Anthropic locked $35 billion in compute capacity with Lambda, the Nvidia-backed cloud provider, tied to a roughly 350 megawatt artificial intelligence data center under construction in Texas. The multi-year agreement represents the largest single-tenant AI infrastructure commitment disclosed to date and marks the first time a foundation model builder has contracted compute at utility-scale capacity before facility commissioning.

Lambda operates as a specialized AI cloud provider, positioned between hyperscalers and bare-metal GPU clusters. The Texas facility, expected online by late 2025 or early 2026, will house an estimated 100,000 to 120,000 H100-equivalent GPUs based on power density assumptions. Anthropic's contract secures exclusive or priority access to a substantial fraction of that capacity. The deal structure is believed to include fixed pricing across the contract term, insulating Anthropic from spot-rate volatility that has pushed H100 hourly costs above $4.50 in certain markets during peak demand windows. Lambda's backing by Nvidia includes not only equity capital but also preferred GPU allocation, giving the provider access to chips often reserved for hyperscaler orders.

The deal matters because it formalizes a new procurement pattern among frontier AI labs: locking years of compute capacity in single transactions rather than scaling month-to-month with AWS, Azure, or Google Cloud. Anthropic's largest known previous commitment was a $4 billion AWS credit facility announced in late 2023. This Lambda agreement is nearly nine times that size and shifts the compute risk from variable cloud bills to a fixed multi-year obligation. For Anthropic, the trade-off is clear: certainty of supply during the training window for Claude 4 and subsequent models, which are expected to require 10x to 20x the compute budget of Claude 3 Opus. For Lambda, the deal converts speculative data center investment into contracted revenue, de-risking the facility before power turns on.

Allocators should track three follow-on events. First, whether Amazon or Google announce similar multi-year, single-tenant AI compute deals in the next 90 to 120 days—silence would suggest hyperscalers are losing pricing power in the foundation model segment. Second, Lambda's next fundraising round, likely in Q2 or Q3 2025, will price the company's ability to replicate this model with other AI labs. Third, Anthropic's next disclosed equity raise or credit facility, expected within six months, will clarify whether this compute lock-in creates new financing needs or was pre-funded by existing backers including Google and Menlo Ventures.

The Texas facility's 350 MW draw is equivalent to roughly 280,000 U.S. homes and positions it among the ten largest AI-specific data centers globally by power capacity. Lambda has not disclosed the facility's exact location, but ERCOT interconnection queues show three projects in that range under development in the Dallas-Fort Worth and Austin corridors, all with 2025 or 2026 target dates.