Anthropic has signed a $10 billion computing capacity agreement with Volta Infra Holdings, a startup founded this year by former Brookfield Infrastructure executives, according to people familiar with the matter. The six-year term makes it one of the largest infrastructure deals struck by a major AI lab.
Volta Infra was founded in 2026 by Daniel Boada and Juan Gumuzio, both previously at Brookfield. The startup has secured backing from Nvidia and Dell, according to a funding round that valued it at $2.4 billion. Anthropic's deal represents a direct bet on Volta's ability to build and operate the data centers required to train and run large language models at scale.
Training and inference for models like Claude require sustained access to high-end GPUs. Rather than build its own infrastructure in-house, Anthropic is locking in capacity from a dedicated provider over a six-year window. The $10 billion commitment averages to roughly $1.67 billion per year.
Volta Infra enters a market crowded with established operators. CoreWeave, Lambda Labs, and Lambda operate similar GPU cloud services. Major cloud providers including Amazon Web Services, Google Cloud, and Microsoft Azure have all expanded their AI compute offerings to compete. Volta's advantage lies partly in its founding team's infrastructure expertise and its backing from hardware makers who have visibility into the next generation of accelerators.
Anthropic has grown rapidly since its 2021 founding. The lab closed a $5 billion funding round in May 2024 and is valued at $20 billion as of its most recent round. It competes with OpenAI and Google DeepMind for both talent and inference market share. Sustained access to compute at favorable terms is essential to that competition.
Volta's ability to deliver on $10 billion in committed capacity over six years will depend on its speed of deployment and hardware procurement in a market where GPU availability remains constrained. If Volta fails to build out capacity on schedule, Anthropic would face the risk of compute shortfalls during peak training runs. The agreement likely includes remedies or step-down provisions tied to delivery milestones.
