Nscale and Anthropic in Talks for $45 Billion AI Cloud Deal

Nvidia-backed Nscale is in talks with Anthropic for a new AI cloud deal worth about US$45 billion, a person familiar with the matter said. The report, carried by The Business Times, gives no closing date for the negotiations and no breakdown of the contract term.

The figure, if confirmed, would rank the agreement among the largest single-customer compute deals yet signed in the AI infrastructure market. It would also mark a sharp step up for Nscale, a cloud provider that counts Nvidia among its backers and has built its business on GPU capacity for AI training and inference workloads.

Both the value and the existence of the talks rest on one unnamed source at this stage. Neither Nscale nor Anthropic has confirmed the discussions publicly.

What do we know about the deal?

The reported number is US$45 billion. That is the only confirmed figure in the story, and it is attributed to a single source rather than to the companies.

What the report establishes:

• Nscale and Anthropic are in negotiations for a new AI cloud agreement.
• The deal value under discussion is approximately US$45 billion.
• The information comes from a person familiar with the matter, not from an official announcement.
• Nscale is backed by Nvidia.

What the report does not establish:

• Whether the parties have signed anything.
• The duration of the proposed contract.
• Which GPU generations or data center sites would serve the agreement.
• Whether the US$45 billion represents total contract value over multiple years or another measure.

Anthropic, the maker of the Claude model family, is an existing large buyer of cloud compute. The company has relied on capacity from major providers including Google Cloud and Amazon Web Services, both of which are also investors in Anthropic. A direct agreement with Nscale would add a dedicated channel aimed at its accelerating compute demand.

Why does Nvidia's backing matter?

Nscale's position as an Nvidia-backed operator sits at the center of the commercial logic. GPU supply remains the binding constraint in AI infrastructure, and allocative priority for Nvidia's highest-demand accelerators shapes which cloud providers can win hyperscale commitments from frontier labs.

An Anthropic commitment of this scale would effectively reserve a large block of Nscale's future GPU capacity before it is built. For Nscale, a signed contract near US$45 billion would provide the revenue visibility that data center financing requires, allowing the company to raise debt and equity against contracted cash flows rather than speculative demand.

The structure mirrors the model that has come to define the current buildout cycle: model developers sign long-term take-or-pay compute agreements, and infrastructure providers use those contracts to fund construction. Each side secures what it cannot source on the open market — the lab locks in scarce GPUs, the provider locks in demand.

How large is this in context?

Multi-billion-dollar compute contracts have become the norm as frontier AI models consume power at data center scale. Recent agreements between major cloud providers and AI labs have run from the tens of billions into the hundreds of billions over their full terms, and a US$45 billion negotiation would place Nscale in territory normally occupied by the largest hyperscalers.

For a Nvidia-backed challenger rather than an incumbent cloud giant, the scale of the reported talks signals how far down the stack AI demand now reaches: model developers appear willing to place frontier-scale commitments with newer operators when capacity and silicon access are guaranteed.

What happens next?

Nothing is binding until the companies announce a signed agreement, and talks of this size routinely change in value or fall apart. If the deal closes near the reported figure, expect confirmation from Nscale and Anthropic with details on capacity, sites and term — the specifics that will determine whether US$45 billion becomes contracted revenue or a negotiating marker in a market where compute commitments keep rising.