Reflection AI, a US startup developing open artificial intelligence models, has signed a landmark computing agreement worth more than $1 billion with European cloud provider Nebius Group, the companies announced.
The deal secures Reflection access to Nvidia's latest GB300 graphics processing units through 2029, with the startup agreeing to pay roughly $150 million per month starting July 1, 2026. For a company that has yet to release its first public-facing model, the commitment represents one of the largest pre-revenue infrastructure bets in the AI industry.
Founded in 2024 by a team including former researchers from leading AI labs, Reflection is developing open-source AI technology and positioning itself in the increasingly competitive race for compute capacity. The startup's strategy mirrors a broader trend in the AI industry: securing physical computing infrastructure years in advance, before models are even released, as demand for Nvidia's chips dramatically outstrips supply.
'Compute is the new oil, and Reflection just bought a refinery before drilling its first well,' one industry analyst noted.
Nebius, the Amsterdam-based cloud infrastructure provider, emerged from the restructuring of Yandex's international operations. The company has been aggressively expanding its GPU cloud capacity to meet surging demand from AI startups and enterprises. The deal with Reflection represents one of the largest single compute contracts Nebius has signed.
The agreement underscores a fundamental reality of the 2026 AI landscape: access to advanced computing hardware has become the single most important strategic asset for AI companies. With Nvidia's GPU supply constrained by TSMC's fabrication capacity and geopolitical tensions affecting chip distribution, startups are increasingly forced to sign multi-year, multi-billion-dollar commitments years in advance to secure the compute they need to train and run frontier models.
Reflection's billion-dollar bet before product launch is emblematic of the era. It echoes similar moves by other AI startups: Anthropic has signed multibillion-dollar cloud deals with Google and Amazon, and OpenAI has secured massive compute commitments from Microsoft. What distinguishes Reflection is the timing — securing this level of infrastructure funding before the company has a publicly known product.
The deal also highlights the rise of European AI infrastructure players. Nebius, along with other regional providers, is carving out a niche as an independent compute supplier, offering an alternative to the dominant US hyperscalers. For Reflection, tapping European capacity diversifies its supply chain in an environment where chip access has become a geopolitical variable.
Reflection's founders have not disclosed details about the model they are training, but the scale of the compute commitment suggests ambitions that rival those of much larger, better-funded competitors. If the bet pays off, Reflection could emerge as a significant new player in the open-weight AI space. If it does not, the $1 billion commitment will stand as a cautionary tale about the risks of the AI infrastructure arms race.
