Nvidia Plans to Acquire Diffusion-Model Startup Inception for at Least $500 Million
Nvidia is in advanced talks to acquire AI startup Inception for at least $500 million, with additional incentives tied to performance targets. Inception’s diffusion large language model technology and research team are expected to support development of the open-source Nemotron models, while its three co-founders plan to join Nvidia full-time. Unlike autoregressive models, the technology generates answers in parallel and repeatedly refines them in full; the report says its speed advantage is more pronounced when generating longer responses.
Why read this
AI score 63
If completed, the deal would fold the diffusion large language model team into Nemotron development, reflecting Nvidia’s expansion beyond supplying computing power into models and talent.
Topics
Body · AI translation
Nvidia is in advanced talks to acquire AI startup Inception in a deal valued at no less than $500 million, with additional performance-based incentives. The move is intended to bring diffusion-model technology and key talent to Nvidia’s open-source model project, Nemotron.
According to a November 19 report by technology outlet The Information, three people familiar with the matter said the negotiations have reached an advanced stage. Inception’s diffusion-based large language model technology and the research team behind it are expected to work directly on developing Nvidia’s open-source Nemotron models. The talks began before Nvidia CEO Jensen Huang’s departure, and there is currently no indication that Nvidia will withdraw because of the leadership change.
Notably, Microsoft had previously approached Inception about a similar transaction, but the talks ultimately failed to progress.
According to the report, people familiar with the matter said the greatest advantage of diffusion models over autoregressive models is speed, particularly when generating longer responses.
Diffusion technology was previously used mainly for image generation and has only in recent years been introduced into language-generating AI by startups such as Inception and major developers such as Google DeepMind. In May this year, Nvidia released a version of Nemotron that combines diffusion with autoregression and other traditional forms of language AI.
Inception was founded about two years ago and is headquartered in the San Francisco Bay Area. It currently has more than 30 employees and has raised $50 million from investors including Menlo Ventures, M12, Mayfield and Nvidia.
The company attracted industry attention in early 2025 with a diffusion-based large language model. Unlike the “autoregressive” models used by products such as ChatGPT and Claude, which generate answers word by word from left to right, diffusion language models first produce an apparently jumbled complete answer and then repeatedly refine it to arrive at the final output.
According to the report, a person familiar with the matter said Inception co-founders Stefano Ermon, Aditya Grover and Volodymyr Kuleshov are expected to join Nvidia full time. The three are prominent machine-learning professors at Stanford University, the University of California, Los Angeles and Cornell University, respectively.
People familiar with the matter also said the researchers plan to extend their work at Nvidia into so-called “physical models” to help robots perceive and process their surroundings.
Nemotron: An Open-Source Strategy to Compete With Anthropic and OpenAI
The talks are the latest in a series of moves by Nvidia to continue expanding the Nemotron team. According to the report, Nvidia has invested tens of billions of dollars in Nemotron-related models, including several billion dollars spent acquiring or preparing to acquire technology and AI researchers. Nemotron-related teams now have hundreds of employees.
Nemotron’s strategic role is to provide corporate and government customers with an alternative to closed-source AI from companies such as Anthropic and OpenAI, while advancing Nvidia’s push into robotic AI. Booz Allen, Palantir and multiple U.S. government agencies unwilling to use Chinese open-source models are already Nemotron users.
Meanwhile, Anthropic and OpenAI are developing their own AI server chips to reduce their reliance on Nvidia GPUs. Nvidia’s continued investment in Nemotron is also viewed as a strategic countermeasure at the AI-model layer. Risk Warning and Disclaimer
Markets involve risk, and investments should be made with caution. This article does not constitute personal investment advice, nor does it take into account the particular investment objectives, financial circumstances or needs of individual users. Users should consider whether any opinions, views or conclusions in this article are appropriate for their specific circumstances. Any investment made on this basis is at the investor’s own risk.
AI Trends