Nvidia in talks to buy Reflection AI, valued at $25bn

The latest:

Nvidia is in early talks to acquire or deepen its investment in Reflection AI, a US developer of open-weight models that the Trump administration hopes can rival cheap Chinese alternatives, people with direct knowledge said. A deal could land within weeks, the people said, though they cautioned the discussions could collapse. Both Nvidia and Reflection declined to comment.

Details:

• The structures on the table: A transaction could take several shapes, the people said, including an acqui-hire in which Nvidia takes on staff and licenses technology without the drawn-out regulatory review that accompanies a full acquisition. Alternatively, Nvidia could make a further equity investment or sign a deal supplying Reflection with more chips and computing power.
• The valuation: Reflection was last valued at $25bn in a March funding round, and Nvidia is already one of its largest shareholders after investing $800mn. The terms under discussion between the two sides could not be established.
• The antitrust workaround: Acqui-hires have grown popular among Big Tech firms seeking to avoid antitrust scrutiny of AI dealmaking. Nvidia used the same format with chip designer Groq in a $20bn transaction in December, a deal that drew a lawsuit this week from Groq employees left out of it.
• The strategic gap: Nvidia, the world’s most valuable company at $5.5tn, already runs an open-weight model family called Nemotron but lacks a frontier-scale model of its own. Buying Reflection would close that gap at the top end of its lineup.
• Huang’s push: In July, chief executive Jensen Huang led an industry call for a domestic open-weight ecosystem, which he said was essential for “creating opportunities for innovation and prosperity across the country.” Open-weight models can be customised and run on a user’s own hardware, often far cheaper to operate than top-tier versions of OpenAI’s ChatGPT or Anthropic’s Claude.
• The buying spree: In September, Nvidia agreed to acquire Hugging Face, a repository holding millions of models and data sets, for $13bn. It has also poured tens of billions into large AI labs including OpenAI, neocloud operators such as Nebius, and a string of smaller groups, generating fresh demand for its own chips.
• The China race: Reflection is viewed as a strategically significant US asset against Chinese rivals such as DeepSeek. This week it unveiled its first public, customisable AI product, which runs on Nvidia chips, and claims the model matches Z.ai’s GLM-5.2, a leading Chinese competitor.
• Who is behind it: Founders Misha Laskin and Ioannis Antonoglou previously worked on Google’s Gemini model. Backers include Sequoia Capital, Disruptive and 1789 Capital, where Donald Trump Jr is a partner.
• Government ties: Reflection has partnered with the Pentagon and the Department of Energy, and struck agreements to develop AI models for US allies including South Korea.

Between the lines:

The shape of any deal may matter more than its price. Nvidia’s Groq transaction showed the acqui-hire route can move a company’s talent and technology without triggering a full merger review, and the employee lawsuit filed this week shows the format carries its own costs. Nvidia already holds the chips, the open-model family and the distribution platform; Reflection would supply the frontier model it lacks.

What’s next

Whether talks produce a deal in the coming weeks, and which structure is chosen — outright acquisition, acqui-hire, equity top-up or a compute agreement. Also watch the Groq employee lawsuit, which tests the acqui-hire model Nvidia has used before.