Anthropic has agreed to buy six years of cloud computing capacity from Volta, an AI infrastructure startup founded earlier this year, in a deal reportedly worth $10 billion. Bloomberg first reported the arrangement, citing people described only as familiar with the negotiations, and TechCrunch reporter Lucas Ropek laid out the details on August 4, 2026. Anthropic itself has not confirmed either figure.
The size of the commitment says more than the price tag does. Locking six years of capacity to a company whose public track record only began this year means Anthropic is betting on a supplier before that supplier has finished proving it can deliver at scale. If Volta stumbles on construction, financing, or chip delivery, Anthropic inherits the delay.
Volta plans to build the underlying facility in Norway together with Bitdeer, a company that built its business mining cryptocurrency before turning toward AI infrastructure. The site is rated at 133 megawatts and will run on NVIDIA’s newest Vera Rubin chip generation, the architecture NVIDIA unveiled earlier this year as its successor to prior GPU lines. Volta also sits inside NVIDIA’s Cloud Partner program, the network of infrastructure providers NVIDIA works with directly to place its chips into commercial data centers.
A dollar figure hides what actually gets built. Chip supply gets most of the attention in coverage of AI infrastructure, but power availability is what actually decides how fast a site this size can come online and how much it can run once it does. Norway offers two inputs few other locations can match at this scale: hydroelectric electricity priced well below the global average, and an ambient climate cold enough to cut the cost of keeping thousands of chips from overheating. Measuring the deal in megawatts rather than dollars is the more honest read, because 133 megawatts is a hard ceiling on how much Claude training and inference this one site can support, while a headline price can still move in negotiation.
The Volta agreement fits a wider pattern that is becoming visible across Anthropic’s recent deal flow. The company has already announced large blocks of AWS Trainium capacity and additional Google TPU capacity, and it is reportedly the largest single customer on a rival lab’s own cloud service. Separately, it is now building an internal chip design team. A lab that owns no fabs and no data centers of its own is assembling supply from every major vendor, and several new ones, at the same time.
Volta had previously described talks with an unnamed AI lab without saying who the counterparty was. Bloomberg’s account rests on sources it identifies only as people familiar with the deal, and TechCrunch noted that Anthropic did not provide details when asked directly for comment. Until one of the companies puts a number on the record, both the $10 billion price and the six-year term should be read as reported, not verified.
Anthropic’s earlier compute agreements with SpaceX and Amazon already show a company that treats every willing supplier as insurance against any single one falling through. Operators negotiating their own multiyear infrastructure contracts should treat the Volta terms, once either side confirms them, as a fresh benchmark for what six-year AI capacity costs per megawatt in a market where power, not silicon, is now the scarcer input.
TechCrunch’s Lucas Ropek reported this deal on August 4, 2026, citing Bloomberg’s original reporting on the Anthropic-Volta agreement.
