstone and Magnetar led a $2.3 billion GPU-backed facility. In May 2024 a $7.5 billion delayed-draw term loan followed, same leads, plus Coatue, Carlyle, CDPQ, DigitalBridge, BlackRock, Eldridge. On 30 March 2026 CoreWeave Compute Acquisition Co. VIII, LLC signed DDTL 4.0: an $8.5 billion facility "entered into primarily to finance capital expenditures required to perform a customer contract, including the acquisition of GPU servers." The press release called it the first investment-grade rated financing "secured by HPC infrastructure and an associated customer contract." Moody's A3. DBRS Morningstar A (low). The investor-relations copy matches the 8-K exhibit. Floating tranche SOFR + 2.25%. Fixed tranche about 5.9%. Maturity March 2032. Anchored by Blackstone Credit & Insurance.
Moody's rated a special-purpose borrower, the chips inside it, and one customer's promise to pay, not CoreWeave the brand. A3 is not a claim that GPUs are now an investment-grade asset class. Strip the named tenant and the same racks do not get A3. The "first investment-grade GPU financing" is tenant credit with a chip lien. Nebius said the same thing in plainer English a year earlier.
Nebius wrote it into a 6-K dated 8 September 2025. Microsoft would take dedicated GPU capacity in Vineland, New Jersey, over five years. Total contract value about $17.4 billion through 2031, or about $19.4 billion if Microsoft took the extras. Cash from the deal would finance part of the capex. The rest: "the issuance of debt secured against the contract in the near term, at terms enhanced by the credit quality of the counterparty." The tenant's rating is the financing.
The Meta Infrastructure Services Agreement exhibit is the contract text. $12 billion of dedicated five-year clusters, NVIDIA Vera Rubin, deployments from early 2027. A further order of up to $15 billion: Meta buys unsold capacity on certain clusters if Nebius does not sell it to someone else. The 6-K is the summary. Nebius said it intends to sell that slice to third parties first. The exhibit even talks about a bankruptcy-remote structure for the financing. Up to $27 billion is the ceiling. $12 billion is the dedicated floor. Do not mash them.
The Q1 2026 10-Q put current debt at $7.547 billion and non-current at $17.312 billion. That is about $24.9 billion of debt on $55.6 billion of assets, of which property and equipment was $36.4 billion. Most of the assets are GPUs and leases.
Backlog is not remaining performance obligation
At year-end 2024 CoreWeave had $15.1 billion of remaining performance obligations, up from $9.9 billion a year earlier. RPO is a GAAP term. It is contracted, not yet recognised, subject to remaining performance.
By 31 December 2025 the company was leading with "revenue backlog" of $66.8 billion. The footnote is the analysis: remaining performance obligations, plus other amounts the company estimates will be recognised under committed contracts, in each case subject to delivery and availability of service. By 31 March 2026 that figure was $99.4 billion. The annual-report letter said no single customer was more than 35% of backlog at year-end 2025, against 85% at the start of the year. The FY2025 10-K repeats the Platinum ClusterMAX claim and the revenue series. Backlog mix is not revenue mix. A new Meta or OpenAI order can dilute Microsoft in the backlog while Microsoft still pays this quarter's invoice.
The S-1 already warned that useful-life estimates for GPUs might be wrong, and that customers might not let the company redeploy old SKUs. Weighted average committed contract length moved from about four years to about five, per the year-end letter. Aircraft lessors have run this structure for decades: buy the asset, attach a lease to an airline, borrow against both, live on what the airframe is worth when the lease ends. A 737 still flies in year twelve. A training SKU the lab treated as current in year one is last year's chip by year three if a successor exists. The customer can still owe under take-or-pay. The rack can still be the wrong rack. The landlord's equity is the residual. The lender's protection is the tenant's name, not the resale market for used H100s.
OpenAI is the other named ceiling. CoreWeave announced an agreement of up to $11.9 billion on 10 March 2025, plus $350 million of stock to OpenAI at the IPO. Reuters confirmed the size the same day. A May expansion of up to $4 billion, then a later expansion of up to $6.5 billion, brought the company's own total to "up to approximately $22.4 billion." "Up to" is a cap. It is not cash received. Q1 2026 added a new $21 billion Meta commitment in March and a multi-year Anthropic agreement. Named ceilings stacked on the same balance sheet. Delivery still has to happen.
On the 26 February 2026 earnings call, management said 2026 capex would be at least $30 billion, more than twice 2025, to support the contracted book. That is guidance on a call, not a 10-K line. Treat it as intent.
The hyperscaler is also the tenant
Microsoft still runs Azure. It also rents GPU hours from firms that borrowed against a Microsoft contract. Those hours show up as CoreWeave or IREN revenue, and as Microsoft opex, not as Microsoft-owned servers on this quarter's balance sheet. The company has not published a memo that says why. The public record is the contracts: CoreWeave, Nebius in Vineland, IREN in Childress. Press has also tied Microsoft to Nscale's Portugal site. Treat that last one as reported, not as a 10-K line.
NVIDIA already sits on more than one side of CoreWeave: vendor, shareholder, and, in the IPO prospectus, a customer. The landlord needs chips. The lender needs the tenant to pay. NVIDIA needs both to keep buying. Those roles can conflict.
On 3 November 2025 IREN said it had signed a five-year GPU cloud contract with Microsoft of about $9.7 billion, including a 20% prepayment, and a Dell equipment order of about $5.8 billion. On 13 August 2026 it said Horizon 1, the first of four 50 MW liquid-cooled phases at Childress, Texas, had been delivered and accepted. Acceptance starts the service term. The other three phases are still a schedule. A signed ceiling is not a live hall. IREN later said it had signed $2.8 billion of additional multi-year AI cloud contracts and lifted a 2026 AI-cloud annualised-run-rate target above $4 billion. That is a company target, not recognised revenue.
The private names repeat the same three objects without an S-1. Lambda said it raised over $1.5 billion Series E in November 2025; Forge puts the round at $5.9 billion post-money. There is no S-1 as of this writing. Sacra's revenue estimates are a research house's model. Use the raise. Crusoe closed an initial $1.375 billion Series E in October 2025 at a valuation above $10 billion, with NVIDIA in the list. The Abilene construction loan already sits in the debt essay. Do not count it twice. Nscale was reported in March 2026 at a $14.6 billion valuation after a $2 billion Series C, with a Portugal build tied to Microsoft. That is a magazine's round-up. Same ingredients. Not a 10-K.
Two products that share a ticker story
Bitcoin miners already had interconnection queues and cheap power. When hashprice fell and H100 rents rose, they used the same sites for GPU halls.
Hut 8's Nueces County, Texas campus is now a 15-year, $9.8 billion lease for 352 MW of IT load, the second phase with the same unnamed investment-grade tenant. Reuters covered the first 352 MW take-or-pay in May. The July follow-on takes the site to 704 MW and $19.6 billion of base-term contract value. First halls for phase 2 are expected in Q2 2028. Applied Digital's Delta Forge 2 lease is 210 MW, 15 years, about $5.2 billion base-term, third deal with the same unnamed hyperscaler. An April lease at Delta Forge 1 was $7.5 billion for 300 MW. Applied Digital said the contracted portfolio was 1.4 GW of critical IT load and about $36 billion of base-term lease revenue.
Those numbers are large. They are not GPU-hour revenue. A 15-year MW lease is a building product. A 5-year GPU take-or-pay is a chip product. Analysts file both under neocloud because both used to mine bitcoin. The depreciation, the residual, and the rollover are different. A hall can host the next SKU. A financed H100 cluster cannot, not without another draw.
Power is the scarce input either way. CoreWeave said it had more than 850 MW of active power at year-end 2025 and about 3.1 GW contracted. The Q1 2026 release said active power had passed 1 GW. IREN's February filing described 810 MW of operating data centres and more than 4.5 GW of secured power, with an AI cloud target of 480 MW in 2026. Interconnection queues, transformers, and liquid cooling are the lead times. NVIDIA can ship a SKU faster than a utility can energise a feeder.
SemiAnalysis ClusterMAX is the industry's own scorecard for whether a cluster actually trains. CoreWeave says it is the sole Platinum in the 2025 ranking. That is a performance claim about the fabric, not a credit claim. A Platinum cluster with one tenant is still one tenant.
Hall, tenant, residual
The A3, the backlog, and the homepage treat three separate conditions as one story: the hall exists, the tenant pays, and the rack is still worth something when the term ends.
The hall has to exist on the date in the contract. CoreWeave's 2026 capex guidance of at least $30 billion is the delivery bill. IREN's three phases still on a schedule are the same risk in public. "Up to" converts when the cluster is accepted, not when the press release hits.
The tenant has to pay. That is Microsoft, Meta, or OpenAI credit, wrapped in a special-purpose borrower. CoreWeave the equity is what is left after that structure. Interest already ate $1.229 billion in 2025. Adjusted EBITDA of $3.093 billion is the line that ignores that bill.
The rack has to be worth something when the term ends, or the take-or-pay has to cover the debt even if nobody wants the SKU. Backlog mix can look diversified while this quarter's invoice is still one name. A government counting GPUs should not add CoreWeave's 850 MW to a sovereign scorecard. That is a private lease, not national compute.
The IPO year is already on the tape. The year that tests the structure is the first time a five-year take-or-pay ends on a GPU generation the lab no longer wants, and the special-purpose borrower still owes Blackstone. Until that happens, the homepage still says cloud. The credit agreement already put the risk on the tenant.