Nvidia in Early Talks to Buy Open-Model Startup Reflection AI, FT Reports

Nvidia in Early Talks to Buy Open-Model Startup Reflection AI, FT Reports

Nvidia is in talks to acquire U.S. open-model startup Reflection AI — or to deepen its already sizable investment — the Financial Times reported on Saturday, October 10, citing people with direct knowledge of the matter. The discussions are at an early stage, and the outcome is far from settled.

The Deal Could Take Several Shapes

According to the FT, a transaction could come as a full acquisition, an acqui-hire in which Nvidia hires Reflection's staff and licenses its technology rather than buying the company outright — a structure that can sidestep lengthy regulatory review — or simply a larger equity investment. Nvidia could also deepen the relationship through a deal for more chips and computing capacity. Multiple sources told the FT that an agreement could be reached in the coming weeks, while cautioning that the talks could still fall apart. Reflection declined to comment; Nvidia did not immediately respond.

Nvidia Is Already the Startup's Biggest Backer

The potential acquirer is no stranger: Nvidia is one of Reflection's largest shareholders, having invested $800 million in the startup, the FT reported. Reflection was founded in 2024 by former DeepMind researchers Misha Laskin, now chief executive, and Ioannis Antonoglou. Laskin led reward modeling for Gemini at DeepMind; Antonoglou co-created AlphaGo and AlphaZero and ran RLHF for Gemini. The company originally pitched itself as a builder of autonomous coding agents and later widened its ambition into an "open frontier AI lab." In April, Laskin told CNBC the startup was raising fresh capital at a pre-money valuation of $25 billion.

Five Days After Beam: A 501-Billion-Parameter Answer to China

The acquisition talks arrive just five days after Reflection unveiled Beam, its first open-weight model, on October 5. Beam is a sparse mixture-of-experts model with 501 billion total parameters, only 23 billion of them active at inference — a design aimed at faster, cheaper runs. Reflection says the model is competitive with Z.ai's GLM-5.2 and is closing in on Alibaba's Qwen 3.8-Max on coding and agentic tasks. Final red-teaming and evaluations are still underway; the weights will be released under an Apache 2.0 license later this month, with early-access signups open now. Laskin's framing: companies and governments want advanced AI they can run and control themselves, "and today's strongest open models largely come from Chinese labs."

Filling the "Western Open-Weight Void"

The strategic context is Washington's discomfort with an open-weight leaderboard dominated by DeepSeek, Alibaba's Qwen, and Moonshot's Kimi, while OpenAI, Anthropic, and Google keep their best models closed and Meta has stepped back from frontier open releases. Reflection carries national-security weight: it has partnerships with the Pentagon and the Department of Energy and has signed agreements to develop AI models for U.S. allies including South Korea. Nvidia chief executive Jensen Huang called in July for a domestic open-weight ecosystem "essential for innovation and prosperity," and in September Nvidia agreed to acquire Hugging Face for $13 billion. Reflection would be another piece on the same board.

Who Feels It: Enterprises and Governments That Want Their Own Models

For enterprise and public-sector buyers, open weights mean downloading a frontier model, running it on their own infrastructure, fine-tuning it freely — no data leaving the building, no API toll booth. A Nvidia–Reflection combination would tighten the bundle of chips, compute, and open models, and make the American open-model camp materially stronger.

What to Watch

Two things: whether Beam's weights and full technical details ship this month as promised, letting independent labs verify the company's efficiency claims — self-reported benchmark tables deserve a skeptical read until outsiders run the same evals; and whether the Nvidia talks harden into a deal in the coming weeks, and what form it takes.