What does CoreWeave do?
CoreWeave, Inc. is a Nasdaq-listed technology infrastructure company trading under CRWV. It operates an AI-native cloud built specifically for accelerated computing rather than a broad, general-purpose cloud. The CoreWeave Cloud platform combines high-density GPU infrastructure, networking, storage, orchestration software, observability, security, and managed services so customers can train models, run inference, move large datasets, and operate agentic workflows at scale.
Which parts of the platform matter most?
The platform spans the full AI workload stack. At the hardware layer, CoreWeave deploys current-generation NVIDIA systems, high-speed networking, dense storage, and liquid-cooled capacity. Mission Control monitors fleet and workload health; SUNK brings Slurm-style scheduling to Kubernetes; and Weights & Biases extends the platform into model experimentation and observability. The aim is to replace a fragmented stack of chip, data-center, networking, software, and operations vendors.
| Research lens | CoreWeave position | Why it matters |
|---|---|---|
| Customers | AI labs, technology companies, and enterprises | Demand is concentrated among organizations capable of purchasing very large compute clusters. |
| Workloads | Model training, inference, reinforcement learning, data movement, and specialized HPC | The platform participates across more of the model lifecycle than raw GPU rental alone. |
| Infrastructure footprint | 43 data centers and more than 850 MW of active power at December 31, 2025 | Physical deployment scale is a binding constraint in AI cloud, not merely a background asset. |
| Contracted capacity | Approximately 3.1 GW at December 31, 2025 | The gap between active and contracted power represents both a growth runway and an execution obligation. |
How does CoreWeave make money?
Why are committed contracts the economic core?
Customers generally buy specified compute capacity under multi-year, take-or-pay commitments, with revenue recognized as capacity is provided. CoreWeave also offers pay-as-you-go access, but committed contracts generated more than 98% of FY2025 revenue. Terms commonly run one to six years; weighted-average duration was about five years at December 31, 2025. Customers often prepay 15% to 25% of contract value, creating deferred revenue and helping fund construction.
What are the main revenue and margin drivers?
| Economic engine | Pricing logic | Margin or cash-flow implication |
|---|---|---|
| Reserved AI compute | Multi-year take-or-pay capacity | Improves revenue visibility, but requires infrastructure to be delivered on schedule. |
| On-demand compute | Pay-as-you-go usage through published and negotiated pricing | Offers flexibility and customer acquisition, but represents a small portion of current revenue. |
| Managed software and services | Bundled or separately priced orchestration, storage, observability, and developer tools | Can deepen switching costs and improve customer productivity without requiring proportional GPU additions. |
| Customer prepayments | Typically 15%–25% of active contract value at FY2025 year-end | Boosts operating cash flow and deferred revenue before the related service is fully delivered. |
How concentrated is the customer base?
Customer concentration is the model’s clearest trade-off. Microsoft generated 67% of FY2025 revenue, while CoreWeave’s top two customers produced approximately 65% of Q1 2026 revenue. Large contracts create backlog and asset-level financing support, but they also give a few buyers substantial influence over deployment timing, renewal economics, and capacity planning. Some customers are also hyperscale competitors, so CoreWeave can simultaneously be a supplier, partner, and potential substitute.
What does CoreWeave’s latest quarter show?
Is growth translating into operating profit?
The Q1 2026 earnings release showed exceptional top-line expansion but continuing GAAP pressure. Revenue increased by $1.096 billion from Q1 2025. Management attributed 38% of that increase to expansion among existing customers and the remainder to new customers. Operating expenses reached $2.222 billion, producing a $144 million operating loss and a negative 7% operating margin. Interest expense of $536 million then pushed the net loss to $740 million, or negative 36% of revenue.
| Metric | Q1 2026 | Q1 2025 | Interpretation |
|---|---|---|---|
| Revenue | $2.078B | $982M | Capacity deployment and new contracts drove 112% growth. |
| Operating income (loss) | $(144)M | $(27)M | Depreciation and infrastructure costs scaled faster than revenue. |
| Interest expense, net | $(536)M | $(264)M | Debt-funded growth remains a major burden below operating income. |
| Net loss | $(740)M | $(315)M | The accounting loss widened despite much larger revenue. |
| Operating cash flow | $2.984B | $61M | Customer collections, prepayments, and working-capital timing generated strong cash inflow. |
| Property and equipment purchases | $(7.695)B | $(1.407)B | Infrastructure investment materially exceeded operating cash generation. |
Why does the expense structure matter?
Technology and infrastructure expense was $1.273 billion, or 61% of Q1 2026 revenue, largely because depreciation and amortization climbed to roughly $1.1 billion. Cost of revenue was $716 million, or 34% of revenue, with rent, power, and data-center costs rising as new sites came online. Together, those two categories consumed about 96% of quarterly revenue before sales, marketing, and administration. CoreWeave can therefore report high adjusted EBITDA while still recording a GAAP operating loss and deeply negative cash flow after capital investment.
Which turning points shaped CoreWeave’s strategy?
How did a crypto-era infrastructure business become an AI cloud?
CoreWeave’s history explains the speed and financing intensity of its model. Founded in 2017, it initially used specialized computing for crypto mining, then discontinued that activity and pivoted to cloud infrastructure. The 2020 platform launch centered the business on GPU orchestration and managed services. From 2023 onward, large contracts and asset-backed financing turned it into a major independent AI-compute operator.
-
2017CoreWeave was founded. Early experience acquiring and operating specialized GPUs created practical supply-chain and fleet-management knowledge.
-
2020The CoreWeave Cloud platform launched, shifting the company toward commercial accelerated computing and managed cloud services.
-
2023Relationships with Microsoft and NVIDIA supported larger-scale deployments and helped establish long-duration, contract-backed infrastructure financing.
-
March 2025CoreWeave priced its IPO at $40 per Class A share, issued 37 million shares, and generated about $1.4 billion of net proceeds before $31 million of offering costs.
-
May 2025The company completed the $1.0 billion Weights & Biases acquisition, extending its position from infrastructure into developer workflows and model observability.
-
September–November 2025Major OpenAI and Meta commitments expanded future demand, while acquisitions including OpenPipe, Marimo, and Monolith broadened reinforcement learning, notebooks, and industrial AI capabilities.
-
January 2026NVIDIA invested $2.0 billion at $87.20 per Class A share and deepened collaboration around future AI-factory capacity and platform alignment.
The acquisitions show a deliberate vertical-integration strategy. The Weights & Biases transaction did not materially change near-term consolidated revenue, but it added software assets, users, and developer relationships that can make CoreWeave more than a capacity reseller.
What gives CoreWeave a competitive advantage in AI infrastructure?
Why can specialization outperform a general-purpose cloud?
AI training and inference require dense clusters, predictable networking, fast storage, power, and operational expertise. CoreWeave designs around those constraints instead of adapting a broad enterprise cloud. Liquid cooling, high-speed interconnects, bare-metal Kubernetes, and workload-specific scheduling can improve utilization and time-to-compute. Early deployment of NVIDIA GB200 and GB300 systems also matters when customers compete for scarce hardware.
Where do software and relationships reinforce the moat?
Mission Control, SUNK, storage services, and Weights & Biases create an operating layer above the GPU fleet. This software can reduce downtime, expose performance bottlenecks, support compliance, and make complex clusters easier to use. The deeper the platform integrates with a customer’s training and inference workflow, the more disruptive migration becomes. Supplier relationships are equally important: CoreWeave’s close NVIDIA alignment supports hardware access and technical collaboration, while OEM, ODM, data-center, power, and network relationships determine whether contracted capacity can be delivered on time.
Who are CoreWeave’s main competitors?
How does CoreWeave compare with hyperscalers and specialist clouds?
The principal competitors are AWS, Microsoft Azure, Google Cloud, Oracle Cloud, and specialist GPU providers. Hyperscalers bring global distribution, broad enterprise relationships, integrated software, and larger balance sheets. CoreWeave counters with AI-specific architecture, rapid accelerator deployment, and managed operations. Boundaries are blurred: Microsoft is a major customer, NVIDIA is both supplier and shareholder, and large customers can build substitute capacity internally.
| Competitor group | Structural advantage | CoreWeave response | Pressure point |
|---|---|---|---|
| AWS, Azure, Google Cloud | Global scale, broad product suites, enterprise procurement access | Purpose-built clusters, specialist support, rapid accelerator deployment | Hyperscalers can bundle compute with data, software, and commercial credits. |
| Oracle Cloud | Large-scale infrastructure and enterprise relationships | AI-first software stack and focused operations | Competes for the same large training and inference contracts. |
| Specialist GPU clouds | Focused offerings and potentially lower overhead | Larger contracted footprint, software acquisitions, and financing access | Price competition can intensify as accelerator supply improves. |
| Customer self-builds | Full control and potential long-run cost savings | Faster deployment, outsourced complexity, flexible geography | Major customers may internalize workloads once facilities are ready. |
How strong are CoreWeave’s cash flow, debt, and capital capacity?
Why is conventional free cash flow deeply negative?
For Q1 2026, operating cash flow was $2.984 billion and property-and-equipment purchases were $7.695 billion, implying about negative $4.711 billion of simple free cash flow before other investing items. CoreWeave must buy equipment, fund construction, secure power, and build data-center systems before much of the related contract revenue is recognized.
How much liquidity offsets the debt load?
The Q1 2026 Form 10-Q reported $11.091 billion of total liquidity, including $2.244 billion of cash, $22 million of marketable securities, and $8.825 billion available under existing facilities. Total debt, net of discounts and issuance costs, was $24.859 billion, with $7.547 billion classified as current. Interest expense doubled year over year to $536 million in Q1 2026. The balance sheet can support enormous deployment, but it leaves limited tolerance for construction delays, customer disputes, refinancing stress, or a sharp increase in the cost of capital.
| Financial capacity metric | March 31, 2026 | December 31, 2025 | Research implication |
|---|---|---|---|
| Cash and cash equivalents | $2.244B | $3.127B | Cash declined despite a $2.0 billion NVIDIA equity investment and heavy financing activity. |
| Total liquidity | $11.091B | $6.862B | Expanded facilities provide deployment capacity but also increase financing dependence. |
| Total debt, net of discounts | $24.859B | $21.373B | Debt rose by $3.486 billion in one quarter. |
| Property and equipment, net | $36.424B | $30.557B | The asset base expanded rapidly as contracted capacity was built. |
| Total liabilities | $50.814B | $45.967B | Liabilities represented most of the $55.573 billion asset base. |
Who owns CoreWeave, and how is control structured?
Why do the Class B shares matter?
CoreWeave has a dual-class structure: Class A carries one vote per share and Class B carries ten. The 2026 proxy statement showed directors and executive officers controlling 72.32% of voting power as of April 15, 2026. Founding leadership therefore retains strong influence over the board, strategic transactions, equity issuance, and capital allocation despite outside ownership of most Class A shares.
| Holder or group | Class A beneficial ownership | Class B beneficial ownership | Total voting power | Why it matters |
|---|---|---|---|---|
| Michael Intrator | 5,289,944 shares | 56,215,770 shares | 38.70% | Co-founder, CEO, president, and board chair; the single most influential voter. |
| Brian Venturo | 422,832 shares | 30,114,514 shares | 20.30% | Co-founder and chief strategy officer with material control. |
| Brannin McBee | 377,569 shares | 21,140,580 shares | 14.59% | Co-founder influence reinforces the founder-controlled governance structure. |
| Magnetar-managed funds | 77,684,206 shares | None reported | 5.34% | Largest disclosed outside Class A holder by beneficial shares. |
| NVIDIA Corporation | 47,213,353 shares | None reported | 3.27% | Strategic supplier and shareholder alignment is commercially important. |
NVIDIA’s January 2026 investment added 22,935,780 Class A shares for $2.0 billion at $87.20 per share. The broader NVIDIA collaboration signals confidence and may improve coordination around future architectures, but CoreWeave remains exposed to one dominant accelerator ecosystem and its pricing, allocation, and product road map.
What opportunities and risks could change CoreWeave’s outlook?
Where could growth exceed the current plan?
Which risks are most financially material?
The main risks connect directly to the financial statements. A major customer can delay, renegotiate, internalize, or decline to renew capacity. Data-center and power delays can create costs before billing begins. Debt service can absorb operating gains, new chips can shorten asset lives, and regulation of AI, power, water, competition, or data security can slow deployment. Cyber incidents could interrupt mission-critical workloads.
Which KPIs matter most, and how should valuation be framed?
What should researchers monitor each quarter?
Why is a DCF unusually sensitive for CoreWeave?
A conventional software DCF can mislead because CoreWeave’s reinvestment is physical and front-loaded. A defensible model begins with contracted revenue conversion and explicit assumptions for deployment timing, utilization, revenue per unit of capacity, power and rent, depreciation, replacement capex, prepayments, interest, debt repayment, and dilution. Terminal value should reflect obsolescence and capital intensity rather than software-like margins.
What is the key takeaway from CoreWeave analysis?
CoreWeave became strategically important by converting scarce AI compute, power, and operating expertise into a purpose-built cloud. FY2025 revenue reached $5.131 billion, Q1 2026 revenue reached $2.078 billion, and backlog approached $100 billion. Its advantages are AI-specific infrastructure, early accelerator access, a growing software layer, multi-year commitments, and willingness to build ahead of demand.
The weaknesses are equally specific. A few customers account for most revenue; the company must deliver a vast contracted power pipeline; Q1 2026 capex far exceeded operating cash flow; total debt, net of discounts, reached $24.859 billion; and interest expense materially widened the loss. Founder voting control allows patient execution but limits outside shareholder influence. For students and investors, the central question is not whether AI compute demand is growing. It is whether CoreWeave can convert backlog into durable, diversified free cash flow before financing costs, obsolescence, construction risk, and customer bargaining power absorb the economics.
5-Year Financial Model
40+ Charts & Metrics
DCF & Multiple Valuation
Free Email Support
Disclaimer
All information, articles, and product details provided on this website are for general informational and educational purposes only. We do not claim any ownership over, nor do we intend to infringe upon, any trademarks, copyrights, logos, brand names, or other intellectual property mentioned or depicted on this site. Such intellectual property remains the property of its respective owners, and any references here are made solely for identification or informational purposes, without implying any affiliation, endorsement, or partnership.
We make no representations or warranties, express or implied, regarding the accuracy, completeness, or suitability of any content or products presented. Nothing on this website should be construed as legal, tax, investment, financial, medical, or other professional advice. In addition, no part of this site—including articles or product references—constitutes a solicitation, recommendation, endorsement, advertisement, or offer to buy or sell any securities, franchises, or other financial instruments, particularly in jurisdictions where such activity would be unlawful.
All content is of a general nature and may not address the specific circumstances of any individual or entity. It is not a substitute for professional advice or services. Any actions you take based on the information provided here are strictly at your own risk. You accept full responsibility for any decisions or outcomes arising from your use of this website and agree to release us from any liability in connection with your use of, or reliance upon, the content or products found herein.