Anthropic has reportedly signed a cloud computing contract worth $35 billion with the Nvidia-affiliated provider Lambda, according to consistent media reports. The computing power is said to come from a roughly 350-megawatt data center in Texas being built by the former Bitcoin miner Hut 8. Both companies have not officially confirmed the deal so far.
Lambda delivers Nvidia chips from a Texas data center
Hut 8, a company that has transformed from Bitcoin mining to AI data center operator, is developing the so-called Beacon Point campus in Nueces County, Texas, with a capacity of around 350 megawatts. The Wall Street Journal first reported the deal on August 31: Nvidia reportedly holds the rental rights to the area, while Lambda installs Nvidia chips there and resells the resulting computing power to Anthropic. Reuters confirmed the agreement with its own independent source. The contract duration remained unspecified in the reports, as did the amount that Lambda pays Nvidia for access to the area.
Neither Anthropic nor Lambda has publicly commented on the reports so far. An Anthropic spokesperson declined to comment, and Nvidia, Lambda, and Hut 8 initially did not respond to press inquiries. The deal amount of $35 billion itself is independently unverified. The nested structure of four involved companies shows how strongly Nvidia now acts as both a supplier and a financier and space renter in the data center business of its own customers.
Additional billion-dollar contracts secure computing power for Claude
The Lambda deal is not the first of its kind. Just at the end of August, Anthropic secured computing power for $45 billion from Nscale, equipped with Nvidia’s new Vera Rubin chips in West Virginia. Two more deals are added according to Bloomberg: $50 billion with the provider Fluidstack and $45 billion with Elon Musk’s SpaceX. Anthropic is responding to a shortage of computing power that arose in early 2026 when the demand for Claude grew faster than the available infrastructure – recently, Anthropic had to reduce usage limits for Claude Code.
At the same time, Anthropic is building its own chip expertise: an in-house chip design team is supposed to lower inference costs, bolstered by former Google TPU head Amir Salek in the compute team. Additionally, Google linked loans of $200 billion to its own chip deliveries to Anthropic in August. The investments come at a time when Anthropic is preparing for its planned IPO in fall 2026 and is presenting investors with a valuation of up to two trillion dollars.
Hut 8 benefits the most from the announcement
Market participants reacted faster than the involved companies themselves: The stock of data center operator Hut 8 rose by up to four percent in pre-market trading. All 18 analysts covering the stock recommend it as a buy, as a review by Yahoo Finance shows. Analyst Mark Palmer from Benchmark raised his price target to $195 – about 143 percent above the then-current price – and called the deal a “confirmation of the ‘Power First’ thesis,” according to which scarce energy is converted into long-term rented infrastructure.
Hut 8 had already signed a 15-year contract worth $9.8 billion with Nvidia for the space at the Beacon Point campus in May, which can total up to $25.1 billion with three extension options. The new Lambda-Anthropic contract builds on this basis, without the involved companies disclosing the financial relationship between the two contracts. For Hut 8, this structure primarily means lower risk: because Nvidia acts as a tenant, the former crypto miner does not bear the occupancy risk of the billion-dollar facility alone.
It remains to be seen whether the four computing power deals, along with the in-house chip team, will be sufficient to finally resolve the shortage before Anthropic goes public in the fall. The total of the known contracts exceeds $175 billion, without a single company having officially confirmed all the figures. It will be crucial whether investors demand more precise contract details before the IPO than journalists have managed to obtain so far.
