Anthropic’s $45B Nscale Deal: What the Compute Spend Signals Before IPO

Anthropic has agreed to spend $45 billion renting AI computing power from cloud infrastructure provider Nscale, according to a person familiar with the matter cited by Reuters on August 27, 2026. The six-year deal taps Nscale's West Virginia data center campus and will run on Nvidia's new Vera Rubin chips. It is one of several massive infrastructure commitments Anthropic is stacking ahead of what is expected to be a major IPO, as the company projects revenue of $190-200 billion by 2028.

What happened

Detail	Fact
Deal value	$45 billion
Provider	Nscale (West Virginia data center campus)
Contract length	Six years
Power capacity	~460 megawatts
Chips used	Nvidia Vera Rubin
Current revenue run rate	$47 billion
2028 revenue projection	$190-200 billion
SpaceX compute payment	$1.25 billion per month through May 2029
AMD commitment	Tens of billions in AI servers, up to $5B investment in Anthropic

Anthropic will rent AI cloud computing capacity from Nscale, a cloud infrastructure provider, over a six-year term. The arrangement centres on Nscale’s West Virginia campus and will be powered by Nvidia’s Vera Rubin chips, the next generation of GPU hardware. Bloomberg News reported the deal first; Reuters confirmed it through a separate source on August 27, 2026. Anthropic declined to comment.

The stated driver is demand for products like Claude Code, Anthropic’s AI coding tool. To serve that demand, the company has been spending heavily across GPU capacity, model training, inference infrastructure, and hiring. This Nscale agreement is one piece of a much larger picture.

How big is Anthropic’s total compute bill?

The Nscale deal does not sit alone. According to SpaceX’s IPO filing, Anthropic agreed to pay SpaceX $1.25 billion per month for compute capacity through May 2029. That capacity spans both of SpaceX’s AI training clusters, Colossus and Colossus II. In July, Advanced Micro Devices announced plans to sell Anthropic tens of billions of dollars in AI servers while also committing up to $5 billion as a direct investment in the company.

Stacking these commitments together, Anthropic is locking in supply across multiple hardware vendors and data center operators simultaneously. That is unusual even by the standards of the current AI infrastructure race, and it points to a company betting very hard on its own revenue forecasts.

Why it matters

Anthropic’s internal projections, reported exclusively by Reuters earlier in August, put 2028 revenue at $190-200 billion. Its current run rate sits at $47 billion. That is a four-fold increase in roughly two years, a target that would require an enormous amount of compute to actually deliver on. These infrastructure deals are, in part, a signal to IPO investors that supply is secured.

For the broader AI market, this deal reinforces how much power (literally, in megawatts) frontier model companies need. The 460-megawatt commitment to Nscale alone is equivalent to powering a mid-sized city. It also puts pressure on competing providers and data center operators to lock in similar partnerships before capacity gets committed elsewhere. You can see related supply dynamics playing out in stories we have covered, like Amazon tripling its Nvidia GPU order to 2 million chips for AWS.

For businesses using Claude or planning to, scale of this kind is generally good news for availability and pricing stability. More compute usually means faster inference and less throttling. But it also means Anthropic’s cost base is enormous, which creates pressure to monetise aggressively.

Our take

The numbers here are hard to contextualise without squinting a little. A $45 billion compute contract paid to a single provider over six years is larger than most countries’ annual tech budgets. The 2028 revenue target of $190-200 billion would make Anthropic bigger than many S&P 500 companies. These are projections, not results, and the gap between a $47 billion run rate and a $200 billion target leaves a lot of room for things to go wrong.

That said, the structure of these deals is interesting. Anthropic is distributing compute across SpaceX, AMD, and now Nscale rather than going all-in on a single hyperscaler like AWS or Google Cloud. That diversification reduces vendor lock-in risk, even if the total spend is staggering.

If you are building on Claude’s API today, the practical implication is that Anthropic is not going anywhere and is actively securing the infrastructure to grow. For businesses thinking about AI integration built on top of frontier models, that continuity matters more than the headline dollar figure.

What to do about it

1. Review your current AI vendor mix. If you rely heavily on a single model provider, consider whether redundancy across providers makes sense for your use case.
2. Watch Anthropic’s IPO filing for actual revenue and cost disclosures. Projections are one thing; audited numbers will tell a different story.
3. If you are evaluating Claude Code or other Anthropic products for your team, this level of infrastructure investment suggests service stability for at least the next several years.
4. Track Nvidia Vera Rubin availability. As Nscale deploys these chips for Anthropic, benchmark performance data will follow. That data is useful for any business comparing AI hosting options.

The bottom line: Anthropic is spending like a company that believes its own projections. Whether those projections are realistic will become clearer once an IPO prospectus forces them to show the actual numbers.

Source: Bing News · Anthropic