The Race for AI Compute Reaches Wall Street as CoreWeave Racks Up $28 Billion in April Deals
For years, the battle for artificial intelligence infrastructure was assumed to be a contest between hyperscale cloud giants. This week, one of the world's most secretive and sophisticated quantitative trading firms made clear that the competition has expanded to an entirely different arena.
Jane Street, the New York-based global trading firm whose algorithmic strategies generate billions of dollars in annual revenue, announced on April 15 that it has committed approximately six billion dollars to CoreWeave's AI cloud platform. The firm also took a one-billion-dollar equity stake, purchasing CoreWeave's Class A shares at $109 each — a total commitment of seven billion dollars to a company that did not exist a decade ago. CoreWeave's shares rose toward $117 in early trading following the announcement.
The deal is remarkable not simply for its size but for what it represents. Jane Street is not building a consumer chatbot or a foundation model. It trains large and complex models on massive volumes of noisy financial data and deploys them at scale to make markets more efficient. The firm said it needs CoreWeave's infrastructure because it allows researchers to move at the pace its competitive business demands — with access to next-generation compute across multiple facilities, including hardware based on Nvidia's recently unveiled Vera Rubin architecture.
"We are delighted to expand our partnership with CoreWeave," the firm said in a statement, adding that the investment underpins its mission to use technology to make markets fairer and more liquid.
For CoreWeave, the announcement was the third major win of a remarkable April. On April 9, Meta Platforms expanded its existing cloud agreement with the company to a staggering twenty-one billion dollars, running through December 2032 and bringing Meta's total commitments to CoreWeave to roughly thirty-five billion dollars. The following day, Anthropic signed a separate multi-year, multi-billion-dollar agreement to run its Claude models on CoreWeave's platform at production scale. Together, the three announcements — from a social media giant, a frontier AI lab, and a Wall Street trading house — underline how broadly the demand for specialised AI compute has spread.
The numbers behind CoreWeave itself are equally striking. As of the end of 2025, the company's revenue backlog stood at $66.8 billion, more than four times the level recorded twelve months earlier. Full-year 2025 revenue reached $5.13 billion. The company now operates 43 active data centres and has contracted more than three gigawatts of power — a figure that puts it in the company of national electricity grids rather than corporate server rooms. Nine of the ten leading AI model providers are counted among its customers.
CoreWeave's rise is, in part, a consequence of structural constraints elsewhere. Tech giants have collectively committed more than $630 billion in capital expenditures this year, yet even that pace struggles to keep up with the demand for specialised GPU clusters needed to train and serve frontier models. Hyperscale cloud providers — Amazon, Microsoft, Google — remain dominant, but their general-purpose infrastructure does not always meet the specific performance and latency requirements of cutting-edge AI workloads. CoreWeave built its business precisely around that gap, offering dense GPU compute optimised for the kind of intensive training and inference work that modern AI demands.
The Jane Street deal introduces a new dimension to that narrative. Financial services firms have been among the most aggressive early adopters of machine learning for decades, but their infrastructure investments have typically been handled quietly and internally. A seven-billion-dollar public commitment to an AI cloud provider is a different kind of signal — one that suggests the firm views its ability to access specialised compute as a strategic asset comparable to its trading algorithms or market access.
It also raises a broader question about who the future customers of AI infrastructure actually are. The assumption has long been that the primary demand would come from technology companies building AI products for consumers or enterprises. The CoreWeave deals of April 2026 suggest the reality is more complex: quantitative finance, frontier model development, social media scale inference, and enterprise software are all converging on the same constrained resource, and all of them are willing to commit extraordinary sums to secure it years in advance.
CoreWeave went public on Nasdaq earlier this year and has spent much of its brief trading history educating investors about the economics of long-term infrastructure commitments. The April deal flow — more than twenty-eight billion dollars in new or expanded contracts across three very different kinds of organisations — has made that education considerably easier.
