Nvidia Is in Early Talks to Buy or Deepen Its Stake in Open-Model Startup Reflection AI

Nvidia already owns a piece of Reflection AI. It is now talking about owning more of it, or the company. The Financial Times reported that Nvidia is in talks to acquire the open-weight model startup or to invest further, citing people with direct knowledge. Bloomberg, following up, said the options on the table include boosting the existing stake or supplying more computing power, and that the discussions are early, with no terms reached. Reuters added an acqui-hire structure: Nvidia hires the staff and licenses the technology rather than buying the corporate shell, a shape that can shorten a regulatory review. A deal could come together in weeks. It can also fall apart. Reflection declined to comment. Nvidia did not immediately respond. Reflection was founded in 2024 by Misha Laskin and Ioannis Antonoglou, both former Google DeepMind researchers. It builds tools that automate software development and open-weight models; its Beam model has been described as matching advanced reasoning benchmarks of China’s GLM-5.2. Nvidia has already invested about $800 million. Other backers include Disruptive, Sequoia Capital, and 1789 Capital. In March the company was valued around $25 billion. In April, Laskin was reported to be seeking new money at that pre-money mark. An acquisition at or above that figure would be one of the larger AI deals of the year. An acqui-hire would be a smaller headline and a larger talent transfer. Why Nvidia Would Want an Open-Model Lab Nvidia sells the chips. An open-weight lab is a customer, a showcase, and, in the current politics of the market, a counter to Chinese open models that Washington would rather not see become the default. The Trump administration has been described as hoping a U.S. open-model company can rival cheaper Chinese alternatives such as DeepSeek. A Nvidia-backed Reflection that stays independent serves that story. A Nvidia-owned Reflection serves it more directly, and it keeps the weights and the researchers inside the company that already supplies most of the training hardware. The acqui-hire option is the tell that lawyers have been in the room. A full acquisition of a $25 billion startup draws a review. Hiring the team and licensing the models can be structured to avoid one, or to face a narrower one. It is also the structure Nvidia and others have used when the asset is the people and the code rather than the contracts. Reflection’s product is models and developer tools. Those travel with the staff. More chips, the third option Bloomberg named, is the cheapest version of the same relationship. Reflection needs compute. Nvidia has it. A larger supply agreement deepens dependence without a change of control. For a startup that was shopping a $25 billion round, a chip deal is not the outcome the investors priced. For Nvidia, it is a way to stay close if the acquisition price or the review looks wrong. What Is Not Agreed No term sheet has been reported. The $800 million already in is a stake, not a commitment to buy the rest. A $25 billion valuation from a funding process is not a purchase price. Open-weight models are also a regulatory object in a way a chip-optimization startup is not: export rules, open-source policy, and the question of whether a dominant hardware vendor owning a leading open model changes the market for everyone else’s models. Those questions are why the acqui-hire is on the menu. Reflection’s own comment is silence. Nvidia’s is silence. The FT’s sources are the record. Early-stage talks that “could still fall apart” are the majority of stories in this category. Amazon’s talks for Decart, Anthropic’s exit from those same talks, and Nvidia’s earlier look at Decart are the pattern: several buyers, one startup, a price that moves, and a close that is not guaranteed. What to Watch Whether the structure that emerges is a purchase, an acqui-hire, or more equity and chips. Any price against the $25 billion March mark. A regulatory filing, which a full deal would eventually require and an acqui-hire might not. Reflection’s next model, and whether it is trained on Nvidia hardware under a new agreement before any sale. How Chinese open-model labs are positioned in the same Washington conversation. Nvidia is in early talks to buy Reflection AI, hire it, or simply fund and feed it more chips. The startup makes open-weight models. Nvidia already put in about $800 million. Nothing is signed. The reason the talks exist is that open models have become a strategic object, and Nvidia would rather they be its object. Digital8Hub will update if a structure or a price is confirmed.