- Nvidia, which has already invested $800 million in Reflection AI, is discussing either a larger stake or a takeover, according to the Financial Times; neither company commented.
- One option on the table is an acqui-hire, in which Nvidia hires Reflection's researchers and licenses its technology, a structure that could avoid a lengthy antitrust review.
- Reflection was raising at a $25 billion pre-money valuation in April and released Beam, a 501-billion-parameter open-weight model, on October 5.
Nvidia is in early talks to acquire Reflection AI or deepen its stake in the New York lab, a move that would put the largest American open-weight model developer inside the world's most valuable chipmaker five days after that lab shipped its first model. The Financial Times reported the discussions on October 10, citing people with direct knowledge, and said a deal could take shape within weeks or still fall apart.
Nvidia is discussing a takeover or a bigger stake in Reflection AI
Reflection was founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou and has positioned itself as the American answer to DeepSeek, Qwen and Kimi in open-weight AI. Nvidia led its $2 billion round in October 2025 at an $8 billion valuation, putting in $800 million, and Laskin said in April that the company was raising again at $25 billion before new money.
The talks follow the October 5 release of Beam, a mixture-of-experts model with 501 billion total parameters and 23 billion active per token. Reflection's launch post says the reinforcement learning phase ran for four weeks on 10,500 Nvidia GB300 GPUs, which makes Beam one of the largest public showcases of Nvidia's current hardware outside the big closed labs.
| Nvidia's existing investment in Reflection | $800 million, in the October 2025 round Nvidia led |
| Reflection's valuation | $25 billion pre-money in April 2026, up from $8 billion in October 2025 |
| Beam, released October 5, 2026 | 501 billion total parameters, 23 billion active per token |
| Nvidia GPUs used for Beam's reinforcement learning | 10,500 GB300s over four weeks |
A Reflection deal would give Nvidia its own US open-model lab
Nvidia already owns pieces of the open-model supply chain. It signed a $12.93 billion agreement to buy Hugging Face, the platform where most open weights are published, which still awaits US antitrust clearance, and in August it agreed to pay Poolside $6 billion to license its coding models while hiring 109 of its staff. Reflection would add a lab that trains frontier-scale open models, the one layer Nvidia does not yet control.
The structure matters as much as the price. Nvidia used licensing and hiring deals for Groq, Enfabrica and Poolside, each of which left a legally independent company standing and skipped the merger filing a purchase would trigger. The Hugging Face acquisition is the exception and is the one deal still waiting on regulators, which explains why an acqui-hire is on the table for Reflection.
Ownership would also put Nvidia in direct competition with labs that buy its chips, including the Chinese open-model developers Reflection was built to challenge, though export controls already bar Nvidia from selling its most advanced parts to them. Whichever form the deal takes, it would leave the leading American open-weight lab funded by, and possibly folded into, its main hardware supplier. The same week, Ukrainian drones showed how exposed that hardware can be when they hit the Yandex site housing its AI supercomputers.
Nvidia's open-model strategy now runs through companies it owns, licenses or funds, and Reflection would close the circle by giving the chipmaker a lab that trains the models, a platform that distributes them and the hardware they run on.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.