Reflection AI Signs $1 Billion Nebius Chip Deal as Open-Model Race Intensifies
Updated
Updated · TechCrunch · Jul 14
Reflection AI Signs $1 Billion Nebius Chip Deal as Open-Model Race Intensifies
3 articles · Updated · TechCrunch · Jul 14
Summary
$1 billion will buy Reflection AI access to Nvidia’s latest chips through Nebius, giving the U.S. startup more computing capacity to train and deploy its open models.
The agreement follows a similar compute deal with SpaceX signed weeks ago, underscoring how AI developers are locking in scarce infrastructure through large partnerships.
Reflection, founded in 2024 by two former Google DeepMind researchers, is valued at $8 billion and has raised nearly $2.6 billion from backers including Nvidia, Sequoia and Lightspeed.
Interest in open-weight models has risen as Chinese rivals improve and U.S. policy pressure grows; last month the Trump administration pushed Anthropic and OpenAI to restrict their most powerful new models.
Nebius is emerging as a major AI infrastructure supplier after a $2 billion Nvidia investment, with separate long-term deals worth up to $27 billion with Meta and $19.4 billion with Microsoft.
As open-source models from rivals proliferate, can U.S. compute controls alone secure long-term AI leadership?
With billions spent before shipping a product, what is Reflection AI's path to profitability in the open-source world?
Can 'open' AI truly thrive when dependent on billion-dollar deals for scarce, proprietary chips?
Reflection AI Secures $1B+ Nebius Deal: The New Frontline in the Global AI Compute Arms Race
Overview
Reflection AI has secured a major multi-year, $1 billion+ deal with Nebius to access Nvidia-powered AI computing infrastructure through 2029, highlighting the fierce competition for high-performance compute in the AI industry. This agreement ensures Reflection AI can train advanced large language models and develop competitive open-source AI, a growing trend as open models offer more flexibility and lower costs. The deal underscores how vital long-term GPU access is for AI innovation, as training cutting-edge models requires immense computing power. In the current AI arms race, securing reliable compute resources is as crucial as developing new algorithms.