Nvidia in talks to acquire or invest more in Reflection AI

Nvidia is in early talks to acquire or further invest in Reflection AI, the open-weight model startup the chipmaker has already backed with $800 million, the Financial Times reported on October 10, 2026.

Reuters, which relayed the Financial Times report, said it could not independently verify the talks. It also reported that neither Nvidia nor Reflection AI responded to requests for comment outside regular business hours.

Reflection AI builds open-weight AI models, which publish their parameters so organizations can inspect, fine-tune and run them on their own hardware. The company was founded in 2024 by Misha Laskin and Ioannis Antonoglou, both former Google DeepMind researchers. Laskin is chief executive.

The talks are at an early stage and could still collapse, according to the report. A deal could take several forms, including a full acquisition of Reflection AI, a larger investment, or an acqui-hire in which Nvidia would hire Reflection's staff and license its technology rather than buy the company outright. The report said an acqui-hire structure could avoid a lengthy regulatory review, and that an agreement could be reached within weeks.

Nvidia is already a major financial backer and strategic investor in Reflection AI, having invested $800 million in the startup, the Financial Times reported. Laskin told CNBC in April that the company was raising fresh capital at a $25 billion pre-money valuation, Reuters reported.

The talks follow Reflection AI's release of Beam, its first open-weight model, on October 5, 2026. Reflection said Beam is a sparse mixture-of-experts model with 501 billion total parameters and 23 billion active, built for coding, reasoning and agentic workloads. The company said it pretrained Beam on 23.8 trillion tokens and ran more than 100 million reinforcement-learning rollouts on 10,500 Nvidia GB300 chips over four weeks. Reflection said it will release Beam's weights, a technical report and a model card later in October. Reuters reported that Beam is aimed at competing in coding and agentic tasks with lower-cost Chinese models such as DeepSeek and Kimi.

Nvidia has bought access to AI talent in this shape before. In August 2026 it agreed to pay Poolside $6 billion to license the startup's model-building software and extended job offers to more than 100 of its engineers, alongside a $1 billion investment, Quartz reported. In December 2025 it paid Groq about $20 billion for its inference technology and top engineers, CNBC reported, and Groq stayed independent. Across three such deals, Nvidia has committed roughly $27 billion, The Next Web reported, each structured as a license and a hiring round rather than a purchase, which avoids the antitrust review a buyout can invite.

For people who build on open-weight models, a closer tie between Nvidia and Reflection AI could keep Beam's weights openly released, or it could pull the company's future models toward Nvidia's own open-weight Nemotron line.

The next signal is whether the talks produce an agreement, which the report said could come within weeks, and the release of Beam's weights, promised for later in October.