Nscale Spent Two Years Buying Power Plants and GPUs. Its Next $1.65 Billion Purchase Was Software.

For the last two years, the fastest way to win in AI infrastructure has been simple, if not cheap: buy land, buy power, buy GPUs, repeat. Nscale, the London-based "neocloud" that emerged from almost nowhere to a $14.6 billion valuation in March, built its entire rise on that formula — striking long-term chip-supply deals with Nvidia and multi-site GPU commitments with Microsoft, including a planned 200,000-GPU buildout and campuses in Norway and Portugal.

On July 30, Nscale changed the formula. It announced a definitive agreement to acquire Anyscale, the software company behind the open-source Ray framework, for roughly $1.65 billion — its first major move up the stack, from owning compute to owning the software that decides how that compute gets used.

What Nscale actually bought

Anyscale isn't a household name, but Ray is quietly load-bearing infrastructure across the AI industry. Built originally at UC Berkeley's AI research lab and commercialized by the same team as Anyscale, Ray is the distributed-computing framework that lets a workload — training a model, serving it, curating its data, running reinforcement learning — spread cleanly across hundreds or thousands of machines instead of choking on a single box. Anyscale layered developer tooling, observability, and managed orchestration on top of that open-source core and sells it to companies running large-scale AI workloads.

The deal brings roughly 200 Anyscale employees into Nscale, with Anyscale continuing to operate under its own brand and serve its existing customers. It's expected to close in the second half of 2026, pending regulatory approval. Notably, governance of the open-source Ray project itself moved to the PyTorch Foundation under the Linux Foundation back in October 2025 — so Nscale is buying Anyscale's commercial business and engineering talent, not control of the community project that made Ray ubiquitous in the first place.

Anyscale had been growing briskly on its own: the company reported 70% sequential revenue growth in its most recent quarter, a healthy jump for a company last valued at $1.38 billion in a 2022 Series C — meaning Nscale is paying roughly a 20% premium over that four-year-old valuation for a business that looks considerably bigger today than it did then.

Why an infrastructure company wants an orchestration company

The logic, as Anyscale put it in its own statement on the deal, is that "together, Anyscale and Nscale can co-design the software layer and infrastructure beneath it." Translated out of press-release language: Nscale already controls the expensive, physical part of the AI compute stack — the power contracts, the data center real estate, the GPU clusters it's filling with Nvidia silicon for Microsoft. What it hasn't controlled is the software that determines how efficiently all of that hardware actually gets used. A GPU cluster sitting half-idle because workloads aren't scheduled well is money leaking out of a business built entirely on the premise that compute is scarce and expensive.

That's the same math that's been driving nearly every AI infrastructure story this summer. Just three days before the Anyscale deal, Nvidia and OpenAI structured a financing backstop around a $500 billion Ohio data center project specifically because the capital math on frontier-scale compute has gotten so large and so circular that no single balance sheet wants to hold the risk alone. A day before that, AMD and Cerebras unveiled a disaggregated inference architecture aimed squarely at squeezing more useful tokens out of every watt of power — because raw chip count has stopped being the only scoreboard that matters.

Nscale buying Anyscale fits the same pattern from a different angle: if you can't cheaply add more power or more chips, you extract more value from the ones you already have. That's an orchestration and scheduling problem, not a chip-fab problem — which is exactly the gap Ray and Anyscale's tooling are built to close.

The bigger shift this points to

It's also a data point for a thesis that's been building quietly since early summer: that the next wave of meaningful AI company-building isn't happening in chatbots or even foundation models, but in the unglamorous middle layer — the software that manages, monitors, and allocates AI infrastructure at scale. Nscale didn't need a bigger model or a flashier consumer product to become one of Europe's most valuable AI companies; it needed power contracts, GPUs, and now, apparently, a scheduler.

Whether that bet pays off depends on something Nscale can't fully control: whether the current pace of AI infrastructure spending is sustainable, or whether it's a bubble waiting for a correction. But for now, the message from one of the AI industry's fastest-growing infrastructure players is unambiguous — in a compute-constrained world, the company that decides how a GPU spends its next millisecond is worth almost as much as the company that owns the GPU.

Sources

Nscale buys Anyscale as it seeks to own more of the AI compute stack - TechCrunch: https://techcrunch.com/2026/07/30/nscale-buys-anyscale-as-it-seeks-to-own-more-of-the-ai-compute-stack/

Nscale to Buy AI Software Startup Anyscale for $1.65 Billion - Bloomberg: https://www.bloomberg.com/news/articles/2026-07-30/nscale-to-buy-ai-software-startup-anyscale-for-1-65-billion

Nscale raises $2bn in Series C funding at $14.6bn valuation - Data Center Dynamics: https://www.datacenterdynamics.com/en/news/nscale-raises-2bn-in-series-c-funding-at-146bn-valuation/

Nscale AI data center Nvidia raise - CNBC: https://www.cnbc.com/2026/03/09/nscale-ai-data-center-nvidia-raise.html