Reflection AI signs $1 billion compute agreement with Nebius
The open-model AI startup is adding access to Nvidia chips as compute capacity becomes a key bottleneck in the AI race.
By Theo Nakamura · Staff Writer
· 3 min read
Reflection AI has agreed to a $1 billion compute deal with Nebius, TechCrunch reported, giving the U.S. startup access to more of the Nvidia chips needed to build and run advanced AI models. For everyday investors tracking the AI trade, the deal is another sign that demand for computing capacity is becoming a central business driver for infrastructure providers.
Compute is the processing power used to train and operate artificial intelligence systems. In practice, that usually means access to high-end chips, data centers and cloud infrastructure. AI companies need large amounts of it to train models on data and to serve users once those models are deployed.
Nebius, a European AI infrastructure company that was formerly the international arm of Russian technology company Yandex, will provide Reflection with access to Nvidia’s newest chips, according to TechCrunch. The agreement follows a similar arrangement Reflection recently signed for access to SpaceX computing resources.
The deal fits a broader pattern across the AI sector: model developers are locking in infrastructure partnerships as they compete for scarce capacity. The largest AI systems require expensive hardware, and companies that can secure chip access may be better positioned to train and deploy their models at scale.
Why Reflection is drawing attention
Reflection was founded in 2024 by two former Google DeepMind researchers, according to TechCrunch. The company is working on open AI models and is currently valued at $8 billion.
Open-weight AI models make key model files available for others to use or adapt, unlike closed-source systems controlled entirely by their developers. TechCrunch reported that Reflection is among several open-weight model developers receiving more attention as the market debates the value and risks of top-tier closed AI systems.
That debate has picked up as data retention concerns have increased and governments have taken a closer interest in model access. TechCrunch reported that the Trump administration last month pressured Anthropic and OpenAI to restrict their most powerful new models, raising worries that access to certain AI systems could be limited quickly.
TechCrunch also pointed to more capable open models from China as one reason mainstream interest in open-source AI has increased. That puts companies like Reflection in a closely watched part of the market, where openness, performance and control are all part of the discussion.
Nebius adds another large AI infrastructure customer
Nebius has been signing large infrastructure agreements of its own. After securing a $2 billion investment from Nvidia, Nebius signed a five-year deal with Meta worth up to $27 billion, according to TechCrunch. Last year, Nebius also signed a multiyear agreement with Microsoft worth up to $19.4 billion.
Reflection has raised close to $2.6 billion from investors including Nvidia, Sequoia Capital and Lightspeed Venture Partners, TechCrunch reported. TechCrunch said it has contacted Reflection and Nebius for more information about the new agreement.
This story draws on original reporting from TechCrunch.
