Nvidia’s $500 Million Inception Deal Could Expand Nemotron

Nvidia is nearing a deal to acquire Inception, a startup developing language models that refine whole responses through diffusion. The acquisition could add a different model approach to Nvidia’s open-source Nemotron lineup, as the company competes with OpenAI and Anthropic.

WTF Index TERMINATOR
◄ Terminator 2 Idiocracy 0 ►

The acquisition could expand Nvidia’s AI models and their use by government agencies, though the story mainly describes a business deal.

Nvidia’s $500 Million Inception Deal Could Expand Nemotron

Nvidia is nearing a deal to acquire Inception, an artificial intelligence startup working on language models based on diffusion. The reported price is at least $500 million, with performance bonuses included for the startup’s more than 30 employees.

The proposed acquisition would bring Inception’s model technology and its co-founders into Nvidia. It could also support Nvidia’s effort to expand its open-source Nemotron models and compete with companies building their own inference chips.

A different way to generate language

Many familiar language models produce text one token at a time. Inception takes another route: its models begin with a full response and refine it step by step, using a diffusion process associated with image generation.

That difference is central to the potential value of the deal. Nvidia would gain a team developing language models with a distinct generation method, which it could bring into its broader model lineup. The source does not describe how Nvidia plans to combine the technology with Nemotron.

Inception most recently released Mercury 3. The company is a two-year-old startup in the San Francisco Bay Area. Its founders, professors Stefano Ermon, Aditya Grover, and Volodymyr Kuleshov, are expected to join Nvidia full time if the deal proceeds.

Why Nvidia may want the startup

The reported aim is to strengthen Nvidia’s open-source Nemotron models, as well as its physical models for robots. The move would add to Nvidia’s work on AI software at a time when it is also facing competition from OpenAI and Anthropic, which are building their own inference chips.

Nemotron already has users beyond Nvidia. Palantir and US government agencies use the models as an alternative to Chinese models, according to the source. Expanding the lineup could therefore matter to organizations that want another model option, while giving Nvidia a stronger position across both AI models and the hardware used to run them.

The source does not say what specific Inception capabilities would be incorporated into Nemotron or when any integration might happen. The deal’s potential significance rests on the technology and staff Nvidia could acquire, rather than on any announced product changes.

What is known about the talks

The Information reported the deal based on three people familiar with the talks. The proposed purchase price is at least $500 million, and the package also includes performance bonuses for Inception’s more than 30 employees.

Talks reportedly began before Jensen Huang’s resignation and after Microsoft expressed interest. The source does not give further details about Microsoft’s interest or explain the reference to Huang’s resignation. Neither Nvidia nor Inception initially commented.

For now, the acquisition remains a reported deal in progress. If completed, it would bring a diffusion-based language model team into Nvidia and could broaden the company’s open-source model efforts. The practical impact will depend on how Nvidia chooses to use Inception’s approach within Nemotron and its work on models for robots.