Anthropic Signs $35 Billion Deal With Lambda, Accelerating the Race for GPUs

Anthropic signed a $35 billion cloud computing contract with Lambda, an Nvidia-backed provider, according to a source cited by Reuters and reported by ET Tech. The deal will bring Nvidia GPU capacity online to meet growing demand for Claude, the company's family of models.

The data center is located in Nueces County, Texas, and is being built by Hut 8, a company that shifted from cryptocurrency mining to AI infrastructure. Capacity is around 350 megawatts. According to the Wall Street Journal, which first reported the deal, Nvidia itself would hold the data center's lease, adding yet another layer to the already tight relationship between chipmaker, cloud provider, and model lab.

Two Giant Contracts in One Week

The detail that shows the scale of Anthropic's aggressiveness isn't the isolated figure, it's the timing. The week before the deal with Lambda, the company said it would spend $45 billion to lease cloud computing power at Nscale's data center campus in West Virginia. Combined, that's $80 billion in compute commitments announced in a matter of days by a company that, according to the report, is heading toward an IPO.

| Deal | Amount | Partner | Location | |---|---|---|---| | Cloud (Nscale) | $45B | Nscale | West Virginia (US) | | Cloud (Lambda) | $35B | Lambda (Nvidia-backed) | Nueces County, Texas (US) |

Hut 8 had reported in July that it had signed a 15-year lease agreement with an "investment-grade customer," with a base contract value of $19.6 billion. The Financial Times later reported that Nvidia was the tenant at Hut 8's 1-gigawatt Beacon Point campus. The deal with Anthropic fits another piece into this puzzle.

What's Behind the Numbers

Demand for computing infrastructure has surged since the arrival of generative AI services, pushing tech companies to invest hundreds of billions of dollars in data centers equipped with advanced chips from Nvidia and other manufacturers, the report notes. Anthropic has been aggressively expanding its computing power precisely to keep up with the growth it projects for products like Claude Code, its AI-assisted coding tool.

This point matters directly to those building software. When a lab commits to $80 billion in compute in a single week, much of that capacity goes toward serving inference (the responses you get when calling the API) and training the next generations of models. Tools like Claude Code, coding agents, and IDE integrations depend on that capacity being online, with low latency and stable availability.

What Changes for Developers in Brazil

The news is about a data center in Texas, but the impact still reaches the editor of developers here. Some concrete vectors:

• Availability and usage limits. Additional GPU capacity tends to reduce pressure on rate limits and the likelihood of degradation during peak hours, something anyone using the Claude API in production feels in the form of 429 errors and queues.
• Token cost. AI infrastructure is expensive, and the market hasn't stabilized pricing yet. Billion-dollar compute contracts are the fixed cost labs need to spread out. For Brazilian developers, who still pay in dollars and feel the exchange rate, the variable to watch is whether this scale translates into a lower price per token or just sustains the race without passing savings on. The source offers no pricing projection, so this remains open.
• Betting on coding as a core use case. Anthropic explicitly ties the expansion to demand for Claude Code. In other words, the company is betting that coding assistants are one of the uses that consume (and pay) the most for inference. For those choosing between Claude, GPT, and open models in their workflow, this is a signal of where Anthropic will focus its improvements.

The Compute Race as a Barrier to Entry

The size of these contracts redraws who can compete. Running a frontier lab is no longer just a matter of research talent, it has also become a matter of access to gigawatts of energy and batches of Nvidia GPUs. Smaller startups and open-model projects are competing for the same silicon queue, and Anthropic, by locking in capacity with long-term contracts, secures its own supply and, in practice, raises the cost of the race for everyone else.

For developers, the practical takeaway is architectural: the more frontier inference depends on a handful of providers concentrating capacity, the more it makes sense to design applications that aren't locked into a single vendor. Abstracting the model layer, measuring cost per request, and keeping the option to route to open or smaller models for tasks that don't require top-tier performance are decisions that gain weight in this landscape of concentrated compute.

What Remains Open

The information came from an anonymous source, and Anthropic, Nvidia, Hut 8, and Lambda did not respond to Reuters' requests for comment. There are no public details about the term of the $35 billion contract, the timeline for the capacity coming online, or how this affects API pricing. What the report confirms is the scale of the commitment and Anthropic's stated bet on coding products as the engine of this demand. The rest, including whether this scale translates into better and cheaper models for those who write code, is what the next wave of announcements will tell.