Nscale just bought Anyscale. Here’s why it matters for multi-cloud neutrality.

Nscale started life two years ago as a crypto-mining outfit. Now it's swallowing Anyscale, the company behind the commercial version of Ray, in a deal reportedly worth $1.65 billion, and suddenly it's asking to be taken seriously as something called a "full-stack AI hyperscaler." That's a big leap, and the interesting part isn't the price tag — it's what happens to neutrality once a GPU landlord owns the software tenants use to decide where to run their workloads.

Anyscale built its reputation on being agnostic. Ray, the open-source distributed computing framework its founders created, was designed to orchestrate Python-based AI workloads across whatever cloud you happened to be using, and Anyscale sold the commercially supported, no-DevOps version of that promise. Nscale, on the other hand, is a neocloud in the strictest sense: it owns the GPUs, the datacenters, the power contracts, the whole vertically integrated stack. Pairing the two means the company that used to help you avoid vendor lock-in is now owned by a vendor.

Nscale's chief product officer, Dan Bathurst, insists nothing changes for existing customers — Anyscale keeps its brand, keeps running on AWS, GCP and Azure, and Nscale wants to win on performance rather than force anyone onto its own infrastructure. He frames it as "neutrality on the platform layer, differentiation on the infrastructure layer," meaning you can still run Anyscale anywhere, but if you run it on Nscale's own hardware, you get a co-engineered, fully optimized stack, from power supply down to the application. Fair enough, on paper.

But analyst Sanjeev Mohan, formerly of Gartner, isn't buying the framing wholesale. He points out that the moment Anyscale's best pricing and sharpest performance land first on Nscale's own fleet, neutrality becomes a marketing label rather than a technical reality. Software running everywhere and software running best somewhere are not the same claim, and customers will notice the gap in cost and speed even if the code technically works on a rival cloud. Mohan also sees this as part of a bigger pattern: standalone orchestration and tooling vendors keep getting absorbed by whoever owns the GPUs, because the economics of co-designing hardware and software only pencil out when one company controls both.

What Nscale actually becomes, in Mohan's view, isn't a hyperscaler rival to AWS or Azure — it lacks the sprawling menu of managed databases and services those giants offer. It's a specialist, a serious option for raw training and inference at scale, competing on a narrower but potentially lucrative slice of the market. The deal still needs regulatory sign-off and isn't expected to close until the second half of 2026, so there's plenty of time for AWS, Google and Microsoft to decide how threatened they actually feel by a former crypto miner with an $1.65 billion appetite.