Nvidia is reportedly in talks to acquire open-source artificial intelligence start-up Reflection AI or increase its existing investment in the company as it seeks to strengthen its position in open-weight AI models and compete with lower-cost Chinese rivals.
According to the *Financial Times* (FT), discussions remain at an early stage and could result in a full acquisition, additional equity investment or a deal involving chips and computing capacity. An acqui-hire arrangement, under which Nvidia would recruit Reflection AI’s staff and license its technology, is also being considered.
A deal could be reached in the coming weeks, although the terms under discussion remain unclear.
Nvidia has already invested $800 million in Reflection AI, which was valued at $25 billion during a funding round in March, the report said.
Founded in 2024 by former Google DeepMind researchers, Reflection AI develops artificial intelligence models and tools for software development. The start-up launched its first open-weight model, Beam, last week, targeting coding and agentic tasks as it seeks to compete with Chinese models such as DeepSeek and Kimi.
Open-weight AI models allow businesses and developers to customise systems and run them on their own infrastructure, potentially reducing costs compared with closed AI services such as OpenAI’s ChatGPT and Anthropic’s Claude.
Although Nvidia has its own Nemotron family of open-weight models, the FT reported that the company lacks a frontier-scale model of its own. A potential deal with Reflection AI could therefore strengthen its position in a market where open-weight systems are emerging as an alternative to closed AI platforms.
Nvidia chief executive Jensen Huang has previously advocated for a domestic open-weight AI ecosystem in the United States. In an essay published in July, he argued that such an ecosystem was essential for creating opportunities for innovation and prosperity across the country.
“In fact, openness may be one of the most important paths to AI safety and security,” Huang wrote, adding that “relying solely on closed models is not inherently safe” because they “can be breached, misused, or fail in ways that outsiders cannot detect”.
Reflection AI has also secured substantial computing capacity through an agreement with SpaceX. Struck in June, the deal gives the start-up access to Nvidia’s GB300 chips at SpaceX’s Colossus 2 facility in Memphis, Tennessee.
The arrangement is valued at $150 million per month and runs through 2029.
