Anthropic signed a $35 billion cloud computing agreement with Lambda, the Nvidia-backed infrastructure provider, tied to a 350 megawatt artificial intelligence data center under construction in Texas. The deal represents one of the largest committed compute purchases in the AI sector and marks Anthropic's latest use of a revenue-share financing structure designed to convert future inference demand into present-day capacity.
The arrangement functions as a long-term capacity reservation rather than immediate capital outlay. Lambda secures guaranteed utilization across a timeline that likely spans five to seven years, while Anthropic gains access to compute infrastructure without the balance-sheet strain of direct ownership or traditional cloud contracts. The Texas facility, once operational in late 2025 or early 2026, will house tens of thousands of Nvidia H100 and next-generation Blackwell GPUs. Lambda has raised over $800 million in equity and debt since 2022, with Nvidia participating in multiple rounds. The 350 MW power allocation places the site among the top twenty AI-focused data centers globally by energy draw.
The agreement reflects a financing model Anthropic has deployed at least three times in twelve months. The company previously structured similar deals with AWS and reportedly with at least one other hyperscaler. These contracts allow Anthropic to scale inference and training workloads without the capital intensity of owning data centers or purchasing chips outright. For Lambda, the deal provides multi-year revenue visibility and justifies the capital required to build out the Texas site. For Nvidia, indirect exposure to Anthropic's growth through Lambda's chip purchases strengthens demand for its next-generation silicon without direct customer concentration risk.
The model also signals a shift in how frontier AI labs manage liquidity. Anthropic raised $7.3 billion across multiple rounds in 2024, including from Amazon, Google, and Menlo Ventures. Rather than deploying that capital into hardware, the company is using revenue commitments to unlock infrastructure built and financed by third parties. This preserves cash for research headcount, safety infrastructure, and model development while transferring capital risk to partners with cheaper cost of capital and longer investment horizons. The structure works only if Anthropic's inference revenue scales predictably—a assumption that depends on enterprise adoption of Claude continuing at current or faster rates.
Operators should monitor Lambda's debt refinancing activity in Q2 2025, as construction timelines for facilities of this scale require staged capital calls. Watch for additional Anthropic capacity announcements in Europe or Asia, where similar revenue-share models could unlock regional compute without direct subsidiary formation. Nvidia's earnings guidance for fiscal Q1 2025, due in late May, will clarify whether Lambda's chip orders appear in data center revenue or are routed through cloud service provider channels.
The Texas facility's power allocation was finalized in January 2025, eight months after Lambda began site preparation. The speed of Anthropic's commitment—reportedly signed within six weeks of Lambda securing the power contract—suggests the company is operating with month-level forward visibility on compute requirements, not quarter-level. That timeline compression is the new planning cadence at the frontier.