Nscale buys Anyscale for $1.65B to lock down the full AI stack

The British AI cloud provider Nscale is acquiring Anyscale in a $1.65 billion deal, Bloomberg reports, citing an anonymous source. It’s a vertical integration play that makes a lot of sense. Nscale has been furiously building out everything from energy to data centers to orchestration software. What it lacked was a sophisticated layer for managing and scaling the actual AI workloads—exactly what Anyscale built its business on after pivoting from a general distributed computing platform when GPT-3 dropped in 2022 and changed the world.

Anyscale started life as the commercial outfit behind Ray, the open-source Python framework for distributed computing. Once large language models began guzzling compute, the company smartly refocused on the unglamorous but essential plumbing for AI: serving and training models, data curation, inferencing, and reinforcement learning at scale. The startup was last valued at $1.38 billion during its 2022 Series C, and it just reported a 70% jump in quarterly revenue. Not bad for a company that had to reinvent itself practically overnight.

The acquisition is a clean fit. Nscale, fresh off a $2 billion Series C that valued it at $14.6 billion, has backing from heavy hitters like Nvidia, Dell, and Nokia. It’s been signing deals with Microsoft, British Telecom, and others to lock up compute and data center capacity. By bringing Anyscale’s developer tools, observability, and workload orchestration in-house, Nscale can now co-design the software and the silicon it runs on—something Anyscale acknowledged neither company could do as effectively on its own.

Anyscale will keep its branding and its roughly 200 employees are joining Nscale. The real question is whether this sparks a consolidation wave among neoclouds all scrambling to offer a one-stop shop for AI compute. When the interface between software and hardware becomes a competitive advantage, leaving it to a third party starts to look like a liability.