# Reflection signs a $1bn AI compute deal with Nebius, its second big capacity grab in a month > Source: > Published: 2026-07-15 14:30:17+00:00 Reflection AI has locked in more computing power. The US startup signed an AI compute deal worth more than $1bn with Nebius, the Amsterdam-based cloud provider, [according to Bloomberg](https://www.bloomberg.com/news/articles/2026-07-14/nebius-to-sell-1-billion-in-ai-capacity-to-startup-reflection). The agreement runs through 2029. It gives Reflection access to Nvidia’s newest GB300 chips. Nebius confirmed the deal. Its shares have more than doubled this year on the AI spending boom. They rose on the news before slipping later in the session. This is Reflection’s second big capacity deal in weeks. Last month it signed a [multibillion-dollar deal with SpaceX](https://thenextweb.com/news/spacex-reflection-ai-6-3bn-compute-deal) for the same class of Nvidia chips, reportedly worth about $150m a month through 2029. ## Who Reflection is Two former Google DeepMind researchers founded Reflection in 2024. It builds open-source models. The pitch is a cheaper, more customisable alternative to the closed systems from OpenAI and Anthropic. The startup is valued at about $8bn. It has raised close to $2.6bn from backers including Nvidia, Sequoia Capital, and Lightspeed Venture Partners. It is also in talks to raise a further $2.5bn at a $25bn valuation, [the Wall Street Journal reported](https://www.wsj.com/tech/ai/nvidia-backed-startup-seeking-to-counter-chinese-ai-eyes-25-billion-valuation-3bd8216c). “The need for open models is clear, and this additional compute capacity will allow Reflection to continue to build and train frontier AI models at scale,” said co-founder and chief technology officer Ioannis Antonoglou. ## A race for compute AI startups are racing to lock in hardware. Demand from businesses adopting the technology is rising faster than new data centres can come online. Capacity has become the constraint, and chip prices keep climbing. Open models are having a moment for two reasons. They are usually cheaper to run and easier to tailor. And last month the Trump administration pressured Anthropic and OpenAI to restrict some of their most powerful models, [per TechCrunch](https://techcrunch.com/2026/07/14/reflection-inks-1b-compute-deal-with-nebius/). That exposed the risk of relying on a provider that can be cut off overnight. ## Nebius steps up Nebius is a neocloud. It rents out AI computing capacity rather than building models. The company split from the Russian internet group Yandex in 2024 and now trades on Nasdaq. Nvidia has backed it repeatedly, from an early [$700m round](https://thenextweb.com/news/nvidia-accel-nebius-ai-chip-funding-700-million) to a later [$2bn investment](https://thenextweb.com/news/nvidia-invests-2-billion-in-nebius). Nebius has also bought its way up the stack, [acquiring the inference startup Eigen AI](https://thenextweb.com/news/nebius-eigen-ai-inference-optimization). The client list is heavy. Nebius struck a five-year deal with [Meta worth up to $27bn](https://thenextweb.com/news/meta-nebius-27-billion-ai-infrastructure-deal), and an earlier one with Microsoft worth up to $19.4bn. The Reflection agreement adds another marquee name. It also shows how much of the open-model push now runs on rented infrastructure. The models may be open. The compute behind them is anything but free. ## Get the TNW newsletter Get the most important tech news in your inbox each week.