- Published on
Nvidia in talks to acquire Reflection AI, has invested $800 million; latter valued at $25 billion
- Authors
- Name
- DeepHorus
Nvidia is in preliminary talks with AI startup Reflection AI about a potential acquisition or expanded investment. People familiar with the matter said the two sides could reach an agreement within the next few weeks, but the form of the deal has not yet been determined. Nvidia had previously invested $800 million in Reflection AI and is one of its largest shareholders. A deal could include a full acquisition, additional equity investment, the provision of chips and computing power support, or hiring employees and obtaining a technology license.
Reflection AI was founded in 2024 by Misha Laskin and Yannis Antonoglou, both of whom previously participated in the development of Google's Gemini model. The company focuses on open-weight AI models and automated software development. At its financing in March this year, its valuation reached $25 billion. This week, the company released its first customizable AI model for the public, mainly serving programming and AI agent applications, and it runs on Nvidia chips.
Reflection AI said the new model's performance is comparable to Zhipu's GLM-5.2. Open-weight models allow users to customize them and run them on their own hardware. Compared with relying entirely on commercial cloud APIs, they offer advantages in deployment cost, data control, and flexibility. The U.S. government views them as an important technological resource for countering low-cost competitors such as DeepSeek.
The talks are still at an early stage, and the final form is undetermined. One possibility is a “talent acquisition,” in which Nvidia hires core employees and obtains a technology license without completing a traditional acquisition. This approach can reduce antitrust review and regulatory procedures. Last December, Nvidia reached a deal with Groq worth $20 billion under a similar arrangement, but that deal was sued over this week by some Groq employees.
In addition to a full acquisition or talent acquisition, Nvidia could also increase its stake or provide more chips and computing resources. Other investors include Sequoia Capital, Disruptive, and 1789 Capital; 1789 Capital's partners include Donald Trump Jr. Nvidia already has the Nemotron series of open-weight models, but it does not yet have a frontier-level foundation model of its own. In July this year, Jensen Huang called on the United States to build a domestic open-weight AI ecosystem. In September this year, Nvidia agreed to acquire Hugging Face for $13 billion.
Reflection AI has established cooperation with the U.S. Department of Defense and the Department of Energy, and has reached agreements to develop AI models for U.S. allies, including South Korea. The U.S. government hopes to support domestic open-weight model developers and reduce dependence on overseas low-cost AI technology. This background of government cooperation makes it a strategically significant company in the U.S. AI industry.
If it acquires Reflection AI, Nvidia will gain its model R&D team, technological achievements, and commercial partnerships, strengthening its competitiveness in the open-weight model market. Nvidia has also invested tens of billions of dollars in OpenAI, Nebius, and several startups to drive demand for computing infrastructure. If the two sides reach an agreement, Reflection AI's technology, team, and government partnerships will become important assets for Nvidia's expansion of its AI software business.
