Anthropic is set to secure $10 billion worth of computing capacity from Volta Infra Holdings, an AI cloud startup founded only months ago, in a six-year agreement that highlights how quickly frontier AI companies are assembling infrastructure beyond traditional cloud providers.

The capacity will come from a data center in Tydal, Norway, where Bitdeer Technologies operates hydropower-fed infrastructure equipped with Nvidia’s latest Vera Rubin chips, according to reporting from The Decoder citing Bloomberg. Volta says the site will provide 133 megawatts, with the capacity handed over in two phases through March 2027.

The deal gives Anthropic another large source of compute as demand for model training and inference expands. It also links the AI developer to a young infrastructure company backed by several of the same financial and hardware interests that are shaping the broader AI supply chain.

A large commitment from a young provider

Volta was founded in early 2026 by former managers from asset manager Brookfield. Reuters reported that the startup is valued at $2.4 billion and has announced a $10 billion AI partnership with Anthropic, while Bloomberg reported that Nvidia and Michael Dell are among its backers.

According to The Decoder’s account, Volta has raised $300 million in venture capital in a round led by Andreessen Horowitz and Altimeter Capital. Nvidia and Dell are also identified as investors. The company has separately established a $5 billion financing pool intended to help customers pay for expensive AI chips upfront.

That financing structure is important because AI data centers require major capital commitments before customers can turn compute into revenue. A provider that can arrange both power and hardware financing may be able to bring capacity online faster than a conventional cloud operator, although the available evidence does not establish how much of Volta’s planned infrastructure is already operational.

The Norway project is tied to Bitdeer, a company better known for cryptocurrency mining. Its hydropower-fed site is being repurposed or expanded for AI workloads using Nvidia’s Vera Rubin systems. Bitdeer’s stock rose 14% after the announcement, according to The Decoder, suggesting that investors viewed the agreement as commercially significant for the infrastructure operator.

Anthropic’s expanding compute network

The Volta arrangement is not Anthropic’s only major infrastructure commitment. The Decoder says the company already has agreements involving Google and Broadcom, Amazon, SpaceX and AMD. The evidence supplied for those arrangements does not provide their individual values, timelines or capacity, so the list is best understood as an indication of Anthropic’s broadening supplier network rather than a complete picture of its available compute.

For Anthropic, using multiple providers can reduce dependence on a single cloud platform and give the company access to different combinations of chips, data-center locations and financing. It may also help the company secure capacity ahead of model launches or usage growth, when demand for advanced accelerators can exceed near-term supply.

The trade-off is operational complexity. Different providers can mean different networking designs, software stacks, security controls and availability terms. Moving workloads across them is not simply a matter of switching a billing account, particularly for large training runs that depend on tightly integrated clusters.

The Volta deal also illustrates a more complicated form of supplier dependence. Nvidia is reported to be both an investor in Volta and the supplier of the chips deployed at the Norwegian facility. That relationship may help Volta obtain capital and hardware, but it also means the economics of the cloud provider are closely connected to the company that supplies its principal computing equipment.

What the evidence confirms — and what it does not

The strongest details available from the source material come from The Decoder’s report, which attributes the six-year term and infrastructure figures to Bloomberg. Reuters independently confirms the central commercial development through its headline and summary: Volta has reached a $10 billion AI partnership and carries a $2.4 billion valuation. Bloomberg’s available headline independently identifies Nvidia and Dell as Volta backers.

The supplied Reuters and Bloomberg material does not include full article text, contract terms or direct statements from Anthropic, Volta, Nvidia, Dell or Bitdeer. As a result, the exact pricing, minimum utilization commitments, cancellation rights, delivery guarantees and revenue recognition schedule remain unclear.

Volta’s statement that it has secured 1 gigawatt of near-term data-center power is a company claim, not an independently verified deployment figure in the available evidence. Likewise, the $10 billion figure describes the value of the compute partnership; it should not automatically be treated as confirmed revenue, installed capacity or cash already paid to Volta.

Those distinctions matter for AI infrastructure buyers. A headline capacity figure can include future delivery, reserved power or hardware procurement rather than compute that customers can use immediately. The handoff schedule through March 2027 indicates that at least part of the commitment is still prospective.

Why builders and enterprise buyers should care

For AI builders, the deal signals that access to compute is becoming a strategic asset negotiated years or months ahead of actual demand. Startups training large models may need to secure power, accelerators and financing simultaneously, rather than relying on on-demand instances after a model is ready to run.

The arrangement also raises questions about portability. Workloads designed around Nvidia’s newest systems may deliver strong performance when the underlying cluster is available, but teams should evaluate how easily they can move training or inference to alternative hardware. Capacity reservations can improve predictability while narrowing flexibility if demand changes or a model strategy is revised.

Enterprise buyers face a related issue when selecting AI services. Anthropic’s growing infrastructure network could support more reliable access to models, but the network’s layered dependencies may make outages, supply delays or financial stress harder to assess. Buyers will want clarity on regional redundancy, service-level commitments, data handling and whether a model provider’s capacity is owned, leased or dependent on future build-outs.

At the market level, Volta’s rapid formation and multibillion-dollar valuation show how capital is flowing toward specialized AI cloud providers that can package power, chips and financing. That can accelerate deployment, but it can also amplify risk if expected AI demand fails to justify the infrastructure being built. The same investors and suppliers appearing across several layers of the stack could strengthen coordination while concentrating exposure to a downturn.

What to watch next

The first signal will be whether Volta delivers the promised 133 megawatts on schedule and whether both phases are available by March 2027. Public information about commissioning, chip installation and customer access will help distinguish reserved capacity from working compute.

Investors and customers should also watch for details on the six-year contract: payment structure, utilization obligations, pricing, termination provisions and responsibility for hardware upgrades. Those terms will determine how much risk Anthropic has taken on if model demand or economics change.

Further disclosures from Volta, Bitdeer or Nvidia may clarify whether the company’s 1-gigawatt power position represents permitted sites, connected facilities or a broader development pipeline. Anthropic’s future infrastructure announcements will show whether Volta is a major pillar of its strategy or one component in a deliberately distributed portfolio.

Creati.ai perspective

Anthropic’s Volta agreement is significant less because it adds another cloud vendor than because it shows how frontier AI companies are financing and assembling compute through networks of specialized providers, chip companies and investors. The model can bring infrastructure online quickly, but it makes ownership, delivery status and counterparty exposure harder to read from headline contract values.

For builders and enterprise buyers, the practical lesson is to treat compute access as a supply-chain decision. Capacity, portability, reliability and financial resilience should be evaluated together. Until Volta discloses more about delivery and contract mechanics, the $10 billion figure is best viewed as a major commitment to future capacity—not proof that the full amount has already become usable compute.