Nscale buys Anyscale for $1.65B to expand its AI compute stack

British AI neocloud Nscale is buying Anyscale, a software startup that helps companies scale AI workloads across data centers and servers, as it seeks to capture more of its customers’ AI spending. Nscale is paying $1.65 billion for Anyscale, Bloomberg reported, citing an anonymous source.

Anyscale was founded by the team behind the open-source Project Ray distributed-programming Python framework and started out building a platform for running projects that needed large amounts of computing power. After the launch of GPT-3 in 2022 put AI in the spotlight, the company pivoted to offer scaling services for serving and training large language models, data curation, inferencing, reinforcement learning, and other tasks. Its platform, built around Ray, offers developer tools, observability, and orchestration.

The deal fits neatly into Nscale’s strategy of building vertically to serve compute needs: the neocloud already runs business lines across energy, data centers, and orchestration software, and Anyscale would add workload management and scaling. “Together, Anyscale and Nscale can co-design the software layer and infrastructure beneath it, something that neither company could do as effectively by optimizing its layer alone,” Anyscale said in a statement.

Nscale raised $2 billion in a Series C round this March that valued it at $14.6 billion, with investors including Nvidia, Nokia, Blue Owl, Dell, and Norwegian industrial giant Aker. It has been putting that money, along with various debt raises, to work securing compute and data center partnerships with Microsoft, British Telecom, and Nordcraft. Anyscale, valued at $1.38 billion in a 2022 Series C round, said its revenue grew 70% in its most recent quarter compared with the previous one. Nscale said Anyscale will keep operating under its own branding and serving its existing customers; the startup’s approximately 200 employees are all joining Nscale.

Why it matters

The deal shows AI neoclouds racing to own more of the compute stack, pairing raw infrastructure with the workload-orchestration software that determines how efficiently that infrastructure gets used, rather than competing on hardware access alone.