Nvidia deepens CoreWeave partnership to scale AI data centers

doubles its stake from a previously disclosed level, according to reporting and company statements, while CoreWeave said the new funds are intended to accelerate procurement of land, power and the shell infrastructure needed to bring large-scale capacity online.

Alongside the equity investment, Nvidia and CoreWeave said they are expanding their commercial relationship, including an agreement for Nvidia to purchase up to $6.3 billion in cloud services through 2032, a structure that can provide revenue visibility while CoreWeave builds out supply ahead of demand.

CoreWeave, which pivoted from crypto mining into a specialized “neocloud” focused on AI workloads, has positioned itself as a GPU-heavy capacity provider for customers running training and inference at scale, and Nvidia has been using stakes and long-term commitments to reinforce the ecosystem of large buyers for its data center hardware.

The deal also broadens the product stack CoreWeave plans to offer. CoreWeave will add Nvidia’s Vera CPUs, extending Nvidia’s reach beyond GPUs into the server processor layer where incumbents include Intel and AMD.

Investment lands amid a broader surge in AI infrastructure spending, as cloud providers, startups and enterprises race to secure GPU capacity for training large language models and running inference. Across the sector, operators are competing for access to power, land and networking equipment, with multi-hundred-megawatt campuses increasingly becoming the standard unit of expansion. CoreWeave’s 5-gigawatt target would place it among the largest dedicated AI data center buildouts currently planned in the US.

Power availability has emerged as a key bottleneck for the next wave of AI data centers. Operators are increasingly prioritizing locations with access to transmission infrastructure and long-term electricity contracts, while utilities and grid operators face pressure to accommodate large, fast-arriving loads. CoreWeave said the fresh capital will be directed toward securing sites and energy capacity early in the development cycle, a strategy aimed at reducing delays between hardware delivery and facilities coming online.