Reflection AI Signs $1 Billion Compute Deal With Nebius
The open-model startup, valued at $8 billion, is stacking up GPU access at speed as demand for AI that nobody can switch off overnight keeps growing.
Key points
- Reflection AI, a U.S. Startup founded in 2024, signed a $1 billion compute deal with European AI infrastructure company Nebius in July 2026.
- Nebius, formerly the international arm of Russian tech giant Yandex, will supply Reflection with access to Nvidia's latest GPU chips.
- Reflection has raised close to $2.6 billion from backers including Nvidia and Sequoia Capital.
- The deal follows a separate agreement Reflection signed weeks earlier to use SpaceX's computing resources.
- Nebius signed a five-year infrastructure deal with Meta worth up to $27 billion shortly after receiving a $2 billion investment from Nvidia.
Reflection AI wants to build models that anyone can download and run without asking a company for permission first. To do that, it needs an enormous amount of raw computing power, and this week it secured another large slice.
The startup signed a $1 billion agreement with Nebius, a European AI infrastructure company providing access to GPUs, the specialised chips that handle the heavy number-crunching AI models require. Nebius was originally the international arm of Yandex, the Russian technology giant, before spinning out independently. TechCrunch AI first reported the deal.
Reflection is two years old, founded in 2024 by two former Google DeepMind researchers, and already valued at $8 billion. It's also struck a separate agreement to access computing resources through SpaceX, the rocket company that operates large data centres. Stacking multiple compute agreements has become standard practice among AI companies racing to train larger models.
Why does this matter to ordinary people?
It matters because of who open-weight models are attracting right now. Open-weight means the underlying code and settings are released publicly for anyone to use or modify.
Interest has spiked for a pointed reason. Last month, the Trump administration pressured Anthropic and OpenAI to restrict their most powerful new models. Businesses and public institutions that had built services on those tools suddenly faced a real concern: access could disappear with a policy change. Our story on Chinese open-weight models dominating developer downloads traced that same anxiety taking hold among developers this spring.
Open models sidestep that risk. If the code is already on your server, no policy reversal can pull it. That calculus, combined with increasingly capable open models from China, has pushed open-source AI firmly into the mainstream.
The practical thing to know, if you're evaluating AI tools right now, is whether the service you depend on is closed-source, meaning the company controls access entirely, or open-weight, meaning you can host it yourself. That distinction could matter considerably if the policy climate shifts again.
Should you worry about who controls AI infrastructure?
Nebius is building itself into a serious infrastructure player fast. Shortly after a $2 billion investment from Nvidia, it signed a five-year deal with Meta worth up to $27 billion. The previous year it signed a multi-year deal with Microsoft worth up to $19.4 billion. Adding Reflection extends a client list that now spans the biggest names in AI.
The pattern worth watching isn't the dollar figures. It's how quickly compute is concentrating among a handful of providers, which means decisions made by a small number of infrastructure companies will shape what AI developers can build, and how reliably they can build it.
