Bottom Line
CoreWeave is not best understood as a normal software company. It is a leveraged AI infrastructure conversion machine: secure power, secure NVIDIA systems, sign take-or-pay style contracts, finance the build with customer-cash-flow-backed debt, and operate the cloud software layer on top.
The company has real scale: $36.4B of net property and equipment, $10.2B of operating lease right-of-use assets, more than 1 GW of active power, and nearly $100B of revenue backlog as of Q1 2026. SEC 10-Q Q1 release
The hard part is that the business is extremely capital intensive and levered. Q1 2026 revenue was $2.078B and adjusted EBITDA was $1.157B, but net loss was $740M because depreciation/amortization and interest are enormous. Interest expense alone was $536M in Q1. Source
What CoreWeave Actually Owns vs Controls
| Asset / right | What is known | Valuation relevance |
|---|---|---|
| GPU servers, networking, related infrastructure | CoreWeave reports $36.4B of net property and equipment as of March 31, 2026. Its filings say all GPUs currently used in its infrastructure are NVIDIA GPUs because customer contracts specify NVIDIA GPUs. Source | This is the productive asset base. It depreciates fast and needs constant refresh, but it is also what backs much of the company's asset-level financing. |
| Data center capacity | CoreWeave operated 43 data centers with over 850 MW of active power at year-end 2025, and surpassed 1 GW active power by Q1 2026. Its own definition of "data centers" includes leased or owned portions receiving active power. Source Source | Most of the value is access to power and dense fit-out capacity, not just the building shell. |
| Leased data centers | Data center leases generally run 5 to 15 years, some with extension options up to 10 years. Operating lease liabilities were $10.05B at Q1 2026. Source | These leases are long-duration obligations. They are economically debt-like and should be included in a full enterprise-value view. |
| Owned / self-developed data centers | CoreWeave says it is beginning to develop its own data centers, including a Kenilworth, New Jersey joint venture. Real Estate NJ reported a nearly $1.8B, 250 MW Kenilworth project and that CoreWeave purchased the facility and adjacent 27 acres for $322M after initially leasing it. Source Source | Ownership is still a minority piece of the footprint, but it matters because owning strategic power-ready sites can improve long-run economics and reduce landlord dependency. |
| Software platform | CoreWeave's platform includes Kubernetes, storage, Mission Control, observability, runtime acceleration, and acquired assets such as Weights & Biases. Source | This is the margin defense. The risk is that the economics still look dominated by hardware depreciation, power, leases, and financing cost. |
The clean answer: CoreWeave owns a very large amount of compute equipment and some owned/self-developed site exposure, but most publicly described data center scale is controlled through long-term leases and provider agreements rather than outright ownership of every facility.
GPU Count and Compute Capacity
Hard disclosures
| Date | Data centers | Active power | GPU count | Source |
|---|---|---|---|---|
| Dec. 31, 2022 | 3 | ~10 MW | >17,000 | IPO prospectus |
| Dec. 31, 2023 | 10 | >70 MW | >53,000 | IPO prospectus |
| Dec. 31, 2024 | 32 | >360 MW | >250,000 | IPO prospectus |
| Dec. 31, 2025 | 43 | >850 MW | Not disclosed | 2025 10-K |
| Q1 2026 | Not updated in release | >1 GW | Not disclosed | Q1 release |
Estimate that makes sense
The only honest estimate starts from power, not from a claimed current GPU number. The Dec. 2024 ratio was:
If that same physical-density ratio held at more than 1 GW, it would imply roughly 690,000+ active GPUs. But that is not a precise current count. Newer GB200/GB300/Rubin-class systems can consume materially more power per GPU, and active power includes more than just the GPU silicon. So the better framing is:
- Minimum hard fact: more than 250,000 GPUs as of Dec. 31, 2024.
- Directionally current estimate: high hundreds of thousands of active GPUs is plausible from the power ramp.
- Modeling range I would use: roughly 500,000 to 700,000 active GPUs, with a sensitivity table rather than one fake-precise point estimate.
Customer Contracts and Revenue Economics
CoreWeave's revenue is mostly from multi-year committed contracts where customers purchase specified capacity on a take-or-pay basis. The company also has pay-as-you-go usage, but that is not the center of the model. Source
RPO includes billed and unbilled consideration from committed contracts, net of estimated variable consideration. The company explicitly adjusts for potential customer credits, delivery delays, and committed capacity that CoreWeave has the right to resell. Source
| Customer / partner | Public economics | What is not disclosed | Read-through |
|---|---|---|---|
| Microsoft | Largest customer in 2025, representing about 67% of FY2025 revenue. IPO prospectus disclosed $81M of 2023 revenue and $1.2B of 2024 revenue under the Microsoft MSA. Source Source | Total remaining contract value, specific GPU SKUs, exact price per GPU-hour, margins. | Using FY2025 revenue of $5.1B, 67% implies roughly $3.4B of Microsoft revenue in 2025. This is customer concentration, not recurring software ARR. |
| OpenAI | Filings show two important references: a March 2025 MSA with up to $11.9B through Oct. 2030, and a later May/Sept. 2025 agreement/order form with up to $6.5B through May 31, 2031. The March agreement included a $350M stock issuance offset against recognized revenue. Source Source | Whether these amounts are additive, partly superseding, or tied to separate order forms is not cleanly reconcilable from public text alone. | I would not add $11.9B and $6.5B together without inspecting the full redacted agreements and later order history. |
| Meta | 2025 10-K disclosed a September 2025 order form initially up to $14.2B through Dec. 2031. Q1 2026 10-Q disclosed a March 2026 order form initially up to about $21B, including new capacity through Dec. 20, 2032 and option capacity through Apr. 10, 2032. Source Source | Exact GPU type mix, delivery schedule by facility, pricing, gross margin. | This is the clearest large post-IPO expansion and a major reason backlog jumped toward $100B. |
| Anthropic | Q1 2026 press release says CoreWeave signed a multi-year agreement with Anthropic to support Claude model development and deployment. Source | Dollar value, duration, pricing, dedicated capacity. | Important strategic logo, but not modelable from public economics. |
| Jane Street | April 2026 announcement: Jane Street committed about $6B to use CoreWeave's AI cloud and invested $1B in CoreWeave Class A shares at $109/share. Source | Revenue timing, capacity delivery schedule, exact facility/GPU allocation. | This happened after Q1, so do not assume it is fully included in March 31 RPO/backlog. |
| NVIDIA | NVIDIA invested $2B in January 2026 at $87.20/share. CoreWeave says all GPUs currently used in infrastructure are NVIDIA GPUs and customers have contractually specified NVIDIA GPUs. Source Source | Long-term supply allocation, actual chip pricing, any side-letter economics. | NVIDIA is supplier, shareholder, and ecosystem sponsor. This lowers supply risk relative to smaller GPU clouds but increases single-vendor dependence. |
Data Center Footprint and Lease Agreements
CoreWeave's public numbers distinguish active power from contracted power. Active power is already serving or ready to serve capacity. Contracted power is future deployment runway, not currently revenue-producing capacity.
| Provider / site | Capacity | Contract economics | Status / interpretation |
|---|---|---|---|
| Core Scientific | About 590 MW of critical IT load across six Core Scientific sites after a 70 MW Denton expansion. Source | Core Scientific projected $10.2B of revenue over 12-year contract terms. That implies about $1.44M per MW-year of critical IT load, before considering exact ramp timing. Source | CoreWeave tried to acquire Core Scientific, but Core Scientific terminated the merger agreement after shareholders did not approve it. So this remains a third-party provider relationship, not CoreWeave-owned sites. Source |
| Applied Digital, Polaris Forge 1, North Dakota | 400 MW total critical IT capacity across three long-term leases. Source | Applied Digital says total anticipated contracted lease revenue is about $11B, including $7B from the first two approx. 15-year leases. Simple math implies about $1.83M per MW-year over 400 MW and 15 years. Source | Initial two leases were executed in May 2025; third 150 MW building is planned for 2027 service. This is leased capacity, not owned by CoreWeave. |
| Galaxy Helios, West Texas | Initial 133 MW critical IT load, plus 260 MW Phase II, plus final option for 133 MW. Galaxy says CoreWeave committed to the full 800 MW approved gross power capacity at Helios. Source Source | Galaxy expects average annual revenue of more than $1B over the 15-year term of the CoreWeave agreements. Source | This is one of the most important power-control assets in the story. Treat the 800 MW as gross power; disclosed critical IT load is lower. |
| Kenilworth, New Jersey | Reported 250 MW data center project. Initial 2024 announcement was a 280,000 sq. ft. long-term lease and $1.2B CoreWeave investment; later reporting says CoreWeave purchased the facility plus adjacent 27 acres for $322M and started a nearly $1.8B project. Source Source | Public reporting mentions a $250M, five-year New Jersey tax credit award. Source | This is the clearest evidence that CoreWeave is moving beyond pure leased shells into at least some owned/self-developed site exposure. |
| Other providers / undisclosed sites | Total contracted power was over 3.5 GW by Q1 2026, which is larger than the fully source-itemized named projects above. Source | Not fully public. | There is meaningful contracted power not cleanly allocated in public releases. Do not assume every GW has identical cost or delivery certainty. |
Lease accounting
| Lease metric | Q1 2026 | Why it matters |
|---|---|---|
| Operating lease ROU assets | $10.182B | Represents controlled leased space/capacity on the balance sheet. Source |
| Total operating lease liabilities | $10.050B | Economically debt-like when valuing enterprise obligations. Source |
| Weighted average remaining operating lease term | 11 years | Long-term infrastructure obligation, not short-term cloud rental. Source |
| Q1 2026 total lease cost | $539M | Includes $388M operating lease cost, $132M variable lease cost, and $19M finance lease cost. Source |
Debt and Capital Structure
As of March 31, 2026, CoreWeave had $25.149B of principal debt. Its debt stack is not generic corporate leverage; much of it is asset-level delayed-draw financing tied to customer contracts and the related infrastructure. Source
| Debt category | Principal at Q1 2026 | Effective rate / maturity |
|---|---|---|
| DDTL 1.0, 2.0, 2.1, 3.0, 4.0 | $11.823B | Effective rates from 7% to 15%; maturities from 2028 to 2032. Source |
| Senior notes and convertible senior notes | $6.338B | 2030 senior notes, 2031 senior notes, and 2031 convertible senior notes. Source |
| OEM and software license financing | $5.036B | Effective rate about 10%, maturities July 2026 to July 2030. Source |
| Revolver | $1.500B | Effective rate about 6%, maturity Nov. 2029. Source |
| Convertible promissory notes and Magnetar loan | $452M | Effective rates about 7% and 12%. Source |
| Total principal debt | $25.149B | Current debt was $7.547B; non-current debt was $17.312B net of discounts/costs. Source |
Q1 2026 cash and cash equivalents were $2.244B, restricted cash was $1.076B, and marketable securities were $22M. Net debt excluding operating leases was therefore roughly $21.8B. Including operating lease liabilities, net debt-like obligations were roughly $31.9B. Source
Post-Q1 financing updates
- April 2026: CoreWeave issued $1.75B of 9.750% senior notes due 2031, then added $1.0B of additional 9.750% senior notes. Source
- May 2026: CoreWeave closed a $3.1B DDTL 5.0 delayed draw term loan facility to support committed deployments. Source
- CoreWeave's May 2026 press release says it secured more than $20B of debt and equity capital year-to-date. Source
Valuation Inputs That Matter
| Input | Current data point | How to use it |
|---|---|---|
| Revenue | FY2025 revenue $5.1B; Q1 2026 revenue $2.078B, up 112% year over year. Source Source | Q1 annualized revenue is about $8.3B, but capacity ramps are lumpy, so this is not a steady-state run-rate guarantee. |
| Adjusted EBITDA | FY2025 adjusted EBITDA $3.093B; Q1 2026 adjusted EBITDA $1.157B. Source Source | Useful for operating scale, but not enough by itself because D&A and interest are central costs, not accounting noise. |
| Net loss | FY2025 net loss $1.167B; Q1 2026 net loss $740M. Source Source | Equity value depends on whether future contracted revenue can outrun depreciation, interest, and replacement capex. |
| Cash flow | Q1 2026 operating cash flow $2.984B, investing cash flow negative $7.708B, financing cash flow $3.914B. Source | Operating cash flow is helped by upfront customer payments. The model still consumes enormous capital during buildout. |
| Backlog | Q1 2026 revenue backlog $99.4B; RPO $98.8B. Source Source | Backlog is the main financing support. Its quality depends on delivery, customer credit, availability requirements, and whether margins hold. |
| Market value snapshot | CRWV closed at $103.77 on May 18, 2026. Q1 filings show about 532M Class A/B shares outstanding; Jane Street's $1B equity investment at $109/share adds about 9.2M shares, implying about 541M shares before other post-Q1 issuance. Source Source Source | That gives a rough market cap near $56B and rough enterprise value near $78B excluding operating leases or $88B including operating leases, before exact post-Q1 debt repayment/draw effects. |
Simple valuation math
At those rough values, the stock is not cheap on near-term revenue or EBITDA, but the backlog is unusually large relative to enterprise value. The real question is not whether demand exists. Given AGI-driven compute demand, demand is the easy part. The hard questions are delivery speed, power access, chip allocation, customer concentration, financing cost, and whether future GPU generations make today's financed fleet obsolete faster than contracted cash flows amortize it.
Big Risks and Unknowns
- Exact current GPU count: not disclosed. Any exact number after Dec. 2024 is an estimate.
- Customer concentration: top two customers were about 65% of Q1 2026 revenue; Microsoft was 67% of FY2025 revenue. Source Source
- Delivery risk: RPO/backlog depends on delivering capacity and meeting availability requirements.
- Debt/refinancing risk: Q1 interest expense was $536M and gross principal debt was already $25.1B before later financing activity. Source
- Technology refresh risk: CoreWeave needs to keep buying new NVIDIA generations. Older GPUs may remain useful for inference, but resale and redeployment values are uncertain.
- Power and facility risk: contracted power is not the same as active revenue-producing power.
- Counterparty risk: customer prepayments and asset-level lending reduce but do not eliminate non-payment or non-performance risk.
What I Would Use as the Baseline Model
| Line item | Baseline value | Comment |
|---|---|---|
| Active power | >1 GW | Use Q1 2026 as current hard disclosure. |
| Contracted power | >3.5 GW | Future deployment runway, not all active. |
| Current active GPU estimate | 500k to 700k | Use as a sensitivity, not a fact. |
| Backlog / RPO | $99.4B backlog / $98.8B RPO | Use RPO for SEC-defined contracted performance obligations; use backlog to understand management's broader committed revenue view. |
| Net debt excluding operating leases | ~$21.8B at Q1 | Update after next filing for post-Q1 notes, DDTL draws, repayments, and equity issuance. |
| Net debt-like obligations incl. operating leases | ~$31.9B at Q1 | Better for economic leverage. |
| Q1 adjusted EBITDA | $1.157B | High margin before D&A and interest. |
| Q1 net loss | $740M | Do not ignore depreciation and financing cost. |
Stock Chart
This chart uses the site's shared chart component when served through the existing web app. If opened directly from disk, the chart may not load because it depends on the site's chart API.
Primary Sources
- CoreWeave Form 10-Q for quarter ended March 31, 2026 - balance sheet, RPO, debt table, lease details, customer concentration, OpenAI/Meta disclosure, NVIDIA GPU dependence.
- CoreWeave Q1 2026 earnings press release - revenue, adjusted EBITDA, revenue backlog, active power, contracted power, Meta/Anthropic highlights.
- CoreWeave Form 10-K for year ended December 31, 2025 - 43 data centers, 850+ MW active power, 3.1 GW contracted power, RPO, Microsoft concentration, software/platform description.
- CoreWeave IPO prospectus, 424B4, March 2025 - historical GPU counts, Dec. 2024 250,000+ GPUs, Dec. 2024 360+ MW active power, Dec. 2024 1.3 GW contracted power, Microsoft/OpenAI early disclosures.
- CoreWeave Q4/FY2025 earnings press release - FY2025 revenue, adjusted EBITDA, backlog, active/contracted power.
- Core Scientific and CoreWeave Denton expansion release, Feb. 26, 2025 - 590 MW critical IT load, $10.2B projected revenue over 12-year terms.
- Core Scientific merger termination release, Oct. 30, 2025 - terminated CoreWeave merger agreement after stockholder approval failure.
- Applied Digital release, Aug. 29, 2025 - Polaris Forge 1 additional 150 MW lease, 400 MW total critical IT capacity, about $11B anticipated lease revenue.
- Galaxy release, Apr. 23, 2025 - Helios Phase II option, 393 MW critical IT load, 800 MW gross approved capacity.
- Galaxy PR Newswire release, Aug. 15, 2025 - $1.4B Helios financing, full 800 MW CoreWeave commitment, more than $1B average annual revenue over 15 years.
- Onyx/Machine/CoreWeave Kenilworth release, Oct. 28, 2024 - 280,000 sq. ft. lease, $1.2B planned CoreWeave investment.
- Real Estate NJ Kenilworth update, Nov. 25, 2025 - nearly $1.8B project, 250 MW, reported $322M site purchase, $250M tax credit.
- CoreWeave/Jane Street release, Apr. 15, 2026 - $6B AI cloud agreement and $1B equity investment at $109/share.
- NVIDIA/CoreWeave release, Jan. 26, 2026 - NVIDIA $2B equity investment at $87.20/share and expanded collaboration.
- CoreWeave April 2026 senior notes 8-K - $1.0B additional 9.750% senior notes due 2031 and prior $1.75B issuance.
- CoreWeave May 2026 DDTL 5.0 8-K - $3.1B delayed draw term loan facility.
- CoreWeave DDTL 5.0 press release, May 18, 2026 - facility closing and more than $20B debt/equity capital secured year-to-date.
- Yahoo Finance CRWV historical data - used only for the May 18, 2026 closing price cross-check.