CoreWeave Diversifies Revenue with Multibillion AI Contracts

CoreWeave secured multiple multibillion-dollar commitments this month that materially diversify its customer base and reduce revenue concentration risk. Key headline moves include Jane Street committing $6 billion for compute plus a $1 billion equity investment at $109 per share, Meta expanding to a $21 billion deal, and a multi-year agreement with Anthropic. These contracts increase CoreWeave's contracted backlog into the tens of billions and shift reliance away from Microsoft, which accounted for roughly 67% of CY25 revenue.

CoreWeave will deploy next-generation Nvidia silicon across its facilities, including access to Vera Rubin GPUs that Jane Street will run on CoreWeave clusters. Management projects a major capacity ramp, with plans to double active GPU capacity and bring all contracted power online by Q4 CY27. Expect sustained capital intensity: management signaled $30-35 billion in capital expenditures this year to buy chips, build data centers, and secure power. The company also carries a significant financing profile, with roughly $30B of debt outstanding, making execution on power, build schedules, and chip deliveries the immediate operational focus.

• •Jane Street: $6 billion compute commitment, $1 billion equity at $109 per share; access to Vera Rubin hardware and CoreWeave tooling.
• •Meta: Expanded multiyear agreement to $21 billion, bringing Meta commitments higher when combined with prior deals.
• •Anthropic: Multi-year capacity agreement as one of several frontier AI lab customers.
• •Existing customers: OpenAI and Microsoft remain large users; Microsoft accounted for a concentrated share of CY25 revenue but recent deals reduce that concentration.

CoreWeave sits in the neocloud segment supplying physical GPU capacity and integration services to frontier labs, hyperscalers, and nontraditional compute consumers. These announcements underscore three converging trends: intensified demand for frontier GPUs, enterprises and trading firms adopting frontier-scale models and hardware, and neoclouds capturing share by pairing proximity to Nvidia supply with flexible commercial terms. Jane Street's move is especially notable because it signals non-AI-native firms will pay frontier pricing and secure advanced silicon for training and inference workloads. The participation of Vera Rubin hardware ties CoreWeave directly to Nvidia's supply cadence, which is the deterministic constraint for many model builders.

The deals change the economics and availability profile for large-batch training and inference. More contracted capacity at CoreWeave means a larger pool of specialty-hosted Vera Rubin resources outside hyperscaler stacks, potentially easing scheduling and procurement friction for labs that cannot or will not run on hyperscaler clouds. For ML engineers and infrastructure teams, this increases vendor choices for high-performance training, but it also raises operational complexity around data locality, interconnect performance, and procurement timing.

Execution on power procurement, chip deliveries, and capacity commissioning will determine whether backlog converts to usable capacity on schedule. Watch quarterly deployment metrics, realized utilization rates, and any changes to contract terms tied to Vera Rubin availability. Also monitor CoreWeave's balance sheet actions to manage heavy capex and roughly $30B of leverage.

"Training large, complex models on massive volumes of noisy data," said Jane Street in its statement, signaling that frontier compute economics are extending into quant trading and finance. The practical upshot is that frontier compute is now a cross-industry commodity, and CoreWeave is positioning itself as a critical aggregator and allocator of that scarce resource.