Cloud ComputingUS
← Sticker Wall
JOURNEY · DETAIL
CoreWeave: The AI Compute Rising Star Transforming from Crypto Mining to GPU Cloud, Tied to NVIDIA
Founded: Michael Intrator, Brian Venturo, Brannin McBee · CoreWeave
JOURNEY
Key Fields
FIELD STAMPSIndustryCloud Computing
RegionUS
ScaleGiant
ChannelB2B
Origin
The three founders were originally commodities traders and energy investors who, in 2017, stumbled upon arbitrage opportunities in cryptocurrency mining and began mining Ethereum with GPUs in a garage. While many on Wall Street dismissed it as a fringe business at the time, they keenly realized that their capabilities in large-scale GPU procurement, data center management, and power optimization could be repurposed. When the price of Ethereum plummeted, they did not liquidate; instead, they began exploring how to convert their vast GPU inventory into rentable compute resources, eventually pivoting to AI model training workloads around 2020 and entering the cloud services market.
Milestones
2017
Inception Turning Point
Michael Intrator, Brian Venturo, and Brannin McBee began mining Ethereum using GPUs, with early mining rigs housed in a co-founder's apartment closet. According to later interviews with CoreWeave co-founders, they initially invested their own capital to purchase graphics cards, earning their first pot of gold during the 2017 cryptocurrency bull market, though the initial scale was limited to a personal-level mining operation.
2018
Expansion Failure
The price of Ethereum plunged by over 80% from its peak in early 2018, causing mining revenue to shrink drastically and leading many miners to shut down. CoreWeave's mining operations approached the break-even point based on electricity costs, forcing the team to reduce the number of active rigs and begin renting out GPU compute at prices lower than mining yields to maintain cash flow. This passive pivot became the technical foundation for their later cloud business.
2020
Pivot Transition
The team discovered that the demand for AI model training on GPUs was far higher than that for cryptocurrency mining, and customers were willing to pay a higher premium. CoreWeave officially shifted its business focus from mining to GPU cloud services, providing bare-metal servers and on-demand compute to machine learning companies and research institutions. That year, the company signed contracts with several AI startups, transitioning its revenue from residual crypto income to cloud service contract revenue.
2023
Breakout Growth
NVIDIA invested approximately $2 billion in CoreWeave, making it a strategic partner in the AI cloud space. That year, CoreWeave secured large compute orders from Microsoft and OpenAI, with OpenAI indirectly procuring CoreWeave's GPU resources via the Microsoft cloud for training models like ChatGPT. In 2023, CoreWeave's revenue reached approximately $700 million, growing several-fold from the previous year, while headcount expanded from under 100 to several hundred.
2025
Capitalization PMF
In 2025, CoreWeave went public on the NASDAQ with a valuation of approximately $35 billion under the ticker CRWV. According to IPO filings, 2024 revenue reached $2.58 billion, a year-over-year increase of over 200% from 2023. At the time of listing, the company had a backlog of over $30 billion in orders, with Microsoft accounting for over 60%. However, the company also disclosed over $12 billion in debt, and most of its revenue came from fewer than 10 customers, indicating extremely high customer concentration.
2026
Risk Exposure Failure
In August 2026, CoreWeave released financial results showing continued revenue growth but widening losses, causing the stock price to drop during trading. Financial reports indicated that full-year 2025 revenue was approximately $6.5 billion, but gross margins were pressured by electricity costs and depreciation/amortization, resulting in a net loss exceeding $2 billion. The company also issued a risk warning, stating that abandoning the exclusive use of NVIDIA chips would entail high technical conversion costs and performance degradation risks, sparking serious market concerns about its single-supplier dependency and the fragility of its business model.
Turning Points
- The 2018 Ethereum crash forced the team to rent out idle GPUs, planting the seeds for the cloud business.
- The 2020 explosion in AI training demand allowed CoreWeave to find a more sustainable path for compute monetization than mining.
- The 2023 $2 billion investment from NVIDIA and orders from Microsoft and OpenAI locked in its position as a core AI cloud provider.
- The 2025 NASDAQ listing provided public market pricing power, but simultaneously exposed massive debt and customer concentration risks.
- In 2026, the coexistence of high revenue growth and widening losses led to a peak in market skepticism regarding the exclusive NVIDIA-tied model.
Failures & Pitfalls
- The 2018 Ethereum crash caused the core mining business to nearly collapse, forcing the company to rent out compute at low prices to survive.
- Early cloud customers were primarily small AI labs; with limited contributions from individual clients, revenue scale remained stagnant at the million-dollar level for a long time.
- Post-IPO in 2025, the debt scale exceeded $12 billion, with interest expenses far higher than those of peer cloud providers.
- Over-reliance on Microsoft orders; in 2024, Microsoft contracts accounted for over 60% of revenue, leaving bargaining power chronically constrained.
- 2026 financial reports showed widening net losses, as high growth failed to translate into positive free cash flow, putting pressure on the stock price.
关键成功要素
- The ability to repurpose GPU assets across cycles is a double-edged sword: it saves lives during mining crashes but makes it difficult to escape the capital-intensive nature of the business.
- NVIDIA's strategic investment provides priority access to chips but creates a fatal dependency on a single supplier.
- Using customer contracts as collateral for debt financing, trading orders for capital expenditure to sustain expansion.
- Focusing on a few hyperscale customers rather than the long-tail market, sacrificing bargaining power for revenue certainty.
Lessons
- Cash flow insolvency is more fatal than losses; the decision to rent out compute during the 2018 crash saved the core assets and the team.
- Deep integration with ecosystem giants accelerates growth but sacrifices independence; a balance must be found between speed and security.
- In a capital-intensive model, widening losses can occur even when revenue doubles, as depreciation, amortization, and electricity costs consume incremental profits.
- Business models with high customer concentration are repeatedly punished by capital markets, no matter how attractive the revenue growth story may be.
Core Data
- 2024 Revenue：$2.58 billion (based on public data, not independently verified)
- 2023 Revenue：Approx. $700 million (based on public data, not independently verified)
- 2025 Revenue：Approx. $6.5 billion (based on public data, not independently verified)
- 2024 Net Loss：Over $2 billion (based on public data, not independently verified)
- IPO Valuation：Approx. $35 billion (based on public data, not independently verified)
- NVIDIA Investment：Approx. $2 billion (based on public data, not independently verified)
- Order Backlog：Over $30 billion (based on public data, not independently verified)
- Total Debt：Over $12 billion (based on public data, not independently verified)
- Microsoft Order Share：Over 60% (based on public data, not independently verified)
- IPO Proceeds：Approx. $1.5 billion (based on public data, not independently verified)
Competitors / Peers
CoreWeave's primary competitors in the GPU cloud space include the three traditional cloud giants—AWS, Microsoft Azure, and Google Cloud—as well as emerging AI infrastructure companies like Lambda Labs, Together AI, and Crusoe Energy. Traditional cloud providers possess more complete service ecosystems and stronger balance sheets, but CoreWeave has gained an early advantage in AI training workloads through priority access to NVIDIA chips and more aggressive debt-fueled expansion. Compared to Lambda Labs, CoreWeave's customers are more concentrated among top-tier AI labs and tech giants; compared to Crusoe Energy, CoreWeave lacks the electricity cost advantages derived from energy assets. As Microsoft and Amazon build their own AI data centers in 2026, CoreWeave faces the risk of losing its largest customers and pressure from the cloud services operated by its upstream supplier, NVIDIA.