Anthropic has finalized a $35 billion cloud agreement with Lambda, which is supported by Nvidia.

      Anthropic has pledged $35 billion to a relatively unknown cloud provider for space in a Texas data center being constructed by a former bitcoin miner. The Wall Street Journal reported on the Lambda agreement on Monday, which was subsequently confirmed by Reuters through a source familiar with the deal.

      Neither company has publicly announced the agreement, and no specific terms have been disclosed. We have not independently verified the report, which is based on unnamed sources. Lambda is backed by Nvidia, and the deal's structure adds an intriguing layer. Lambda holds the lease for a data center in Nueces County, Texas, developed by Hut 8, while Nvidia has a separate agreement with Hut 8 to ensure the capacity, with Lambda utilizing Nvidia chips in the process.

      Following this chain of transactions, Nvidia is involved at three different points despite not being a formal party in the agreement. It has invested in the tenant, secured the site, and provides the hardware that populates it. The chipmaker has been establishing this framework throughout the year, aiding cloud providers in securing the necessary financing and real estate to deploy its systems.

      This follows the same rationale behind the revenue-sharing program it launched in July and paused last week due to internal antitrust concerns. For Anthropic, this deal is among several substantial agreements made in quick succession. Earlier this month, it agreed to pay $45 billion over six years with Nscale for capacity in West Virginia. Amazon separately announced that Anthropic plans to spend over $100 billion on AWS in the next decade, aiming for five gigawatts of capacity and access to Trainium3 silicon. In comparison, the $35 billion commitment is not the largest among its agreements.

      The pattern is now well established. Anthropic has signed a $9.1 billion, 20-year deal with bitcoin miner Riot Platforms, a $10 billion agreement with a startup that was just a week old, and a larger package with Google and Broadcom. Anthropic is also in preliminary discussions to lease $10 billion of compute from Meta, which is a noteworthy negotiation between two companies that compete for the same engineers and enterprise budgets.

      What connects these companies is a preference for multiple suppliers over a single one. A company with a singular product model has a clear incentive to avoid relying on one compute provider to run it. Beneath the diversification, there lies a concentration issue. Spreading contracts among various providers may appear resilient until one realizes that most are utilizing the same vendor’s chips in facilities that vendor helped to establish.

      Bitcoin miners frequently show up in these agreements for practical reasons. They possess grid connections, land, and power contracts acquired over several years, which are now the most scarce resources in the industry and significantly harder to obtain than chips. Hut 8 is the second miner-turned-operator in Anthropic’s supply chain after Riot. The transformation of crypto infrastructure into AI infrastructure has quietly emerged as one of the significant structural changes over the past two years.

      Texas is also playing a major role in these developments. Its affordable land, distinct grid, and faster permitting process make it the go-to choice for those in urgent need of a gigawatt. None of these commitments represent costs that have already been incurred; they are multi-year obligations against revenue that has been growing rapidly but hasn’t yet reached the levels assumed by the contracts.

      Anthropic’s run rate has been increasing quickly enough to justify the figures, having exceeded $30 billion earlier this year. Whether it can achieve the levels these contracts expect remains uncertain, distinct from the question of whether it is growing. That is the gamble in each of these agreements. Anthropic is securing now the capacity it will require if demand continues, at prices established before knowing whether it will materialize.