Nvidia in Talks to Buy Reflection AI, Open-Weight Startup

Nvidia is in early-stage talks to deepen its investment in or acquire Reflection AI, the open-weight AI startup it already backed with an $800 million check, the Financial Times reported on Oct. 10. The talks are preliminary. A deal could arrive within weeks or fall apart entirely. Neither company commented.

Reflection, founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou, was last valued at $25 billion in a March 2026 round. On Oct. 5 it released Beam, its first open-weight model, built for coding and agentic tasks and positioned against cheaper Chinese models like DeepSeek and Kimi.

The structure of any deal matters as much as the price. According to the FT, the options include a full acquisition or an acqui-hire, in which Nvidia would hire Reflection's staff and license its technology. Bloomberg confirmed that an acqui-hire is under consideration, partly because it could dodge the long antitrust review a full takeover would invite. Nvidia used the same playbook late last year, hiring Groq's team and licensing its technology in a deal valued at $20 billion, per the Seoul Economic Daily.

Why it matters

Nvidia has more money and more compute than almost anyone on the planet. It does not need Reflection's servers. What it cannot manufacture is a small, dense team that ships frontier open models. Open-weights models have become the most contested lane in AI: they let developers and governments run capable models without renting access from a closed lab, and they have become a geopolitical issue as Chinese labs like DeepSeek and Kimi undercut Western prices.

Reflection's Beam is built for exactly the workloads Nvidia's chips sell into, coding and agents. If Nvidia owns the team building the best open models, it owns a lever over how those models are optimized, and on whose silicon they run best.

The context

The math is what makes this moment striking. Nvidia led a $2 billion raise for Reflection at an $8 billion valuation in October 2025, writing an $800 million check itself. Five months later the company was worth $25 billion. Renting a startup's future with a minority stake gets expensive when the valuation triples in a year. At some point, the rent-vs-own calculation flips toward ownership.

The acqui-hire option is equally telling. Regulators in the U.S. and Europe have made large AI acquisitions slow and expensive to review. Hiring the people and licensing the IP reaches nearly the same destination with less scrutiny. Whether that is clever dealcraft or a dodge depends on your view of antitrust law, but the Groq precedent means Nvidia has already shown it works.

There is also the political layer. Washington has been pushing for strong American open models as a counterweight to Chinese labs. Nvidia's CEO has made that argument publicly. Buying Reflection would give that argument a flagship product.

Key numbers

• $800 million: Nvidia's investment in Reflection AI in October 2025
• $25 billion: Reflection's valuation in its March 2026 round
• $20 billion: value of Nvidia's Groq acqui-hire late last year, per Seoul Economic Daily
• Oct. 5, 2026: date Reflection released its Beam model
• 2024: year Reflection was founded

What happens next

The talks are early, and early talks die more often than they close. If Nvidia proceeds, watch which structure it picks. A full acquisition signals confidence it can clear regulators or a willingness to fight them. An acqui-hire signals the opposite, and it would confirm the Groq deal as Nvidia's standard operating procedure for buying AI talent. Either way, Reflection's next model release, and its pricing against DeepSeek and Kimi, will tell the market whether the $25 billion valuation was the start or the peak.