Reflection AI, a US startup founded in 2024 by two former Google DeepMind researchers, has signed a US$1 billion compute deal with Nebius, an Amsterdam-headquartered AI infrastructure company, to access Nvidia’s latest chips for training and deploying its models.

Reflection said it plans to release open weights, publish related research, and open-source its software and reinforcement-learning tools, making large compute reservations central to its strategy.

Nebius provides AI infrastructure through its AI cloud, server, and data center operations. The company emerged after Yandex sold its Russian assets in 2024 and the Dutch parent changed its name to Nebius Group.

The agreement follows Reflection’s earlier deal with SpaceX, under which the startup would pay US$150 million a month from July 1, 2026, through 2029 for immediate access to Nvidia GB300 chips and supporting hardware at SpaceX’s Colossus 2 site near Memphis.

Nebius has also disclosed multi-year infrastructure agreements with Meta and Microsoft, but the headline values reflect total contract value over several years rather than near-term recognized revenue, as the company said demand still outstrips supply.

Reflection was valued at US$8 billion and has raised close to US$2.6 billion from backers including Nvidia, Sequoia Capital, and Lightspeed Venture Partners.

🔗 Source: TechCrunch