- Nvidia is reportedly exploring a full acquisition of Reflection AI, an increased equity stake or an acqui-hire arrangement involving the startup’s researchers and technology. Negotiations remain preliminary.
- Reflection AI, valued at $25 bn before new financing in April 2026, has received approximately $800 mn from Nvidia and recently introduced Beam, an open-weight model with 501 bn parameters.
- A potential transaction would strengthen Nvidia’s position in open-weight AI development, following its $20 bn technology licensing and recruitment deal with Groq and its nearly $13 bn agreement to acquire Hugging Face.
Nvidia is in preliminary discussions to acquire US artificial intelligence startup Reflection AI or expand its existing investment, according to the Financial Times, citing people familiar with the negotiations. The potential transaction would deepen Nvidia’s involvement in developing open-weight AI models as competition intensifies among American and Chinese technology companies.
The discussions could result in a full takeover, additional equity financing or an acqui-hire arrangement in which Nvidia recruits key researchers and licenses Reflection AI’s technology without acquiring the entire business. Another possibility involves providing additional computing infrastructure and semiconductor capacity.
No agreement has been reached, and the negotiations could end without a transaction. People familiar with the discussions indicated that a deal could emerge within weeks, although neither company has confirmed the proposed structure or financial terms.
Nvidia is already a major shareholder in Reflection AI, having invested approximately $800 mn in the company. The startup’s other investors include Disruptive, Sequoia Capital and 1789 Capital, whose partners include Donald Trump Jr.
Reflection AI announced a $2 bn funding round in October 2025. In April 2026, chief executive Misha Laskin confirmed that the company had closed another financing round at a $25 bn pre-money valuation.
The valuation represents Reflection AI’s estimated worth before the addition of new investment capital and does not establish the price Nvidia might pay for an acquisition.
Founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, New York-based Reflection AI initially concentrated on software development automation before expanding into foundation models for enterprises, developers and government organizations.
On October 5, 2026, the company introduced Beam, its first publicly announced open-weight AI model, designed for coding, reasoning and autonomous AI-agent applications.
Beam uses a mixture-of-experts architecture with 501 bn total parameters, of which 23 bn are activated during processing. The architecture is intended to reduce computational requirements while maintaining competitive performance.
The model was undergoing final safety testing and evaluation ahead of the planned wider release of its weights. Reflection AI says Beam can compete with leading Chinese AI systems, although its performance claims remain subject to independent evaluation.
Open-weight AI models allow developers and businesses to download, modify and deploy models on infrastructure they control, subject to licensing restrictions. This offers an alternative to proprietary AI services, including those supplied by OpenAI and Anthropic.
The technology has attracted attention from US policymakers seeking domestic alternatives to lower-cost Chinese models developed by companies such as DeepSeek, Alibaba and Z.ai.
Reflection AI has also established relationships with US defense and energy agencies and reached agreements to develop AI models for countries including South Korea.
Its first public model runs on Nvidia hardware, connecting the startup’s software development directly to Nvidia’s computing ecosystem.
Nvidia already develops its Nemotron family of open-weight AI models but has increasingly emphasized partnerships and investments to expand its presence in advanced model development.
A closer relationship with Reflection AI could provide Nvidia with additional research capabilities, proprietary technology and a stronger position in the market for customizable AI systems.
The potential transaction would follow several significant investments and agreements involving AI companies.
In September 2026, Nvidia agreed to acquire Hugging Face, a widely used platform hosting AI models, datasets and developer applications, for approximately $12.93 bn. The transaction remains subject to closing conditions, including regulatory approvals.
Nvidia previously entered a technology licensing and recruitment arrangement with AI chip developer Groq, reportedly valued at approximately $20 bn. The structure allowed Nvidia to license intellectual property and hire senior personnel without completing a conventional corporate takeover.
That agreement has subsequently faced legal challenges from former Groq employees concerning its financial arrangements.
Similar acqui-hire structures have become more prominent in the technology industry because they can offer access to specialized talent and intellectual property without requiring a full acquisition. Such transactions can nevertheless attract regulatory scrutiny.
A potential Reflection AI deal could raise questions about competition and access to advanced computing resources, particularly given Nvidia’s dominant position in AI semiconductor infrastructure.
For Nvidia, additional investment in model developers also reinforces demand for the processors, networking systems and computing platforms used to train and deploy artificial intelligence.
The outcome will depend on negotiations over ownership, technology licensing and personnel arrangements. For now, Nvidia’s discussions with Reflection AI remain preliminary, with no confirmed acquisition price, additional investment commitment or binding transaction.
