Reflection announced Tuesday that it inked a deal to buy $1 billion worth of compute from neocloud Nebius to train its open source models. The deal comes on the heels of the startup’s agreement with SpaceX to pay $150 million each month through 2029 for compute capacity from Elon Musk’s Colossus data center.

That’s a lot of compute for a company that is yet to launch a model publicly. But cofounder and CTO Ioannis Antonoglou says the company plans to finally release its open source model later this year.

Reflection started training its models last October and was primarily focused on research before that, Antonoglou says. Cofounded in March 2024, the Brooklyn-based startup initially started out building AI coding agents that improved using reinforcement learning — a technique where AI systems learn through trial and error and rewards and penalties. And it saw a massive opportunity in building open source models. Along with sourcing compute, the startup has been heavily focused on recruiting, growing its team from just 30 last September to 230 employees today. “We are flying the plane as we’re building it,” he says.

Currently, the most powerful open source models are being developed by Chinese companies. Because these models are magnitudes cheaper, U.S. startups are increasingly using them to reduce costs. That traction has fueled the rise of breakout AI companies in China. Chinese AI pioneer DeepSeek, for example, is in talks to raise $1.5 billion at a $71 billion valuation and is preparing to go public next year, Bloomberg reported. Moonshot AI, a Chinese AI model developer, is in talks to raise $2 billion at a $20 billion valuation, Forbes........

© Forbes