In today's AI economy, a guaranteed supply of chips can be worth more than the model itself. Reflection AI, a US startup building open-weight models, has agreed to buy more than $1 billion in compute capacity from European infrastructure firm Nebius, the companies said on July 14, 2026. The multi-year deal runs through 2029 and gives Reflection access to Nvidia's latest chips.
Who Reflection is
Founded in 2024 by two former Google DeepMind researchers, Reflection is positioning itself as an open-source, open-weight alternative to closed labs like OpenAI and Anthropic. It is valued at roughly $8 billion and has raised about $2.6 billion from backers including Nvidia, Sequoia Capital and Lightspeed — capital that is now flowing straight into locking down infrastructure.
Why the deal matters to Nebius
Nebius, an Amsterdam-based AI-cloud provider that was formerly the international arm of Russia's Yandex, gets a marquee, long-dated customer commitment. Its shares (NBIS) rose about 3% on the announcement, made by Reflection co-founder Ioannis Antonoglou. For a "neocloud" competing with the hyperscalers, a billion-dollar anchor tenant is validation.
Stacking compute
This is not Reflection's only capacity play. The Nebius agreement is separate from an earlier deal with SpaceX for compute, underscoring how open-model challengers are assembling supply from multiple providers rather than betting on one. Securing multi-year access to frontier GPUs is the price of staying in the race against far better-capitalized rivals.
The compute land grab
The deal is another data point in a defining trend of 2026: access to chips, power and data-center space has become the primary competitive moat in AI. Long-term contracts that guarantee capacity through the end of the decade give startups the revenue visibility and training runway that used to come from the model alone — and make infrastructure providers like Nebius central players in who gets to build frontier AI.