EN
Search
Photo: The Decoder

Nvidia reportedly nears $500 million deal for diffusion-model startup Inception

Nvidia is close to buying Inception, a San Francisco Bay Area startup that builds diffusion-based language models, for at least $500 million, according to a report by The Information. The report cited three people familiar with the talks, and The Decoder relayed it. Neither Nvidia nor Inception had commented initially.

Inception was founded about two years ago and has more than 30 employees, according to the report. The deal would also include performance bonuses for those staff. The startup most recently released a model called Mercury 3.

Inception's approach differs from the chatbots most people know. It uses diffusion, the method known from image generation. Conventional models such as those behind ChatGPT and Claude generate text one token, or small chunk of words, at a time. Inception's models instead refine a full response in successive steps.

According to the report, the acquisition is meant to strengthen Nvidia's open-source Nemotron models and its physical models for robots. The move would also help Nvidia contend with OpenAI and Anthropic, which are both developing their own inference chips, the chips that run AI models once they are trained. Nemotron is already in use at Palantir and at US government agencies, where it serves as a non-Chinese option.

The talks began before Nvidia CEO Jensen Huang's resignation, and after Microsoft had expressed interest in Inception, the report said. The reporting does not describe a signed agreement, and the deal had not been announced at the time.

Inception co-founders Stefano Ermon, Aditya Grover and Volodymyr Kuleshov, who are also professors, are expected to join Nvidia full time, according to the report.

Source