Nvidia in early talks to acquire or deepen investment in open-model startup Reflection AI

Nvidia is in talks to acquire or deepen its investment in Reflection AI, a US startup that builds open-weight AI models.

Open-weight models are AI models whose inner settings are public, so users can customise them and run them on their own hardware. They can be far cheaper for businesses to operate than the most advanced versions of OpenAI's ChatGPT or Anthropic's Claude. President Donald Trump's administration hopes Reflection will rival cheap Chinese alternatives such as DeepSeek.

The talks are at an early stage, and a deal could take several shapes. One option is an acqui-hire, where Nvidia would hire staff and license technology without the long regulatory review that comes with a full acquisition. Nvidia could also put more money into Reflection, or provide it with more chips and computing power.

Nvidia is already one of Reflection's largest shareholders and has invested $800mn in the company. Other backers include Sequoia Capital, Disruptive and 1789 Capital, where Donald Trump Jr is a partner. Reflection was last valued at $25bn in a March funding round. The people familiar with the talks could not establish the terms under discussion.

A deal could come in the coming weeks, according to multiple people with knowledge of the matter. They cautioned that the discussions could fall apart.

Acqui-hires have become more popular among Big Tech companies that want to avoid antitrust scrutiny of AI deals. Nvidia used a similar format with chip designer Groq in a $20bn deal in December. Former Groq employees who felt left behind have since sued, with the lawsuit coming this week.

Nvidia is the world's most valuable company, and it has been on an investment spree for the past two years to strengthen the AI economy and its own place in it. It already has an open-weight family of models called Nemotron, but it lacks a frontier-scale model of its own. Frontier-scale means among the largest and most capable models available.

In recent months Nvidia has focused more on building a market for open-weight models. In July, chief executive Jensen Huang called for a domestic open-weight ecosystem, which he said was essential for creating opportunities for innovation and prosperity across the country. In September, the $5.5tn company agreed to acquire Hugging Face, a repository for millions of models and data sets, for $13bn.

Nvidia has also put tens of billions of dollars into large AI labs such as OpenAI, into neocloud businesses such as Nebius, and into a string of smaller groups. That in turn creates more demand for its chips, which remain the key commodity for companies building or running AI models.

Reflection is seen as strategically important for the US in its AI race with Chinese rivals such as DeepSeek. This week it unveiled its first public, customisable AI product, which runs on Nvidia chips. Reflection claims the model is on a par with Z.ai's GLM-5.2, a leading Chinese competitor. That claim comes from the company.

Reflection's founders, Misha Laskin and Ioannis Antonoglou, previously worked on Google's AI model Gemini.