CoreWeave to Host Millions of New GPUs for Meta, Gains Soros Backing
Nvidia’s multiyear expansion with Meta will route millions of Blackwell and Rubin GPUs through cloud partners like CoreWeave into Meta’s data centers for AI training and inference. Soros Fund Management added CoreWeave exposure, though analysts flag three-year risks from hyperscalers, high capital intensity, and potential customer insourcing.
1. Nvidia and Meta Expand Partnership
Nvidia and Meta extended their multiyear, multi-generational agreement, securing millions of Blackwell and Rubin GPUs plus Grace CPU-only servers for Meta’s data centers. The deal also covers networking hardware and Confidential Computing capabilities for WhatsApp, with Vera CPU-only systems slated for 2027 deployment.
2. CoreWeave’s Role as Nvidia Cloud Partner
As a member of Nvidia’s Cloud Partner program, CoreWeave will host the newly provided GPUs for client rental and usage. Increased chip allocations for Meta are expected to drive higher utilization rates and revenue growth for CoreWeave’s data center operations.
3. Soros Fund Management Adds CoreWeave
Soros Fund Management increased its exposure to CoreWeave in its latest portfolio filing, signaling confidence from a major hedge fund. This move underscores growing institutional interest in specialist AI infrastructure providers.
4. Analyst-Raised Structural Risks
Industry analysts warn of three primary risks over the next three years: hyperscalers like AWS and Google Cloud offering competitive services, persistent high capital expenditures limiting return on invested capital, and major customers potentially insourcing compute as supply constraints ease.