Jane Street buys $1B stake as CoreWeave commits $6B compute

CoreWeave will supply $6 billion in GPU-based compute capacity to quantitative trading firm Jane Street, the companies announced Wednesday. Jane Street also purchased $1 billion of CoreWeave Class A common stock at $109 per share. The compute will come from multiple CoreWeave data center facilities to support the firm’s trading and research operations.

The agreement covers GPU resources across CoreWeave’s sites for large-scale model training and inference. The announcement did not specify how much capacity is contracted versus on-demand, nor did it provide a delivery schedule. The equity purchase gives Jane Street a minority ownership stake; CoreWeave shares rose about 1.5% on Wednesday to roughly $119.04.

The deal follows an agreement announced about a week earlier in which AI developer Anthropic plans to run its Claude models on CoreWeave infrastructure. CoreWeave provides GPU-accelerated cloud services for AI workloads, a segment often called ‘neocloud.’ GPUs handle parallel computations used in training and running modern AI models, while traditional cloud providers more commonly use CPUs for general-purpose tasks.

Analysts at Bernstein compared CoreWeave with neocloud peers IREN and Nebius and found CoreWeave to have a stronger commercial position, citing a revenue mix of contractual agreements and on-demand sales and a diverse customer base. CoreWeave reported that nine of the top 10 AI model providers use its platform.

CoreWeave was founded as Atlantic Crypto in 2017 and shifted focus to AI infrastructure in 2019. The company supplies large pools of GPU capacity to AI developers, research groups and commercial users at a time when GPU availability has tightened for many organizations.

Neither firm disclosed additional financial terms, the duration of the compute commitment or a deployment timeline. CoreWeave is publicly traded and its shares reacted modestly to the announcement.

Demand for GPU-based cloud services has risen as AI model sizes and training costs have increased. Some former crypto miners and data center operators have repurposed hardware and facilities to serve AI workloads.