Anthropic Secures $35 Billion with Lambda, Transforming Nvidia into Landlord, Chip Provider, and Equity Holder in the Same Contract
A six-year contract for 350 MW in Texas closes the week in which Anthropic amassed $130 billion in computing commitments. Nvidia anchors chip supply, equity, and the lease itself.
The Agreement and the Structure Behind It
Anthropic is reported to have signed a six-year contract valued at $35 billion to reserve computing capacity with Lambda, a cloud provider backed by Nvidia. Bloomberg, Reuters, and the Wall Street Journal reported the terms on Monday, September 1. Neither Anthropic nor Lambda have officially confirmed the deal as of the publication of this article.
The contracted capacity is around 350 megawatts, allocated in a data center developed by miner and operator Hut 8 in Nueces County, Texas. In July, Hut 8 had announced that a high-grade investment tenant signed two 15-year leases covering 704 MW on the campus, with a contracted base value of $19.6 billion, without naming the counterparty. The tenant, now identified, is Nvidia. Nvidia will sublease the space to allow Lambda to install the chips and sell computing services to Anthropic.
Nvidia Takes on Three Roles Simultaneously
The structure deserves attention. Nvidia appears in the contract as a supplier of the chips that will run in the data center, as a shareholder in Lambda, and as an anchor of the lease with Hut 8. Every dollar that Anthropic pays to Lambda fuels purchases of Blackwell and Rubin GPUs from Nvidia, increases Nvidia's stake in a customer of its own silicon complex, and helps dilute the real estate commitment of the manufacturer in Nueces County.
For CFOs needing to model concentrated exposure, this format changes the conversation. It is uncommon for the dominant silicon layer supplier to simultaneously undertake supply contract risk, equity risk, and real estate risk concerning the same customer. Industry analysts describe the model as circular financing. The criticism, when well-articulated, is that the demand for chips might start to depend on a customer sustained, in part, by the cash from the manufacturer itself. The valid counterargument is that Nvidia operates from a position different from that of VCs burning cash: it closed the January quarter with over $43 billion in cash and reports an operating margin exceeding 60%. One does not disqualify the other.
One Week, Three Contracts, $130 Billion in Commitments
The contract with Lambda adds to the $45 billion agreed upon with Nscale for capacity in West Virginia and the $50 billion with Fluidstack, both reported in the preceding two weeks. Thus, in less than fifteen days, Anthropic stacked $130 billion in infrastructure commitments, a figure greater than the combined annual revenue of Oracle, Cisco, and IBM. The company currently operates in the range of $5 billion in annualized revenue and is trending towards a valuation close to $1 trillion, according to recent reports from Fortune.
Here’s the data that weakens the very thesis of spending offset by scale: the cost of running queries on frontier models has fallen rapidly over the past two years, and competition among Anthropic, OpenAI, Google, and xAI is compressing the price per million tokens in each generation. The Claude Opus 4.7 currently costs a fraction of what Claude 3.5 Sonnet charged in 2024. Contracting 350 MW for six years assumes that inference demand doubles several times before the price curve necessitates renegotiation. It is not impossible, but it is not free to assume.
What Changes for the Corporate Buyer
CIOs negotiating multi-year AI contracts with Anthropic, OpenAI, or Google are beginning to see the same logic trickle down into SLAs. Those contracting dedicated capacity gain predictability but concentrate risk at two points: the solvency of the model provider and the health of the silicon supplier. In the United States, hyperscalers and independent operators are competing for hospital and financial contracts with minimum reservation clauses. In Germany and the United Kingdom, entities regulated by BaFin and the FCA already require contingency plans in case a model provider’s access is interrupted, a requirement that the EU’s AI Act and DSA only make more explicit. In Brazil, the same reasoning is beginning to appear in RFPs from major banks that standardize Claude and ChatGPT in critical operational flows.
What Anthropic signed with Lambda is not merely a supply line. It's a piece of the sector's capital structure being hand-written, contract by contract.