Nvidia nears Inception deal for at least $500 million, sources say


The proposed acquisition includes performance bonuses for the startup’s more than 30 employees, but neither company has signed or announced a deal.
Nvidia is nearing a deal to buy Inception, a San Francisco Bay Area startup that builds language models, for at least $500 million, The Information said, citing three people familiar with the talks. The Decoder relayed the account, which says the proposed terms include performance-based bonuses tied to targets for more than 30 employees. Neither company initially commented, and nothing has been signed or announced.
The talks began after Microsoft expressed interest in Inception and before Nvidia CEO Jensen Huang resigned. Nvidia has since appointed its chief financial officer, Colette Kress, as interim CEO.
Inception’s latest model, Mercury 3, uses diffusion, a process also used to generate images. It refines a response step by step, working on multiple tokens—pieces of text—at once. Conventional models behind ChatGPT and Claude generate text one token at a time.
Inception says Mercury 3 runs 4–8 times faster than GPT-6 Luna and Claude Haiku 5.5 and produces 1,068 tokens per second on Nvidia graphics processors.
A person familiar with the talks told The Information that Inception’s technology and team could help develop Nvidia’s open Nemotron models. The Decoder said the acquisition also aims to strengthen Nvidia’s models for robots as and develop their own inference chips, which run models to produce responses.

Inception was founded in mid-2024. Its $50 million seed round included Microsoft’s M12 fund and Nvidia’s venture arm NVentures. Reuters said in May 2026 that Microsoft was in discussions with Inception, though those talks might not result in a deal.
Inception’s three co-founders are expected to join Nvidia full time: CEO and Stanford professor Stefano Ermon, UCLA professor Aditya Grover and Cornell professor Volodymyr Kuleshov.
The Inception discussions are separate from Nvidia’s talks with Reflection AI, a US developer of models whose parameters are openly available. In October, the Financial Times said those early talks concerned a possible purchase or additional investment and could fall apart.
Separately, Inception says it is winding down the Mercury API, a service customers use to access its models. The service and its dashboard will remain available through 28 December 2026. After that, requests will stop being served and customers will lose access to the platform and data they submitted.



The chipmaker’s shares were down 3% during November 20 trading, putting it on course for a seventh straight weekly decline as markets remained open.
2 min

The reported Persian Gulf transfer came before a US warning that ships receiving cargo at sea from blockaded vessels may be treated as blockade evaders if the cargo was loaded at an Iranian port or coastline even if they never visit Iran.
3 min

The Japanese online brokerage says information from up to 11.25 million accounts and a separate category of up to 4.47 million accounts may have been exposed in a test-system intrusion.
2 min