Skip to main content
Live
Main content

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

The proposed deal would bring all three Inception co-founders to Nvidia and fold diffusion language models into its open Nemotron lineup.

Jaeden Schafer
Editor in Chief · · 4 min read
Nvidia logo

Nvidia is nearing a deal to acquire Inception, a two-year-old startup that builds large language models using diffusion, for at least $500 million, according to The Information, which cited three people familiar with the talks. Neither Nvidia nor Inception has confirmed the deal, and terms could still change.

The proposed agreement would bring all three Inception co-founders, Stefano Ermon, Aditya Grover and Volodymyr Kuleshov, to Nvidia full time. Inception's more than 30 employees would be eligible for additional earnout awards tied to performance targets. The company is based in the San Francisco Bay Area.

Inception's Mercury models generate text differently from the autoregressive models behind ChatGPT and Claude, which produce one token at a time. Borrowing the diffusion technique from image and video generation, Mercury drafts multiple tokens in parallel and refines the full response over successive passes. Inception says that approach makes generation faster and cheaper to run.

Key facts

  • 01Nvidia is nearing a deal to buy diffusion LLM startup Inception for at least $500 million, per The Information.
  • 02Co-founders Stefano Ermon, Aditya Grover and Volodymyr Kuleshov would join Nvidia full time.
  • 03Inception has more than 30 employees and raised a $50 million seed round in November 2025 that included Nvidia's NVentures.
  • 04Inception says its Mercury 3 model reaches 1,068 tokens per second on Nvidia GPUs.
  • 05An Inception notice says the Mercury API stays available through Dec. 28, 2026.

The company's speed claims are aggressive. Inception says Mercury is five to ten times faster than speed-optimized models from OpenAI, Anthropic and Google at matching accuracy, and that its latest model, Mercury 3, reaches 1,068 tokens per second on Nvidia GPUs. Those are the company's own figures and have not been independently verified.

“for the same cost, you get more things out.”
— Stefano Ermon, Inception co-founder

Inception released Mercury 2 in February 2026, billing it as the first reasoning-capable diffusion LLM, and followed with Mercury 3 in October. Inception says Mercury 3 has more than 40 times the total parameters of Mercury 1 while holding similar generation speed.

For Nvidia, the prize is its open model family. The deal is aimed at strengthening Nvidia's Nemotron models, which Booz Allen Hamilton, Palantir and several US government agencies already use, and Inception's researchers are expected to apply diffusion techniques to Nvidia's physical AI models for robotics. Nvidia is also building Nemotron 4, an open model family with at least 1 trillion parameters.

Nvidia is already an Inception shareholder. Its venture arm, NVentures, joined Inception's $50 million seed round in November 2025, alongside Menlo Ventures, M12 and Mayfield.

The talks began before Jensen Huang resigned as Nvidia CEO on Nov. 12 and followed Microsoft's interest in a similar deal. CFO Colette Kress is serving as interim CEO while the board searches for a permanent successor. Nvidia has not said whether the leadership change affected the timing or terms.

Related · from this week
Nvidia lines up Southeast Asian governments and enterprises on Nemotron models
Jaeden Schafer · 5 min read →

An Inception purchase would be Nvidia's second major talent-and-technology deal in under a year, after it brought Groq's inference technology and key engineers in-house through a $20 billion asset deal in December 2025.

Existing Mercury customers face a deadline either way. A wind-down notice from Inception says the Mercury API and platform dashboard remain available through Dec. 28, 2026, after which requests stop being served. The notice does not tie the shutdown to an acquisition.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Business

Nvidia lines up Southeast Asian governments and enterprises on Nemotron models
Business

Nvidia lines up Southeast Asian governments and enterprises on Nemotron models

At AI Day Singapore, HTX, Viettel, YTL and Sea Limited detail production deployments built on Nvidia's open Nemotron stack.

Jaeden Schafer5 min read
Nvidia logo
Business

Jensen Huang tells Dreamforce every enterprise is becoming an AI company

Nvidia and Salesforce unveiled Koa, a CRM reasoning model built on Nemotron 3 Super, trained on 27 years of CRM data with 3x fewer errors.

Jaeden Schafer5 min read
Nvidia logo
Business

Palantir taps NVIDIA Nemotron open models for US government AI

The deal puts customizable frontier models inside air-gapped federal environments, with agencies keeping the weights they train.

Jaeden Schafer5 min read