CoreWeave
US GPU cloud running AI data centres for hyperscalers and model labs
CoreWeave is an American cloud-computing company based in Livingston, New Jersey, founded in 2017 as Atlantic Crypto and renamed in 2019. It runs 51 AI data centres across the United States and Europe, renting out GPU capacity to customers including Microsoft, OpenAI, Meta and Anthropic.
- 51 AI data centres across the US and Europe as of 2026
- 250,000 GPUs reported in 2025, up from 13 US sites in 2024
- Went public on Nasdaq on 28 March 2025, raising $1.5 billion
- Microsoft accounted for about 67% of 2025 revenue
- Total debt of roughly $35.1 billion as of 30 June 2026
What it is
CoreWeave is an American cloud-computing business headquartered in Livingston, New Jersey. It rents out graphics processing units (GPUs) and the data centres around them to companies running large AI workloads. As of 2026 it operates 51 specialised AI data centres in the United States and Europe, built for parallel computing rather than general-purpose hosting. Some sites serve several customers at once; others are set aside for one client or a single project. The firm sells access to high-performance computing centres and supercomputers, along with cloud services and AI infrastructure. Its platform is built on Kubernetes, the open system for running software across many machines, and includes Mission Control software for checking hardware performance.
How it came about
Four founders started the company in New Jersey in September 2017 under the name Atlantic Crypto: commodities traders Michael Intrator, Brian Venturo and Brannin McBee, plus Peter Salanki. It mined ethereum with GPUs until the 2018 crypto crash, then renamed itself CoreWeave in 2019 and sold cloud infrastructure instead. Demand surged in 2022 and 2023 as AI took off. Nvidia put in $100 million in April 2023, and a $2.3 billion debt facility followed that August, secured against H100 chips. Funding continued through 2024, and the company listed on Nasdaq on 28 March 2025, raising $1.5 billion. Acquisitions since include Weights & Biases, OpenPipe and Monolith AI.
The compute
CoreWeave built its fleet around Nvidia hardware. It spent roughly $100 million on H100 chips in mid-2022, and was among the first cloud providers to take delivery of Blackwell systems in late 2024. In February 2025 it became the first cloud to offer Nvidia GB200 NVL72 racks, and IBM used those clusters to train its Granite models. By July 2025 it was the first to run Blackwell Ultra GB300 NVL72 commercially, with Dell supplying the servers. The Plano supercomputer for Nvidia holds over 3,500 H100s. The software layer is Kubernetes-native, aimed at GPU-heavy jobs, with Mission Control for hardware checks.
Where the GPUs are
The company owns data centres in the United States and Europe. It ran three US sites in 2022, thirteen in the US plus two in the UK by 2024, and 32 with 250,000 GPUs in 2025, reaching 51 by 2026. A London headquarters and two UK data centres opened in 2024, alongside a planned $978.6 million UK investment. It leased a 280,000 square foot building in Kenilworth, New Jersey for a $1.2 billion site. The $1.6 billion Plano, Texas supercomputer for Nvidia covers about 450,000 square feet. CoreWeave also rents capacity from Core Scientific.
Customers and deals
Microsoft is the largest customer, accounting for about 67% of revenue in 2025 and over 60% in 2024. OpenAI signed a five-year contract worth roughly $12 billion in March 2025. Meta Platforms expanded its infrastructure partnership by $21 billion in April 2026, and Anthropic agreed a multi-year deal the same month for Nvidia GPU capacity to run Claude inference, meaning live model answering. IBM trains Granite on CoreWeave clusters, Stability AI was an early client, and Cohere announced a multi-billion-dollar site built and run by CoreWeave in December 2024.
Money and ownership
Revenue reached $5.13 billion in 2025, with a net loss of $1.2 billion and assets of $49.3 billion. First-quarter 2026 revenue was $2.08 billion, and backlog approached $100 billion. Investors include Nvidia, which put in $2 billion in January 2026 at $87.20 a share, plus Coatue Management, Magnetar Capital, Blackstone, Cisco, Jane Street, Fidelity Management and Macquarie Capital. A $2.3 billion debt facility in 2023 used H100s as collateral, a first. Goldman Sachs, JPMorgan Chase and Morgan Stanley led a $650 million credit line in 2024 and managed the 2025 IPO.
Its place in AI
CoreWeave grew from a crypto miner into one of the largest specialist GPU clouds, and its Nasdaq debut in March 2025 was the biggest AI-related listing by amount raised, according to Dealogic. Nvidia described its Plano, Texas site as the fastest AI supercomputer in the world in September 2023. It was first to offer several Nvidia systems through the cloud, including GB200 NVL72 and Blackwell Ultra GB300 NVL72. IBM's Granite models trained on its clusters. In 2025 it joined the New Jersey AI Hub, a project of the state economic development authority, Plug and Play and Princeton University.
Questions and criticism
Borrowing is heavy. As of 30 June 2026 total debt was about $35.1 billion, split roughly $31.4 billion recourse and $3.7 billion non-recourse, structured against hyperscaler contract commitments. Revenue is concentrated in one buyer: Microsoft made up around 67% in 2025. The company also reported a $1.2 billion net loss for 2025. Its business depends on continued access to Nvidia chips and on large power-hungry data centres.
Questions people ask about CoreWeave
What does CoreWeave actually sell?
It rents out GPU-powered computing: access to high-performance data centres and supercomputers, cloud services and AI infrastructure. Customers include model labs and hyperscalers running large training and inference jobs. Some sites serve many clients, while others are dedicated to one customer or a single project.
How did a crypto miner become an AI cloud?
It started as Atlantic Crypto in 2017, mining ethereum on graphics cards. After the 2018 crypto crash it renamed itself CoreWeave in 2019 and sold its GPU stock as cloud computing. When AI demand rose in 2022 and 2023, that hardware base let it grow quickly.
Who are its biggest customers?
Microsoft dominates, at about 67% of revenue in 2025. OpenAI signed a five-year deal worth roughly $12 billion in 2025, Meta expanded its partnership by $21 billion in 2026, and Anthropic agreed a multi-year contract for Claude inference capacity. IBM, Cohere and Stability AI are also named.
What hardware does it run?
Mostly Nvidia. It bought H100 chips early, was among the first to take Blackwell systems in 2024, first offered GB200 NVL72 racks in early 2025, and first deployed Blackwell Ultra GB300 NVL72 commercially in July 2025, with Dell building the servers.
How is it funded and is it listed?
It listed on Nasdaq as CRWV on 28 March 2025, raising $1.5 billion. Nvidia invested $2 billion more in January 2026 at $87.20 a share. Other backers include Coatue, Magnetar Capital, Blackstone, Cisco, Fidelity and Macquarie. Debt stood near $35.1 billion by mid-2026.
Good reading and sources
- Official websiteOfficial site
- How a Scrappy Cryptominer Transformed Into the Multibillion-Dollar Backbone of the AI BoomWired, Article
- Nvidia-backed GPU cloud provider CoreWeave surges to $19 billion valuationCNBC, Article
The facts on this page come from the Wikipedia article "CoreWeave", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryNeocloud rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.