CoreWeave

CoreWeave is a United States cloud provider that operates large clusters of NVIDIA GPUs sold as on-demand capacity for artificial intelligence workloads. It belongs to the category of neoclouds — specialised providers that grew out of the shortage of AI compute in hyperscaler data centres during 2023 and 2024.^[1]

History

The company was founded in 2017 as the Atlantic Crypto Company, a cryptocurrency mining operation, and converted its data centres to GPU cloud computing in 2019 when mining economics collapsed, rebranding as CoreWeave.^[2] Its founders built the business around a single premise: workloads like large language model training do not need general-purpose server features, so stripped-down, densely GPU-packed racks are cheaper per FLOP than AWS or Azure.

CoreWeave listed on Nasdaq in March 2025 in one of the largest technology offerings of that year, and NVIDIA — whose chips form essentially all of its inventory — is a strategic investor in the company.^[3] Demand has since scaled sharply: the company reported second-quarter 2026 revenue of US$2.575 billion, up 112 percent year over year, with a backlog of roughly US$104 billion and full-year capital expenditure guidance of US$35–39 billion.^[4]

Technology and Business Model

CoreWeave's platform is built on dense GPU clusters — thousands of accelerators connected over high-bandwidth fabrics such as InfiniBand and InfiniBand's Ethernet successor Spectrum-X, with liquid cooling as the default deployment mode.^[5] Instead of making customers assemble orchestration themselves, it offers workload-specific managed services: a Kubernetes control plane tuned for distributed training, an inference platform with automatic scaling, and reserved capacity contracts that guarantee GPUs will be available when large jobs start.

Its economics rest on long-term capacity commitments from major AI laboratories. Microsoft has signed multi-billion-dollar rental agreements, and OpenAI contracted for multiple gigawatts of capacity in a 2025 deal reported to be worth over US$200 billion across its term.^[6] The same structure creates the main risks: heavy debt-financed capital spending, intense customer concentration, and competition from the hyperscalers' own capacity expansion.

Why It Matters for AI

Compute, rather than algorithms, is the binding constraint on frontier AI development, and neoclouds like CoreWeave function as a pressure valve on that constraint. They let AI labs scale training runs without waiting on cloud platform queue times, and they force hyperscalers to compete on GPU availability and pricing instead of lock-in alone.^[1]

>See Also

• Neocloud
• GPU Cluster
• AI Data Centres in Malaysia
• Liquid Cooling
• Sovereign AI
• National AI Compute Exchange