Nvidia Is in Talks to Buy Reflection AI. The US Open-Weights Champion May Become a Chip Company's Lab.
The Financial Times reported Friday that Nvidia is in talks to acquire Reflection AI, the US developer of open-weight models, or to deepen its investment in the company. Reuters, Bloomberg and Benzinga all picked it up within hours. The talks are early. One structure on the table is an acqui-hire, in which Nvidia hires Reflection's staff and licenses the technology rather than buying the company outright. The FT says an agreement could be reached within weeks, and that it could also fall apart.
Reflection is the company we covered on October 2 for Beam, and it has been the designated American answer to DeepSeek for a year. Nvidia led its $2 billion round in October 2025 with a reported $800 million of its own money at an $8 billion valuation, and Reflection has been raising this year at a reported valuation around $25 billion, with JPMorgan's security and resiliency initiative named as a possible participant. It has signed more than a billion dollars of compute with Nebius and a reported GB300 arrangement through SpaceX. Nvidia also shows up on the cap table of Nous Research, whose Series B closed Tuesday, so the chipmaker now has a position in both of the open-weight labs that matter in the US.
What makes this an agent story rather than a semiconductor story is what Reflection actually sells. Its pitch since the start has been open weights plus the training and RL infrastructure to turn them into autonomous coding and research agents, which is exactly the layer where Nvidia has been building its own software bets, from NeMo-DCR last week to the Sentry monitor two weeks ago. If Nvidia absorbs Reflection, the company that sells the chips ends up owning a frontier open-weight lab, and the US open-model ecosystem's biggest independent becomes a division.
The uncomfortable reading is the one the Trump administration will have to answer: it has leaned on Reflection as the sovereign open-source rival to Chinese models, and an acqui-hire would be the third time this year that a talent-and-license deal replaced an acquisition that regulators would have had to review. Manus went through a version of that with Meta. Nothing is signed. But the fact that a $25 billion open-weight lab is even discussing being folded into its GPU supplier tells you how the economics of training frontier models look to the people inside them.
FT (via Reuters): https://www.investing.com/news/stock-market-news/nvidia-in-talks-to-invest-further-in-reflection-ai-or-buy-it-ft-reports-4942221
Benzinga: https://www.benzinga.com/m-a/26/10/62288564/report-nvidia-eyes-reflection-ai-takeover-as-the-ai-arms-race-heats-up
← Back to all articles
Reflection is the company we covered on October 2 for Beam, and it has been the designated American answer to DeepSeek for a year. Nvidia led its $2 billion round in October 2025 with a reported $800 million of its own money at an $8 billion valuation, and Reflection has been raising this year at a reported valuation around $25 billion, with JPMorgan's security and resiliency initiative named as a possible participant. It has signed more than a billion dollars of compute with Nebius and a reported GB300 arrangement through SpaceX. Nvidia also shows up on the cap table of Nous Research, whose Series B closed Tuesday, so the chipmaker now has a position in both of the open-weight labs that matter in the US.
What makes this an agent story rather than a semiconductor story is what Reflection actually sells. Its pitch since the start has been open weights plus the training and RL infrastructure to turn them into autonomous coding and research agents, which is exactly the layer where Nvidia has been building its own software bets, from NeMo-DCR last week to the Sentry monitor two weeks ago. If Nvidia absorbs Reflection, the company that sells the chips ends up owning a frontier open-weight lab, and the US open-model ecosystem's biggest independent becomes a division.
The uncomfortable reading is the one the Trump administration will have to answer: it has leaned on Reflection as the sovereign open-source rival to Chinese models, and an acqui-hire would be the third time this year that a talent-and-license deal replaced an acquisition that regulators would have had to review. Manus went through a version of that with Meta. Nothing is signed. But the fact that a $25 billion open-weight lab is even discussing being folded into its GPU supplier tells you how the economics of training frontier models look to the people inside them.
FT (via Reuters): https://www.investing.com/news/stock-market-news/nvidia-in-talks-to-invest-further-in-reflection-ai-or-buy-it-ft-reports-4942221
Benzinga: https://www.benzinga.com/m-a/26/10/62288564/report-nvidia-eyes-reflection-ai-takeover-as-the-ai-arms-race-heats-up
Comments