Reflection AI inks $1B Nebius deal as open model compute wars heat up

The race to secure raw computing power just got a lot more expensive. Reflection AI, the two-year-old startup founded by ex-Google DeepMind researchers, has locked in a $1 billion compute deal with Nebius, the European AI infrastructure firm that spun out of Yandex. The agreement gives Reflection access to Nvidia’s latest chips, a resource that has become the geopolitical and commercial bottleneck of the entire AI industry.

This isn’t a one-off shopping spree. The Nebius pact lands just weeks after Reflection signed a separate deal to tap SpaceX’s computing resources. The startup, now valued at $8 billion and armed with roughly $2.6 billion from backers including Nvidia and Sequoia, is stockpiling silicon at a pace that suggests it’s preparing for a massive training run — or several. For a company building open-weight models, that kind of hardware commitment is a direct challenge to the walled gardens of OpenAI and Anthropic.

The timing is politically charged. Last month, the Trump administration pressured both Anthropic and OpenAI to restrict access to their most powerful models, a move that sent a chill through enterprises worried about having their AI supply chain yanked overnight. When combined with increasingly capable open models emerging from China, that regulatory squeeze has turned “open source” from a philosophical preference into a strategic imperative. Reflection is betting that demand will only accelerate.

Nebius, for its part, is quietly becoming the arms dealer in this conflict. Since securing a $2 billion investment from Nvidia, the company has inked a five-year deal with Meta worth up to $27 billion and a multi-year agreement with Microsoft valued at $19.4 billion. A billion-dollar commitment from a startup — even one as well-funded as Reflection — signals that the battle between open and closed AI is no longer a debate. It’s a balance sheet war. The question now is whether all this compute translates into models that actually matter, or if we’re watching the infrastructure equivalent of a land grab where nobody’s quite sure what to build yet.