Skip to content
Theme
OnDevice Tools

AIAI Models & Research

Nvidia nears $500 million deal to buy diffusion model startup Inception

Nvidia is reportedly close to acquiring the two-year-old AI startup Inception for at least $500 million, adding diffusion-based language models to its open-source Nemotron lineup and robotics work.

Illustration of an Nvidia logo beside a stylised diffusion model diagram on a dark background

Key takeaways

  • Nvidia is reportedly close to acquiring Inception for at least $500 million, per The Information via The Decoder.
  • Inception builds language models with diffusion rather than token-by-token generation and recently released Mercury 3.
  • The deal would support Nvidia's open-source Nemotron models and its robotics efforts.
  • Co-founders Stefano Ermon, Aditya Grover and Volodymyr Kuleshov are expected to join Nvidia full time.
In this article
  1. What the reported deal involves
  2. Why Inception's technology matters
  3. What Nvidia plans to do with it
  4. Background to the talks

What the reported deal involves

Nvidia is close to buying the AI startup Inception for at least $500 million, according to The Decoder, which cites three people familiar with the talks who spoke to The Information. The startup is two years old and works from the San Francisco Bay Area.

The deal would reportedly include performance bonuses on top of the base sum. Neither company has commented so far, so the terms remain unconfirmed.

Why Inception's technology matters

Illustration: AI, NVIDIA logo, business, price and the labels Inception, Mercury 3, Nemotron

Inception builds language models using diffusion, the technique familiar from image generation. Instead of producing text one token at a time, as models behind ChatGPT and Claude do, its models work on a complete response and improve it across several passes.

The company has more than 30 employees and recently launched Mercury 3, its newest model.

What Nvidia plans to do with it

The acquisition is meant to strengthen Nvidia's open-source Nemotron models and its work on physical models for robots, The Decoder reports. That could help Nvidia compete with OpenAI and Anthropic, both of which are building their own inference chips.

Nemotron is already used by Palantir and US government agencies as an alternative to Chinese models. Inception's diffusion approach could make these models faster or more efficient, though the sources do not confirm any specific performance gains.

Background to the talks

According to The Decoder, talks began before the resignation of Nvidia CEO Jensen Huang and after Microsoft had expressed interest in Inception. These details come from unnamed sources and have not been officially confirmed.

If the deal goes through, it would be a notable bet on diffusion as an alternative to standard text generation. It would also give Nvidia a team with strong academic roots in the technique.

Sources

  1. Nvidia bets at least $500 million on Inception's diffusion models to strengthen its open-source Nemotron lineupThe Decoder

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.