Anthropic signed a $35 billion cloud services agreement with Lambda, the Nvidia-backed infrastructure provider, tied to a 350-megawatt AI data center in Texas. The contract represents roughly five years of compute at Anthropic's current burn rate, effectively prepaying the infrastructure for Claude's next generation of training runs. Lambda now has $35 billion in forward revenue visibility from a single counterparty, and Anthropic has locked power and chips at a fixed cost through the back half of the decade.
The structure is unusual. Most frontier labs rent compute on shorter cycles or build their own clusters. Anthropic chose neither. The five-year commitment suggests the company sees stable demand for inference at scale and wants to derisk power availability in a market where 350 MW of contiguous capacity is no longer a given. Lambda gets balance-sheet certainty. Anthropic gets guaranteed access to the chips OpenAI and Google are also bidding for. The Texas site is purpose-built, which means it bypasses the retrofit constraints that have delayed other hyperscale AI projects by six to nine months.
The deal also clarifies Anthropic's path to $10 billion in annualized revenue, the threshold most allocators use to justify its $60 billion private valuation. If the company is spending $7 billion per year on compute alone, it needs either enterprise contracts at margins sufficient to cover that infrastructure or a consumer product with unit economics that hold at scale. Claude Pro subscriptions will not cover $35 billion over five years. That means the revenue model tilts toward API access, enterprise licensing, or embedded use cases where Anthropic collects per-token fees at volume. The Lambda deal front-loads the capex, which suggests Anthropic expects one of those channels to hit inflection in the next twelve to eighteen months.
Lambda's involvement is worth isolating. The company has Nvidia's backing, which gives it preferential access to H100 and H200 allocations. By committing $35 billion to Lambda instead of AWS or Azure, Anthropic signals it values chip access and power guarantees over the ecosystem integrations that come with hyperscalers. That matters for operators evaluating which infrastructure providers can actually deliver multi-hundred-megawatt projects on time. Lambda has now built one. The Texas site will be the reference architecture for the next wave of single-tenant AI clusters.
Allocators should track three things: whether Anthropic raises another equity round in the next six months, which would indicate the $35 billion compute commitment strained liquidity; whether Lambda announces similar deals with other labs, which would confirm the single-tenant model is replicable; and whether Texas power prices spike in 2026, which would reveal whether Anthropic locked favorable rates or simply locked availability. The site goes live in phases starting Q3 2025. If the first phase delivers on schedule, Lambda becomes a credible alternative to the hyperscaler oligopoly.
Anthropics's forward infrastructure is now clearer than its forward revenue. That is the trade.