Anthropic Signs $10B Compute Deal With Six-Month-Old Startup Volta: 133MW Norwegian Data Center Runs on Hydropower

Anthropic just wrote a $10 billion check for compute — to a company that was founded in January.

Three Key Takeaways

Six years, $10 billion, backed by Norwegian hydropower. The deal with Volta Infra secures 133 megawatts of capacity at a Norwegian data center running entirely on renewable hydroelectric power. The facility is operated by Bitcoin miner Bitdeer and will be delivered in two phases: December 31, 2026, and March 31, 2027. The data center will be equipped with Nvidia's latest Vera Rubin chips. J.P. Morgan and another major financial institution are arranging $1.3 billion in credit support for the contract.

Volta is six months old and already valued at $2.4 billion. Founded in January by former Brookfield Asset Management executives, Volta leases AI compute capacity and helps clients finance expensive chip purchases. Alongside this deal, Volta announced a $300 million venture funding round at a $2.4 billion valuation. A six-month-old company landing a $10 billion contract is unheard of in traditional infrastructure.

Anthropic's compute hunger is far from satisfied. This is not Anthropic's first big compute play. Amazon has invested over $8 billion into the company, and Google has put in billions more. But diversifying its own compute channels is clearly a strategic priority — Anthropic needs supply that is not locked to any single cloud provider.

WangDou's Take

Ten billion dollars in compute from a company that did not exist six months ago — only the 2026 AI industry could produce a deal this surreal. But peel it apart and the logic holds: Volta's founders come from Brookfield, where arranging multibillion-dollar infrastructure financing is a day job; Norwegian hydro is cheap and stable with natural PUE advantages; Bitdeer already has facilities ready for conversion. What Anthropic is really buying is not a startup's services — it is an independent compute pipeline that bypasses AWS and GCP. When you are burning tens of billions a year training models, putting all your eggs in one cloud provider's basket is the kind of thing that keeps you up at night.

Source: Bloomberg, TechCrunch, The Decoder