Skip to content

AI

Nscale pays $1.65B for Anyscale, the people behind Ray

British neocloud Nscale is reportedly buying Anyscale for $1.65 billion. The goal: own more of the stack between your GPUs and your invoice.

Cover photo: DrHughManning / CC BY-SA 4.0

TL;DR

  1. Nscale is buying Anyscale for $1.65 billion, per Bloomberg citing an anonymous source.
  2. Anyscale's team built Ray, the open source Python framework for distributed computing.
  3. Nscale already sells energy, data centers and orchestration. Workload scaling is the next shelf.

What happened

British AI neocloud Nscale is buying Anyscale, a software startup that helps companies spread AI workloads across data centers and servers. The price tag is $1.65 billion. That figure comes from Bloomberg, which cited an anonymous source, so treat it as reported rather than announced.

Anyscale was founded by the team behind Ray, the open source distributed programming framework for Python. It began as a platform for projects that needed serious amounts of computing power.

Then GPT-3 arrived in 2022 and the world discovered AI. Anyscale pivoted toward scaling services for serving and training large language models, plus data curation, inferencing and reinforcement learning. Its platform is built around Ray and adds developer tools, observability and orchestration.

Nscale's pitch is vertical. It already runs business lines across energy, data centers and orchestration software. With Anyscale, it adds workload management and scaling to the pile.

Why it matters

This is a neocloud moving up the stack. Renting out GPUs is one business. Owning the software that decides how your training run or inference job gets spread across those GPUs is a stickier one.

If Nscale controls the layer where customers actually write and run their scaling logic, it captures more of their AI spending. That is the stated motive here, and it is not subtle.

The deal also says something about Ray. A widely used open source framework with a commercial company behind it is now heading inside an infrastructure provider. Nothing in the reports says Ray stops being open source, but teams that depend on it will want to watch the roadmap.

For the rest of the market, it is one more sign of consolidation in the compute stack. Chips, power, data centers and now orchestration are getting bundled under fewer roofs.

Renting GPUs is a business. Owning the software that tells them what to do is a better one.

What to do with it

  • Check your Ray dependency.If your pipelines run on Ray or the Anyscale platform, note who owns it after the deal closes.
  • Read the contract, not the press release.Ask your current compute vendor what happens to pricing and support if your orchestration layer changes hands.
  • Compare bundled and unbundled.One vendor for energy-to-scaling may be simpler, but it is also one vendor to be locked into.
  • Wait for confirmation.The price is reported, not announced, so hold off on strategy decks that quote it as fact.

Ciao, sapiens. The robots now have a landlord, and he wants the whole building.