NVIDIA is in talks to boost its investment in Reflection AI or offer the US startup computing power, a move that would strengthen ties with a key developer of open-weight models, according to a person familiar with the matter.

The discussions are at an early stage, and no terms have been reached, said the person, who asked not to be identified because the matter is private. Nvidia did not immediately respond to a request for comment, while Reflection declined to comment.

The Financial Times previously reported on the talks, saying they could potentially lead to an acquisition offer. Possible deal structures include an “acqui-hire,” in which Nvidia would hire staff and license technology; an agreement to provide the startup with more chips; or an increase in Nvidia’s equity investment, the newspaper said.

The startup debuted its first open-weight model, Beam, earlier in October, pushing into a market where Chinese startups have a strong presence. With such software, developers release the model’s parameters, or weights, allowing others to freely use and modify the technology.

Beam is designed to perform well for business customers and developers on coding and AI agent tasks while costing less than many competitors’ products, Reflection said on Monday (Oct 5). It said Beam scores comparably on advanced reasoning benchmarks to GLM-5.2, a flagship open model from China’s Z.AI.

The startup, which was founded by two former Google DeepMind researchers, has held talks to raise US$2.5 billion at a valuation of US$25 billion, the Wall Street Journal previously reported.

Reflection has also struck billion-dollar deals with Elon Musk’s SpaceX and cloud provider Nebius Group to secure computing power to develop its models.

A Reflection tie-up would add to a frenzy of transactions by Nvidia in recent months. The chipmaker, which is the world’s most valuable company, has used its financial might to cultivate the broader AI industry.

The deals have included a US$13 billion agreement to buy AI model company Hugging Face in September and a roughly US$20 billion licensing pact with the chip startup Groq.

Nvidia chief executive officer Jensen Huang is also a major proponent of open-weight models.

The hope is that US development of the technology could curb startups’ use of Chinese open-weight models, which are much cheaper to use than ones from American companies like OpenAI and Anthropic, while not far behind in intelligence.

The US government is also encouraging the development of such models.

“We need to ensure America has leading open models founded on American values,” reads US President Donald Trump’s July 2025 AI action plan. “Open-source and open-weight models could become global standards in some areas of business and in academic research worldwide.” BLOOMBERG