Nvidia is engaged in early-stage talks to acquire or expand its investment in Reflection AI, a US-based start-up developing open-weight artificial intelligence models, according to sources familiar with the discussions. Reflection AI is backed by Nvidia, which is already one of its largest shareholders, as well as venture capital firms including Sequoia Capital and 1789 Capital, where Donald Trump Jr. is a partner.
The potential deal could take several forms, including an "acqui-hire" arrangement. This approach would allow Nvidia to onboard Reflection’s staff and license its technology, potentially avoiding the extensive regulatory scrutiny that can accompany a full acquisition. Other options under consideration include a further equity investment or agreements to supply Reflection with additional chips and computing resources. While multiple people familiar with the matter suggested a deal might be finalized in the coming weeks, they emphasized negotiations remain fluid and could stall. Reflection was last valued at $25 billion during a funding round in March.
Nvidia declined to comment on the discussions, as did Reflection AI.
Acqui-hires have become a preferred structure among large technology companies aiming to navigate antitrust concerns related to AI acquisitions. Nvidia executed a similar strategy in December with chip designer Groq, a transaction valued at $20 billion that has recently led to a lawsuit from former Groq employees.
As the world’s most valuable company, Nvidia has been investing aggressively in the AI sector over the last two years to maintain its leadership in the industry. Despite its existing open-weight AI model lineup branded as Nemotron, Nvidia currently lacks a frontier-scale model developed wholly in-house. The company has increasingly prioritized fostering an ecosystem of open-weight models—AI systems whose parameters are publicly available, allowing users to customize and deploy them on their own hardware. These models can be more cost-effective for businesses compared to top-tier proprietary systems like OpenAI’s ChatGPT or Anthropic’s Claude.
Nvidia’s CEO Jensen Huang has been vocal about the need for a domestic open-weight AI ecosystem, describing it as vital for enabling innovation and economic growth in the United States. In September, Nvidia announced a $13 billion agreement to acquire Hugging Face, a major repository hosting millions of AI models and datasets.
Reflection AI is viewed as a strategically important entity in the US effort to compete with Chinese AI firms such as DeepSeek. Recently, Reflection launched its first public customizable AI product operating on Nvidia’s hardware. The company claims its model matches the capabilities of Z.ai’s GLM-5.2, a prominent Chinese competitor. Reflection’s founders, Misha Laskin and Ioannis Antonoglou, previously contributed to Google’s Gemini AI model.
The firm has documented partnerships with the Pentagon and the Department of Energy, and it has secured agreements to develop AI technologies for allies including South Korea.