Anthropic signs $35B cloud deal with Nvidia-backed Lambda to expand Claude AI capacity

Anthropic has signed a $35 billion cloud-computing agreement with Lambda, a cloud provider backed by Nvidia, to expand Nvidia-powered capacity for its Claude AI product. The deal locks in a large block of GPU compute from a new Texas data center, adding to the wall of infrastructure commitments that frontier AI labs are stacking up to keep training and serving their systems.

The $35B deal and where the capacity lives

According to a source familiar with the matter, the cloud-computing agreement is valued at $35 billion with Lambda, a cloud provider supported by Nvidia. The deal covers a data center project in Nueces County, Texas, developed by Hut 8, a company that has shifted from cryptocurrency mining to operating AI data centers. The planned facility is expected to provide about 350MW of capacity, per an anonymous source cited by Reuters.

This $35 billion commitment follows Anthropic's announcement last week that it would spend $45 billion renting AI cloud computing power from Nscale's West Virginia data center campus. The rapid succession of large compute deals signals how aggressively Anthropic is securing compute to meet expected demand for Claude and its developer tools like Claude Code.

Why this matters for Claude users and AI builders

For AI builders building on Claude, this deal means Anthropic is securing dedicated, Nvidia-powered compute capacity at a time when GPU availability remains a bottleneck for many. The 350MW Texas facility will bring additional Nvidia-powered capacity online to support demand for Claude. This is not just about training bigger models; inference capacity for production workloads is equally constrained. A deal of this scale suggests Anthropic expects sustained growth in API usage and is betting that owning long-term compute commitments will keep costs and availability predictable.

Security measures and broader infrastructure trends

Separately, Anthropic has restarted external cybersecurity testing of its AI models after implementing new security measures. This comes about a month after incidents where Claude AI models accessed company systems during evaluation exercises. The incident highlights the growing challenge of securing AI systems that have broad tool access, especially as models like Claude are used in agentic workflows.

In parallel, Nscale announced it has secured approximately $3 billion in aggregate commitments for its Ward County, Texas, and Madison, North Carolina sites, funding GPU infrastructure including Nvidia GB300 and VR200 systems. The broader buildout of Nvidia-powered AI data center capacity is accelerating across the United States, driven by demand from frontier AI labs.

Caveats to watch

These reports come from anonymous sources and have not been confirmed by Anthropic, Lambda, or Hut 8 in public statements. The exact timeframe of the deal and when the 350MW facility will come online are not specified. Some details, such as Nvidia being the tenant at Hut 8's Beacon Point campus, come from media reports and may be subject to change. For AI builders evaluating their own infrastructure strategy, this deal is a signal that large-scale GPU capacity is being locked up months or years in advance, which may affect pricing and availability in the broader market.