The Story, Not the Spin
Deniro News

The Story, Not the Spin

Nvidia in Talks to Acquire Reflection AI, the Open-Weight Lab It Backed

The chipmaker is weighing a deeper investment or an outright purchase of the open-weight startup, the Financial Times reported. The talks are early and could still fall apart.

A cracked-open silicon microchip releasing streams of code and neural-network nodes toward an interlocking corporate block
Illustration: Deniro News

Nvidia is in talks to buy Reflection AI or deepen its investment in the open-weight startup, the Financial Times reported Saturday. The chipmaker already backs Reflection with an $800 million investment, and the discussions now cover a range of options — a full acquisition, a bigger equity stake, or a deal to license the startup's technology and hire its researchers, a structure known as an acqui-hire.

The acqui-hire route matters because it can sidestep a formal merger filing and the lengthy regulatory review that comes with one. The FT, citing people with direct knowledge of the matter, said an agreement could be reached in the coming weeks — but cautioned that the talks could still fall apart. Reflection declined to comment on the report, and Nvidia did not immediately respond.

Reflection was founded in 2024 by Misha Laskin and Ioannis Antonoglou, former Google DeepMind researchers, with a pitch to build frontier-grade open-weight models in the United States. Last Monday the company released Beam, a 501-billion-parameter model that activates 23 billion parameters for any given task. Reflection said Beam matches Z.ai's GLM-5.2 on reasoning benchmarks while using three to four times less inference compute — claims that have not been independently verified. The full weights and a technical report are due under an Apache 2.0 license later this month.

The startup is one of the most valuable private AI labs in the country. Laskin said in April that Reflection was raising fresh capital at a $25 billion pre-money valuation, nearly triple the $8 billion mark it held in October 2025, when Nvidia first wrote its check. Open-weight models are the other lane in the AI race — their parameters are published, so developers can inspect, modify, and run them — and that lane has been dominated by cheaper Chinese systems like DeepSeek and Moonshot's Kimi.

A deal would extend Nvidia's reach past the chips every major lab buys from it. Acquiring Reflection, or simply owning more of it, would give the company a stake in the models themselves, not just the silicon underneath. The acqui-hire structure has drawn regulatory attention before: Microsoft's 2024 hiring of most of Inflection AI's team prompted scrutiny from the FTC, a warning that the arrangement does not always dodge oversight.

More from Deniro News