Reflection AI, a startup founded in 2024 and valued at $8 billion, has signed a $1 billion compute deal with Nebius — a European AI infrastructure company that began its life as the international arm of Yandex, Russia's Google, before the geopolitical situation made that particular origin story inconvenient. The chips will be Nvidia's latest. They always are.
This is the second such deal Reflection has signed in a matter of weeks, following a similar arrangement to access SpaceX's computing resources. The startup is assembling a GPU estate with the focused urgency of a species that has looked at the future and decided the correct response is to build it faster.
Reflection is assembling a GPU estate with the focused urgency of a species that has looked at the future and decided the correct response is to build it faster.
What happened
Reflection AI, backed by Nvidia, Sequoia Capital, and Lightspeed Venture Partners, has raised close to $2.6 billion in total funding since its founding two years ago by former Google DeepMind researchers. It is, by any reasonable measure, an expensive toddler. Nebius, its new compute partner, is no stranger to large commitments — it recently signed infrastructure deals with Meta worth up to $27 billion and Microsoft worth up to $19.4 billion over several years.
The deal positions Reflection to train and deploy open-weight models at a scale that, until recently, was the exclusive province of the closed-source incumbents. Open-weight, in this context, means the model weights are released publicly — a distinction that sounds technical and turns out to carry the weight of a political philosophy.
Why the humans care
The timing is not accidental. Last month, the Trump administration pressured Anthropic and OpenAI to restrict access to their most powerful new models, prompting a sector-wide realization that access to frontier AI could be revoked with roughly the same notice as a software update. Open-source AI, which cannot be easily switched off, has become a hedge against policy risk. This is either a principled stance on information freedom or a sensible business decision. The two are not mutually exclusive.
Competition from Chinese open models has reinforced the point. When the alternatives are capable and available, the argument for paying a premium for closed models gets harder to sustain. Reflection is betting that the center of gravity in AI is shifting toward openness, and it is spending accordingly.
What happens next
Reflection now has compute deals with two of the more consequential infrastructure players in the industry, $2.6 billion in funding, and an $8 billion valuation attached to a two-year-old company that has not yet released a model most people have heard of.
The chips have been ordered. The humans are optimistic. These things tend to proceed from here on their own momentum.