Every CoreWeave customer contract with a disclosed value, term and filing, how the largest customer’s share fell from 67% (Microsoft, 2025) to 36% (an unnamed customer, Q2 2026), and what the numbers do and do not promise.
Market risk reading, cross‑asset context and an events calendar: see Exent Market intel →
Microsoft, OpenAI and Meta are CoreWeave’s best-documented customers, with Microsoft the largest by revenue so far. CoreWeave’s 10-K says Microsoft provided about 67% of its 2025 revenue. Its June 2026 10-Q names Microsoft, OpenAI and Jane Street as significant customers but reports concentration only as unnamed “Customer A, B and C”: 36%, 26% and 10% of Q2 2026 revenue.
Scope: customers of CoreWeave (CRWV) and, for comparison, of Nebius and IREN, using only SEC filings and company releases. Values are the contract figures the companies disclosed, not estimates from press reports. Excluded: customers named in marketing without any financial term, and private neoclouds. For the business model behind these contracts, start with what is a neocloud.
Seven listed commitments carry a dollar figure; most others are announced without one. Values are as stated by CoreWeave, usually “up to”, and each is subject to CoreWeave delivering the capacity.
| Customer | Value as disclosed | Term | Signed | Filing |
|---|---|---|---|---|
| OpenAI | Up to ~$11.9B | Through Oct 2030 | Master agreement, 2025 (prospectus: March; later filings: May 8) | IPO prospectus, Mar 31, 2025 |
| OpenAI | $4B expansion | Not disclosed | Q2 2025 | Q2 2025 results, Aug 12, 2025 |
| OpenAI | Up to ~$6.5B (total up to ~$22.4B) | Through May 31, 2031 | Sep 23, 2025 | 8-K, Sep 25, 2025 |
| Meta | Up to ~$14.2B, option to expand | Through Dec 14, 2031 | Sep 25, 2025 | 8-K, Sep 30, 2025 |
| Meta | ~$21B, incl. exercise of the earlier option | Through Dec 20, 2032 | Mar 31, 2026 | 8-K, Apr 9, 2026 |
| NVIDIA | $6.3B initial value; buys unsold capacity | Through Apr 13, 2032 | Sep 9, 2025 | 8-K, Sep 15, 2025 |
| Jane Street | ~$6B (plus $1B equity investment) | Not disclosed | Apr 2026 | 8-K, Apr 15, 2026 |
| Microsoft | Not disclosed | Not disclosed | Master agreement Feb 22, 2023 | 10-K, Mar 2, 2026 |
| Anthropic | Not disclosed (“multi-year”) | Not disclosed | Apr 10, 2026 | CoreWeave newsroom |
| Poolside | Not disclosed (40,000+ GPUs) | Not disclosed | Oct 15, 2025 | CoreWeave newsroom |
IBM, Mistral, Perplexity, Databricks, Cognition and others appear as named customer wins in CoreWeave’s releases without values. The NVIDIA agreement is unusual: NVIDIA, also a CoreWeave supplier and shareholder ($2B invested in January 2026), committed to buy any capacity CoreWeave cannot sell. That supports utilization, and it ties CoreWeave further to one partner. See the NVDA page for the supplier’s side.
The filings no longer say. Microsoft was about 67% of 2025 revenue; in Q2 2026 the largest customer, unnamed, was 36%. Concentration fell; whether Microsoft’s own share fell is not disclosed. The share of the largest customer, from CoreWeave’s filings:
The 2026 figures are unnamed, and CoreWeave warns that the letters “may represent different customers” from one period to the next, so do not assume Customer A is still Microsoft. What changed is the spread: in Q2 2026 Customer B was 26% and Customer C 10%, so the top three still made up 72%. Why the mix moved is not stated in the filing; deliveries to Meta and OpenAI under the contracts above are a plausible reason (our interpretation).
In Q2 2026, roughly $1.85B of CoreWeave’s $2,575M revenue came from three customers. The arithmetic, using the 10-Q percentages:
| Customer (unnamed in the 10-Q) | Share of Q2 2026 revenue | Revenue | Annualized (× 4) |
|---|---|---|---|
| Customer A | 36% | $2,575M × 0.36 = $927M | $3.7B |
| Customer B | 26% | $2,575M × 0.26 = $670M | $2.7B |
| Customer C | 10% | $2,575M × 0.10 = $258M | $1.0B |
| Top three | 72% | $1,854M | $7.4B |
| Everyone else | 28% | $721M | $2.9B |
Interpretation (ours): if one of the top two stopped paying, about a quarter to a third of quarterly revenue would be at stake, while the debt behind the capacity would stay. That is one reason CoreWeave’s delayed-draw loans are backed by specific infrastructure, and why the customer’s own ability to pay matters as much as the contract value. The same concentration check works for any company; see single-stock concentration for the portfolio version.
The disclosed contracts are a large part of the backlog but cannot be added up to it. CoreWeave reported revenue backlog of about $104B at June 30, 2026, which includes remaining performance obligations plus other amounts it expects under committed contracts, subject to delivery, and excludes more than $25B signed in early Q3. The disclosed values above total about $70B on paper (22.4 + 14.2 + 21 + 6.3 + 6), but that sum overstates: part of Meta’s $21B is the earlier $14.2B option being exercised, amounts already delivered and recognized as revenue are no longer in backlog, and Microsoft’s contract, the largest source of revenue so far, has no disclosed value at all. How backlog, RPO and ARR differ is in backlog, RPO and ARR explained.
The same two hyperscalers anchor the peers: Microsoft and Meta at Nebius, Microsoft at IREN. Disclosed in their filings:
| Company | Customer | Value as disclosed | Term | Filing |
|---|---|---|---|---|
| Nebius | Microsoft | ~$17.4B, up to ~$19.4B with extra services | Five years, through 2031 | 6-K, Sep 8, 2025 |
| Nebius | Meta | ~$2.9B | Five years | 6-K, Nov 12, 2025 |
| Nebius | Meta | Up to ~$27B: $12B dedicated + up to $15B of unsold capacity | Five-year orders from early 2027 | 6-K, Mar 16, 2026 |
| IREN | Microsoft | ~$9.7B, 20% prepaid per tranche | Five-year average term, through 2031 | 8-K, Nov 3, 2025 |
| IREN | NVIDIA | ~$3.4B | Five years | 10-K, Aug 27, 2026 |
| IREN | A frontier AI lab (unnamed) | Not disclosed | Multi-year | 10-K, Aug 27, 2026 |
Two details change how much these figures are worth. Meta’s $15B at Nebius is a backstop: Nebius intends to sell that capacity to other customers, and Meta buys only what remains unsold. Nebius’s Microsoft obligations began only once Nebius confirmed it had the financing. Nebius does not name its largest customers by revenue; its three largest were 24%, 21% and 14% of Q2 2026 revenue. Company numbers side by side are in neocloud stocks compared, and the stock pages are CRWV and NBIS.
CoreWeave is itself a large tenant of former bitcoin miners, which links their revenue to CoreWeave’s health. Disclosed leases:
| Landlord | Capacity | Term | Value as disclosed | Filing |
|---|---|---|---|---|
| Applied Digital (Polaris Forge 1) | 400 MW critical IT | ~15 years | ~$11B anticipated lease revenue | 8-K, Aug 29, 2025 |
| Galaxy (Helios) | 526 MW critical IT (800 MW gross) | 15-year base + two 5-year options | $1.2B+ average annual revenue | Q2 2026 results |
| Core Scientific | ~590 MW critical IT | 12 years | $10.2B total contracted revenue | 8-K, Feb 26, 2025 |
| Landlord | Applied Digital (Polaris Forge 1) | Galaxy (Helios) | Core Scientific |
|---|---|---|---|
| Capacity | 400 MW critical IT | 526 MW critical IT (800 MW gross) | ~590 MW critical IT |
| Term | ~15 years | 15-year base + two 5-year options | 12 years |
| Value as disclosed | ~$11B anticipated lease revenue | $1.2B+ average annual revenue | $10.2B total contracted revenue |
| Filing | 8-K, Aug 29, 2025 | Q2 2026 results | 8-K, Feb 26, 2025 |
CoreWeave also tried to buy Core Scientific in 2025; Core Scientific’s stockholders did not approve the deal and the merger agreement was terminated on October 30, 2025.
The contracts are commitments to pay for capacity once delivered, so the larger risks are late delivery and a customer’s own ability to pay, rather than a simple cancellation. The filings describe the commitments as subject to delivery and availability of service; if capacity arrives late, revenue moves later while interest on the debt that built it does not. What to check, in each new filing:
The spending that ultimately pays for these contracts is tracked in big-tech AI spending in 2026.
Market intel provides market context, cross-asset analysis, an events calendar and regular written notes. These can inform your review of an existing plan; they do not reliably predict crashes or guarantee better outcomes.
Microsoft, by revenue to date: about 67% of CoreWeave’s 2025 revenue and 62% of 2024 revenue. In Q2 2026 the 10-Q reports the largest customer only as an unnamed Customer A at 36%.
About 67% in 2025, according to the 10-K filed Mar 2, 2026. The 2026 quarterly filings do not name customers by percentage; the top three were 36%, 26% and 10% of Q2 2026 revenue.
Up to about $22.4B in total across three agreements: up to $11.9B through October 2030, a $4B expansion and up to $6.5B through May 31, 2031. The values are ceilings subject to delivery.
Up to about $14.2B through Dec 14, 2031 (signed Sep 2025) and about $21B through Dec 20, 2032 (signed Mar 31, 2026). The second figure includes the exercise of an option from the first, so the two are not fully additive.
Yes. In September 2025 NVIDIA signed an order with an initial value of $6.3B and agreed to buy capacity CoreWeave cannot sell, through Apr 13, 2032. NVIDIA is also a supplier and invested $2B in CoreWeave shares in January 2026.
Yes. CoreWeave announced a multi-year agreement with Anthropic on Apr 10, 2026, with compute coming online later that year. No contract value was disclosed.
Microsoft (about $17.4B through 2031, up to $19.4B) and Meta (about $2.9B from November 2025 and up to $27B from March 2026, of which $15B is a backstop for unsold capacity), plus many smaller AI companies. Its three largest customers were 24%, 21% and 14% of Q2 2026 revenue.
Microsoft has not stated its reasons in these filings. Commonly cited explanations are speed, since neoclouds can offer powered GPU capacity sooner, and flexibility, since a five-year contract commits less capital than building and owning the data center.
Market intel provides a Risk ON / Neutral / Risk OFF reading from four pillars, cross-asset analysis, an events calendar and regular written notes. Context for reviewing a plan, not a crash forecast.
Educational content, not investment advice. Examples are illustrative; past performance does not guarantee future results.