Nvidia and Nscale cash in as Anthropic's compute bill hits billions

Nvidia just reported $96.2 billion in quarterly revenue, with its data centre division alone pulling in $89 billion, more than double what it made a year ago. The company is forecasting $108 billion for the next quarter, which would make it the first chip company to break $100 billion in a single quarter. Those are extraordinary numbers, and the reason behind them is not complicated: AI labs cannot get enough compute, and they are spending at a pace that was difficult to imagine even two years ago.

The latest illustration of that appetite comes from Anthropic, which has signed a $45 billion, six-year deal to rent compute from Nscale, a British AI infrastructure company founded only in 2024. That deal will be powered by Nvidia's Vera Rubin chip system, a six-chip architecture that represents the current cutting edge of GPU design, with capacity expected to come online from Nscale's West Virginia data centre in late 2027. The beneficiaries here are obvious: Nvidia sells the hardware, Nscale sells the capacity, and Anthropic writes the cheques.

Anthropic's compute spending spree, deal by deal

The Nscale agreement is not a one-off. Over the past eight months, Anthropic has signed compute deals totalling well over $60 billion across a string of suppliers. Earlier this month it committed $10 billion to Volta, a cloud startup founded in January 2025, for capacity from a Norwegian data centre. In July it signed a $5 billion deal with AMD. In May it disclosed a computing arrangement with SpaceX, reportedly delivering $1.25 billion worth of capacity per month from two SpaceX data centres. In April it expanded partnerships with both Amazon, gaining access to an additional 5 gigawatts of compute, and with Google and Broadcom for further power capacity. The pace and scale of these commitments suggests that for Anthropic, securing raw compute has become as strategic as building the models themselves.

Nscale's rapid rise on the back of big-name contracts

Nscale is a striking case study in how quickly infrastructure companies can scale when hyperscaler and frontier-lab demand is this intense. Founded in 2024, it had already secured a contract with Microsoft before landing the Anthropic deal. A $45 billion commitment from one of the most-funded AI labs in the world will cement its position as a serious player in European and transatlantic AI infrastructure. For buyers of AI tools, the rise of Nscale and similar companies like Volta signals that the data centre market is diversifying beyond the established hyperscalers, which could eventually introduce more competitive pricing, though that benefit is unlikely to filter down to end users any time soon.

Nvidia's Vera Rubin chips at the centre of the infrastructure race

The Nscale deal puts Nvidia's Vera Rubin system at the heart of Anthropic's future compute stack. Vera Rubin combines six different chip types working together and is the successor to the Blackwell architecture that already drives much of Nvidia's current data centre revenue. The fact that a company founded in 2024 is already committing to deploy Vera Rubin at scale, with a major AI lab as its anchor tenant, underlines how quickly the hardware generation cycle is turning. Nvidia's Edge Computing segment, which includes its consumer gaming business, brought in $7.2 billion last quarter, a reminder that the real money is in the AI infrastructure buildout, not gaming.

What this means for businesses choosing AI tools

For organisations evaluating AI platforms, this infrastructure spending matters in a practical way. Anthropic's aggressive compute acquisition is explicitly aimed at keeping pace with OpenAI and Google, both of which are pursuing the same strategy. Labs that win the compute race are better positioned to run larger models, reduce latency, and hold down inference costs. The risk is that long-term supply commitments, particularly deals locked in until 2031 and 2033, could reduce flexibility if model efficiency improves faster than expected. For now, though, the message from both the earnings data and the deal flow is consistent: the compute infrastructure layer is where the structural money in AI currently sits, and Nvidia, Nscale, and a handful of other suppliers are collecting most of it.