CoreWeave

Nvidia supplies CoreWeave’s GPUs, owns a stake, and has agreed to buy unsold cloud capacity under a $6.3 billion order through April 2032. That three-sided relationship was already taking shape when CoreWeave priced its Nasdaq IPO on March 27, 2025. The company sold 37.5 million shares at $40, down from a plan for 49 million at $47–$55, raising $1.5 billion. Shares opened at $39 and closed the debut at $40.

CoreWeave rents Nvidia GPUs by the hour. It began in Ethereum mining and shifted to GPU cloud services. By mid-2023, it had 700 clients, including Microsoft, a $2 billion valuation, and fewer than 250 staff. Before the IPO, Nvidia held a 5.96% stake; it reportedly placed a $250 million anchor order. It made a further $2 billion investment in January 2026. See The AI Capex Supercycle for the spending cycle around the company.

TEXXR’s first cited CoreWeave item dates to November 2021. Coverage remained thin through 2023 and 2024, when the reported deals included $2.3 billion of Nvidia-collateralized debt and roughly $10 billion in Microsoft contracts running through 2030. A Bloomberg retrospective identifies Blackstone as lead on that $2.3 billion 2023 private-credit loan. Bloomberg calls it a blueprint for a chip-backed debt market it expects to reach $1 trillion; that forecast is market-wide, not CoreWeave debt.

CoreWeave’s S-1 coverage exposed the dependency: Microsoft supplied 62% of 2024 revenue, and the top two customers supplied 77%. CoreWeave denied a Financial Times report that Microsoft had pulled back from some agreements. Separately, Semafor reported that Microsoft passed on a nearly $12 billion capacity option that OpenAI picked up. The FT put debt due by end-2026 near $7.5 billion and compared the financing risk to WeWork’s.

New names followed. OpenAI signed an $11.9 billion contract for five years before the IPO and took $350 million of CoreWeave stock. Meta committed up to $14.2 billion in September 2025, then added $21 billion in April 2026 for 2027–2032. Anthropic signed a multiyear deal that April. Jane Street committed roughly $6 billion in compute spending and took a $1 billion equity stake. By May, OpenAI stood to hold about $2.6 billion of combined CoreWeave and Cerebras stock. It pays a supplier it partly owns; see Cerebras.

The expansion did not remove execution risk. CoreWeave agreed to buy Core Scientific for $9 billion in July 2025, but Core Scientific shareholders rejected it that October. CEO Michael Intrator reported construction delays at a data-center provider in November; CRWV fell 16.3% that session. In December, the Wall Street Journal counted a $33 billion market-value loss over six weeks and reported short-seller Jim Chanos’s criticism. By June 2026, Bloomberg counted $2.3 billion in founder stock sales since the August 2025 lockup. Its October 2026 report counted $3.3 billion sold by three co-founders since the March 2025 IPO, including $2.4 billion in 2026. The starting dates differ. On the supplier side, Nvidia CFO Colette Kress said August 2025’s $2.2 billion “net other income” was “driven by gains in a publicly-held equity security”; Sherwood News named CoreWeave. See Nvidia.

CoreWeave added customers. Nvidia supplies chips, holds shares, and commits to buy capacity. OpenAI buys capacity and holds shares. Expansion is documented; post-2024 customer concentration is not disclosed in the cited record.

TEXXR’s July 18, 2026 graph query counted 267 hyperedges—recorded relationships that can involve multiple entities—touching CoreWeave. Financial relationships were 52.4% of that extraction and partnerships 13.9%, versus controversy at 12.4% and acquisitions at 9.4%. A separate CoreWeave–Nvidia–Microsoft–OpenAI query found 98 recorded relationships over 24 months, with each entity connected to the others. Those counts describe the graph’s record, not transaction volume.

The graph also flags a November 2025 commentary claim that CoreWeave effectively served five customers: OpenAI, Microsoft, Google, Meta, and Nvidia. TEXXR marks it as commentary, not a confirmed customer disclosure. A newer October 4 interview identifies Richard Ahlfeld as CoreWeave’s SVP of Physical AI. He discussed failures when AI models meet real-world checks, and the roles of synthetic data and physical tests. The interview establishes a role and a technical focus, not a new capacity contract.

Positioning: how much of CRWV is sold short, on the record brokers file twice a month. It lags by design — read it as where the crowd stood, not where it stands.

Order flow is signed before it becomes revenue — the earliest of the three views. From the capital-flow ledger.

• Coverage, not demand. TEXXR’s quarterly count is the number of corpus items mentioning CoreWeave per calendar quarter. It recorded one item in each of 2023Q4 and 2024Q1, then 17 in the IPO quarter, 2025Q1. The latest complete quarter, 2026Q3, had five, a derived 50.6% below the prior eight-quarter average of 10.1. This is a change in coverage, not evidence that capacity demand changed.
• Customer shares and conversion. Hold the S-1 concentration baseline against the next disclosure. Q1 2026 revenue was $2.08 billion, up 112% year over year, with $99.4 billion of backlog. Q2 revenue was $2.58 billion, up 112% year over year, above the $2.56 billion estimate. Q2 backlog was $104 billion: a derived $4.6 billion, or 4.6%, above Q1. The test is revenue as power comes online. The company put active power at 1.5 GW, up nearly 500 MW, and contracted power near 3.7 GW.
• Financing cost. The FT described CoreWeave’s data-center financing through special-purpose vehicles. Its comparison sized CoreWeave’s amount at $2.6 billion, versus Oracle’s $66 billion and Meta’s $30 billion. Separately, Bloomberg reported in September 2026 that a CoreWeave-tied data center raised $1.1 billion in junk bonds at 98.5 cents on the dollar, yielding 9.25%—about 2.7 percentage points above similarly rated debt. Bloomberg described a tied data center, not a direct CoreWeave bond issue. In October, the New York Times reported CoreWeave bond yields above 15%. That is a different instrument from the tied data center’s September bonds, not a like-for-like yield change. CoreWeave told the Financial Times it was targeting insurers’ capital to fund chip spending more cheaply. Watch whether it secures financing, and on what terms.
• Cash against buildout. The DeadRisk market layer puts FY2025 revenue at $5.1 billion versus $1.9 billion in FY2024, a 168% rise. Operating cash flow was $3.1 billion versus $2.7 billion; capex was $10.3 billion versus $8.7 billion. Free cash flow was negative $7.3 billion versus negative $6.0 billion. The 10-K puts debt at $21.4 billion versus $7.9 billion, before leases. The DeadRisk market layer reports a different total debt series: $29.8 billion in FY2025 versus $10.6 billion in FY2024. Do not treat the two series as the same measure. Reuters reported in July 2026 that CoreWeave was exploring hedges against falling memory and storage-chip prices. The market provider’s August 26 calendar schedules earnings for November 11, 2026; that date can move.
• Price context. The DeadRisk market layer records an October 28, 2026 close of $63.93 versus $61.82 on October 21. Its October 28 reading was up 7.3% from the prior session but down 24.9% over 30 days and 52.6% over one year. From the October 2 close of $89.62, the October 28 close was down a derived 28.7%.

CoreWeave lives or dies by the same constraint as these names — small equal-weighted groups, one constraint each. Not investment products: see all categories.