Nvidia is in early talks to acquire Reflection AI, hire its team or increase its investment in the open-weight model developer, according to the Financial Times. People familiar with the matter said an agreement could come within weeks, though the talks could still collapse. The report matters because Reflection is already one of Nvidia's largest model bets, with $800 million invested and a $25 billion valuation.
Possible deal structures and Reflection background
Nvidia has three paths under discussion, the paper reported. It could buy Reflection outright, increase its stake possibly alongside a bigger supply of chips and computing power, or complete an acquihire. Under an acquihire, Nvidia would hire Reflection's staff and license its technology without taking control of the company. That structure would likely avoid a lengthy regulatory review.
Nvidia used both acquisition models in the past year. In December, inference chip startup Groq licensed its technology to Nvidia in a deal reported at $20 billion, with founder Jonathan Ross and other senior executives moving to the chipmaker while Groq kept running its cloud business. On Sept. 3, Nvidia agreed to pay $12.9 billion for model hosting platform Hugging Face in a straight purchase. The Reflection talks follow that pattern of pairing capital with talent and infrastructure.
Reflection was founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. Nvidia led its $2 billion round in October 2025 and has invested $800 million in total, according to the report. Laskin told CNBC in April that the latest round closed at a $25 billion pre-money valuation. Other backers include Sequoia Capital Operations, Lightspeed Venture Partners and DST Global. Neither Nvidia nor Reflection has commented publicly on the talks.
What a deal would mean for enterprise AI buyers
The report came five days after Reflection unveiled its first open-weight model, Beam, a mixture-of-experts system with 501 billion parameters. Coding agents were the company's first focus, and it positions Beam against lower-cost Chinese open models such as DeepSeek and Kimi on coding and agentic work. For companies building software automation, a closer Nvidia tie could mean better optimization of such models for Nvidia hardware and faster availability through established distribution channels.
Compute commitments shape the commercial risk around Beam. Training ran on Nvidia GB300 systems that Reflection rents from Space Exploration Technologies Corp. under a June agreement worth up to $6.3 billion. Payments of $150 million a month cover capacity at SpaceX's Colossus 2 data center in Memphis through 2029, while Nebius Group agreed in July to sell Reflection another $1 billion in AI capacity. A small buyer gains little leverage over those costs, while a large enterprise can negotiate reserved capacity and clearer service terms.
The marker to watch is whether any agreement appears within weeks and in what form. An outright purchase would signal Nvidia's intent to control an open-weight coding model stack, while a larger investment plus chips would point to capacity-led partnership. An acquihire with a license would leave Reflection independent but move expertise inside Nvidia. The chosen structure will define pricing, access and support conditions for business users of Beam-based agents.
