Open-source AI is drawing another mega-deal. Reflection AI, a startup deeply backed by NVIDIA, is negotiating a new funding round of roughly $2.5 billion at a pre-money valuation target of $25 billion — up from the $20 billion figure floated in early March. The company builds open-source frontier models and is seen by capital markets as a strategic U.S. counterweight to the rise of China's DeepSeek.
Valuation Rockets 15x in Six Months, JPMorgan Eyes Entry
Reflection AI's valuation trajectory is staggering. In March 2025, the firm was valued at just $545 million. By October of the same year, after a $2 billion round led by NVIDIA with participation from Sequoia, Lightspeed, and former Google CEO Eric Schmidt, its valuation hit $8 billion — a 15-fold jump. Now, less than half a year later, the target has more than tripled again. Potential investors in this round include existing backers Disruptive and NVIDIA, plus traditional finance heavyweight JPMorgan Chase through its "Security and Resilience Program." If completed as planned (post-money valuation of ~$27.5 billion), it would be one of the largest financing deals in open-source AI history.
60-Person Team with Geopolitical Ambitions
Founded in 2024 and headquartered in New York, Reflection AI was launched by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. The team started with autonomous coding agents before pivoting to open-source frontier AI models. Despite numbering only around 60 people, it has attracted intense capital interest — driven largely by geopolitical dynamics. Multiple international media outlets note that with Chinese open-source models like DeepSeek gaining traction, the U.S. needs a homegrown force to set global open-source AI standards. Reflection AI fits that bill.
NVIDIA's Hardware Ecosystem Play
NVIDIA is not just a financial backer; it supplies chips and computing infrastructure, signaling a deeper strategic bet. Reflection AI is widely viewed as a key piece of NVIDIA's effort to build a global open-source AI network on its hardware. Although the company has yet to release its flagship Mixture-of-Experts model, investors are confident. Its planned "open weights" approach — releasing model weights while retaining control over training data and pipelines — balances openness with business defensibility, a formula that resonates strongly in today's AI capital market.