Nvidia is in talks to acquire Reflection AI or deepen its investment in the US startup, the Financial Times reported on Saturday, October 10, 2026, citing people with direct knowledge of the matter. The news comes just five days after Reflection unveiled Beam, its first frontier open-weight model, and confirms the interest of the largest maker of artificial intelligence chips in securing a role in the segment of models that anyone can download and modify.
According to the British newspaper, the talks are at an early stage and a deal could take several forms. One option is an outright purchase of the company. Another is a so-called acqui-hire, in which Nvidia would hire the staff and license the technology without formally absorbing the company, an approach that could avoid a lengthy regulatory review. An increase in Nvidia’s equity stake or a commitment to supply more chips and computing capacity are also on the table.
The FT said an agreement could be reached in the coming weeks, while warning that the discussions could still fall apart. Reuters, which picked up the report, said it could not independently verify it and that neither Nvidia nor Reflection immediately responded to requests for comment. Bloomberg, for its part, cited a person familiar with the matter who confirmed that talks are taking place, that they are at an early stage and that no terms have been agreed. Reflection declined to comment to that agency.
The relationship between the two companies is not new. Nvidia is already one of Reflection’s main financial backers and strategic partners, with an investment of $800 million, according to the Financial Times. The startup was founded in 2024 by Misha Laskin and Ioannis Antonoglou, two former Google DeepMind researchers, and specializes in tools that automate software development, one of the fastest-growing uses of artificial intelligence in business.
In April, Laskin, who serves as chief executive, told CNBC that the company was raising fresh capital at a pre-money valuation of $25 billion. The Wall Street Journal had previously reported that Reflection was in talks to raise $2.5 billion at that valuation. The company has also struck billion-dollar deals with Elon Musk’s SpaceX and with cloud provider Nebius to secure the computing capacity it needs to train its models.
The immediate context is Beam. On October 5, Reflection introduced that mixture-of-experts model with 501 billion total parameters, aimed at coding, reasoning and agent tasks. The company says Beam is designed to perform well for business customers and developers at a lower cost than many competitors, and that it scores comparably on advanced reasoning benchmarks to GLM-5.2, the flagship open model from Chinese company Z.ai.
That last point explains much of the strategic interest. Chinese open-weight models, such as those from DeepSeek, Qwen and Kimi, have become a very cheap option for startups and companies around the world, without trailing too far behind the closed systems of OpenAI or Anthropic in capability. In the United States, concern has grown that this price advantage could turn Chinese models into the de facto standard for many applications.
The US government has made that concern explicit. The artificial intelligence action plan the White House published in July 2025 states that the country needs to “ensure America has leading open models founded on American values,” and adds that open-source and open-weight models could become global standards in some areas of business and academic research. According to the FT, President Donald Trump’s administration hopes Reflection can rival low-cost Chinese alternatives.
For Nvidia, the deal would fit into a very active streak of transactions. Bloomberg notes that the company, the world’s most valuable, has used its enormous financial power to cultivate the artificial intelligence ecosystem in recent months, with deals such as a $13 billion agreement to buy model platform Hugging Face in September and a licensing pact worth roughly $20 billion with chip startup Groq.
Its chief executive, Jensen Huang, is also one of the most visible advocates of open-weight models. The business logic is clear: the more powerful models run on Nvidia hardware, and the more companies adapt and deploy them in their own data centers, the greater the demand for its processors. A competitive American developer of open models strengthens that cycle and reduces the appeal of alternatives optimized for other chips.
The possible acqui-hire structure is also telling. In recent years, several large technology companies have turned to deals in which they hire a startup’s founders and team and license its technology instead of buying it outright. These arrangements allow talent to be absorbed quickly, but they have drawn the attention of competition regulators, who question whether they are used to sidestep merger control. A deal of this kind involving Nvidia would be closely scrutinized in Washington and Brussels.
Many questions remain open. There is no indication of the price a purchase might reach, nor of whether Reflection would keep its strategy of publishing model weights under permissive licenses if it came under Nvidia’s control. It is also unclear how a deal would affect its computing agreements with SpaceX and Nebius. The developer community that has started testing Beam will be watching for any sign of a change in that openness policy.
For now, what exists is a press report about preliminary negotiations, with no official confirmation from either party. But the move shows where competition is heading: the race is no longer fought only among labs selling access to closed models, but also over control of the open-model ecosystem on which thousands of companies will build their products. If the deal materializes, Nvidia would go from chip supplier to direct player in that battle.
