Reflection’s $1B Compute Deal With Nebius—AI Arms Race
Reflection AI signed a $1 billion compute deal with Nebius, the ex-Yandex infrastructure firm, to secure Nvidia's latest chips as open-model demand surges.
In Brief
- Reflection AI signed a $1 billion compute deal with European infrastructure firm Nebius to access Nvidia’s latest chips.
- Nebius, formerly Yandex’s international arm, also has up-to-$27B and $19.4B deals with Meta and Microsoft.
- Reflection, valued at $8 billion, has raised roughly $2.6B from Nvidia, Sequoia and Lightspeed since its 2024 founding.
The race for AI compute just got more expensive. Reflection AI, a US startup building open-weight models, signed a $1 billion compute deal with Nebius, a European AI infrastructure company, in a sign that the scramble for training capacity is now a balance-sheet event.
Nebius — formerly the international arm of Russian tech giant Yandex — will give Reflection access to Nvidia’s latest chips, according to TechCrunch. The agreement lands just weeks after Reflection signed a similar arrangement to tap SpaceX’s computing resources, and it mirrors a wave of partnerships as AI firms lock in hardware ahead of demand.
The timing is strategic. Open-weight developers like Reflection have gained attention as debate rages over the value of closed, top-tier models — especially after the Trump administration pressured Anthropic and OpenAI to restrict their most powerful releases, raising fears that access could be pulled overnight, as Frontierbeat documented in its anti-AI protest coverage.
The Compute Squeeze Is Reshaping AI
Compute is the bottleneck that decides which labs can train frontier models, and the largest players are buying it in bulk. Reflection’s back-to-back Nebius and SpaceX deals show startups pre-reserving capacity years before they need it, treating GPU access as a strategic asset rather than a line item.
Nebius is quietly becoming a critical neutral supplier. Shortly after securing a $2 billion investment from Nvidia, it signed a five-year infrastructure deal with Meta worth up to $27 billion, and last year it struck a multi-year agreement with Microsoft worth up to $19.4 billion, details laid out on Nebius’s site. That pipeline makes it a swing provider between the hyperscalers’ own clouds.
The demand underside is visible across the sector. TeraWulf’s CEO recently argued that “not all megawatts are created equally” in the AI buildout, a reminder that power, location and chip mix — not just raw capacity — separate winners, as CoinDesk reported. Reflection is effectively paying to secure all three.
Why Open Models Are Winning Capital
Reflection’s fundraising tells the story. Founded in 2024 by two former Google DeepMind researchers, it is already valued at $8 billion and has raised close to $2.6 billion from backers including Nvidia, Sequoia Capital and Lightspeed Venture Partners. That investor roster is itself a compute advantage: Nvidia’s capital and chips travel together.
The open-weight thesis is gaining ground precisely because of restriction anxiety. With more capable open models emerging from China and US policymakers signaling they will curtail the most powerful closed systems, mainstream interest in open-source AI has climbed. Reflection sits at the intersection of both forces.
The company’s positioning also reflects a broader strategic shift documented in Nadella’s warning about AI labs hoarding data, where the scarce resource is no longer the model architecture but the infrastructure to run it. A $1 billion compute pact is the clearest possible vote on where the constraint really lives.
FAQ
What is the Reflection-Nebius deal?
Reflection AI signed a $1 billion agreement with Nebius for access to Nvidia’s latest chips. It follows a similar deal Reflection struck to use SpaceX’s computing resources, as the startup secures training and inference capacity ahead of demand.
Who is Nebius?
Nebius is a European AI infrastructure company that was formerly the international arm of Russian tech giant Yandex. Beyond the Reflection deal, it has large infrastructure agreements with Meta and Microsoft, positioning it as a major neutral compute provider.
Why does compute matter so much for AI startups?
Training and deploying frontier models requires massive GPU capacity, and access to the newest chips is the main constraint on what a lab can build. Startups like Reflection are signing billion-dollar pacts to guarantee that supply rather than competing for it spot-market style.
