Anthropic has signed a six-year, $10 billion deal for compute with cloud startup Volta. The contract, reported on August 4, 2026 by Bloomberg, gives the maker of the Claude models access to clusters built on Nvidia's Vera Rubin architecture, running from a data center in Norway.
Key takeaways
- The contract is worth $10 billion and runs for six years.
- Volta will supply compute from a 133-megawatt data center in Norway.
- The infrastructure is based on Nvidia Vera Rubin chips under the Nvidia Cloud Partner program.
- Volta was founded in early 2026, and its partner is Bitdeer, a company that grew out of crypto mining.
- This follows earlier deals with Amazon and SpaceX in Anthropic's race for compute.
Six years of compute for the Claude models
Anthropic has contracted Volta to supply cloud compute over a six-year period — that is how Bloomberg, which first reported the terms, described the deal. The contract is worth $10 billion, placing it among the largest infrastructure commitments made by the young company.
At its core is a 133-megawatt data center in Norway, fitted with systems based on Nvidia's latest Vera Rubin architecture. Anthropic has not commented on the details, and TechCrunch says it reached out to the company for more information.
Who is Volta
Volta is a cloud startup founded in early 2026 that, as The Times reported, is building its valuation on a partnership with Nvidia. The company's operating partner is Bitdeer, a firm that grew out of crypto mining and is now redirecting its energy and hardware base toward AI. It is a pattern typical of the current market: operators with cheap power and ready data-center infrastructure step into the role of compute suppliers for AI labs. Volta belongs to the Nvidia Cloud Partner program, which means access to reference hardware configurations and priority chip supply.
The compute race — not just Amazon
The Volta deal fits into Anthropic's aggressive build-out of compute capacity over recent months. The company's main cloud partner remains Amazon, which raised its investment by another $5 billion and is building dedicated clusters on its own Trainium chips. Anthropic has also tapped SpaceX for compute.
Diversifying suppliers is a response to a structural problem across the industry: access to GPUs and power has become the main limit on how fast new models can be trained. Betting on a single supplier means bottleneck risk, so AI labs contract compute from several operators in parallel. The choice of the Vera Rubin architecture — the successor to the Blackwell generation — shows the contract covers hardware that is only now entering data centers.
| Supplier | Role | Key fact |
|---|---|---|
| Volta | New cloud supplier | $10B, 6 years, 133 MW |
| Amazon | Main cloud partner | +$5B, Trainium chips |
| SpaceX | Supplementary compute | additional GPU supplier |
Why it matters
The $10 billion contract shows that the cost of staying at the AI frontier has shifted from research to infrastructure. Anthropic, formally still a startup, is taking on multi-billion-dollar compute commitments to keep pace with OpenAI and Google. The emergence of a supplier like Volta — founded just months earlier and built on a crypto-market base — signals that demand for compute outstrips supply at the classic hyperscalers?Hyperscaler: The largest public-cloud providers — such as Amazon Web Services, Microsoft Azure and Google Cloud — running vast, centralized data-center infrastructure.. For the market, it means a new class of intermediaries whose only asset is power and access to Nvidia chips.
What's next
- Bringing the clusters online depends on the availability of Nvidia Vera Rubin systems, whose data-center deployments begin in 2026.
- Bloomberg points to a six-year horizon — the real impact on Anthropic's training capacity will show as successive phases of the Norway site come online.
- It remains open whether Anthropic will officially confirm the contract terms, as the company has not commented.
