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

How It Unfolded

Reuters reported on October 10, citing the Financial Times: Nvidia is in talks with open-source AI startup Reflection AI, with options on the table including additional investment or an outright acquisition. The report cites multiple people familiar with the matter.

The talks are at an early stage. Whether a deal happens, and in what form, remains undecided. Reflection AI declined to comment; Nvidia had not responded to Reuters’ request for comment as of press time.

Three Deal Structures

Per the FT, the two sides are discussing three transaction forms: additional investment; a full acquisition; or an “acqui-hire” (Nvidia poaching the core team while licensing the technology).

The third is the one to watch. An acqui-hire buys the people and licenses the tech, sidestepping lengthy antitrust review. Nvidia used the same playbook late last year to land AI chip company Groq for $20 billion — the largest deal in Nvidia’s history.

Sources say a deal could come together in the coming weeks. But the report stresses: the talks could still fall apart.

Company Background

Reflection AI was founded in March 2024 by two former Google DeepMind researchers: Misha Laskin and Ioannis Antonoglou. The company’s pitch is building “open-weight frontier models,” hoping the open route can take on low-cost Chinese models like DeepSeek and Kimi.

In April this year, CEO Laskin told CNBC the company was raising at a $25 billion pre-money valuation. In October 2025, Nvidia invested $800 million in Reflection AI as part of a $2 billion funding round. In barely a year, the valuation has nearly tripled.

On October 5, Reflection AI released its first open-weight model, Beam: a 501-billion-parameter mixture-of-experts model with 23 billion activated parameters, trained on 23.8 trillion tokens, with a 1-million-token context length. The company claims Beam’s reasoning benchmarks match Zhipu’s GLM-5.2 while using only a third to a quarter of the inference compute — the company’s own numbers, not yet third-party verified. Full weights and the technical report are planned for release under the Apache 2.0 license later this month.

The Buyer’s Logic

The buyer’s logic is straightforward: Nvidia’s business is built on other people training models — the bigger and more numerous the models, the more GPUs sold. Owning an open-source lab is like gaining a large customer that consumes compute continuously, and it keeps the “open-source ecosystem” compass in Nvidia’s own hands.

The other context is policy. The Trump administration has openly backed cultivating American open-source models to compete with China. Acquiring a homegrown U.S. open-source lab fits the policy winds — and comes with the “Western DeepSeek” branding.

What to Watch Next

First, whether the talks land: a deal could come within weeks, per reports. Second, the weight promise: Beam’s Apache 2.0 full weights are due this month — whether acquisition talks affect the open-source cadence. Third, the price: whether the $25 billion valuation climbs further on the back of these talks.