Anthropic bets $10B on a 6-month-old cloud startup to run Nvidia's next-gen chips

Anthropic just signed a six-year, $10 billion deal with Volta Infra Holdings — a cloud provider that literally didn’t exist six months ago. The compute will come from a hydropower-fed data center in Tydal, Norway, operated by Bitcoin miner Bitdeer Technologies. Inside those racks: Nvidia’s latest Vera Rubin chips. Volta says the facility runs at 133 megawatts, with capacity delivered in two phases through March 2027. Bitdeer’s stock popped 14% on the news.

Volta was founded in early 2026 by former Brookfield managers and has moved fast. Alongside the Anthropic deal, the startup raised $300 million in venture funding from Andreessen Horowitz, Altimeter Capital, Nvidia, and Michael Dell. That round values Volta at $2.4 billion. The company also assembled a $5 billion financing pool designed to help customers swallow the brutal upfront cost of cutting-edge AI chips. Not bad for a company that barely has a Wikipedia entry.

Anthropic isn’t putting all its chips on one table. The company already has compute agreements with Google, Broadcom, Amazon, SpaceX, and AMD. But the Volta deal represents something different — a direct line to Nvidia’s next-generation hardware through a startup that Nvidia itself has a stake in. That circular relationship is starting to raise eyebrows. Nvidia invests in Volta, then supplies the chips Volta rents to Anthropic. Critics argue these tangled dependencies could magnify losses if the AI boom cools faster than the infrastructure buildout assumes.

I’ve watched enough cloud deals to know that $10 billion commitments to six-month-old companies aren’t normal. This is hyperscale spending at startup speed. The Norway location matters too — cheap, clean hydropower is becoming a strategic asset in AI, and Bitdeer’s existing mining infrastructure gives Volta a faster path to deployment than building from scratch. The real question is whether Anthropic sees demand justifying this scale, or whether it’s simply terrified of being locked out of compute while competitors hoard capacity. Either way, the compute land grab is accelerating, and the lines between chipmaker, cloud provider, and AI lab keep getting blurrier.