Separating AI hype from reality.

Nvidia Reported in Advanced Talks to Acquire Diffusion AI Startup Inception for $500M+

November 20, 2026 · Nvidia · Funding

Nvidia is in advanced discussions to acquire Inception, a two-year-old startup building diffusion-based large language models, for at least $500 million plus performance-based earnouts, according to a November 19 report from The Information citing three people familiar with the talks [2][4][5]. Neither company has confirmed the deal, and terms could still change.

What Inception builds

Inception, founded by Stanford professor Stefano Ermon (CEO), UCLA professor Aditya Grover, and Cornell professor Volodymyr Kuleshov, develops "diffusion large language models" (dLLMs) under the Mercury brand. Unlike conventional autoregressive models that generate text token-by-token, diffusion models start with a rough full draft and refine all tokens in parallel over multiple steps—the same principle behind image generators such as Stable Diffusion and Sora [3][6].

Inception claims its Mercury models run 5–10× faster than speed-optimized models from OpenAI, Anthropic, and Google while matching their accuracy [6]. The company's own benchmarks show Mercury 3 exceeding 1,000 tokens per second with a 1 million-token context window [3]. These speed claims are vendor-reported and have not been independently reproduced across workloads [3]. Inception raised a $50 million seed round in November 2025 led by Menlo Ventures, with participation from Mayfield, Innovation Endeavors, Microsoft's M12, Snowflake Ventures, Databricks Investment, and Nvidia's venture arm NVentures [6].

Nvidia's expanding model strategy

The reported talks fit a pattern of Nvidia moving beyond hardware into model ownership. Inception's technology would feed into Nvidia's Nemotron portfolio of open models, which are already used by Booz Allen Hamilton, Palantir Technologies, and several U.S. government agencies [2]. The Information's report, as relayed by The Decoder and summarized in [3], says Inception's team could also contribute to Nvidia's "physical AI" models for robotics, where diffusion methods are already common in research.

Nvidia has been investing in AI labs and model makers across the year, including through its NVentures venture arm and the separately named Nvidia Inception startup program [3]. The company's existing investment in Inception via NVentures means the chipmaker already held a financial stake in the startup it is now reportedly seeking to acquire [5].

Deal structure and timing

The reported price—roughly 10× Inception's seed round amount—is notably below the >$1 billion valuation Reuters said the startup sought in May 2026, when Microsoft and SpaceX were also reported to be interested [4]. The package includes earnouts for Inception's 30+ employees tied to performance targets [2][4]. All three co-founders would join Nvidia full-time [2][5].

The Information reports that talks began before Nvidia's board accepted Jensen Huang's resignation as CEO. An SEC filing shows the board accepted Huang's resignation on November 11, 2026, effective immediately [7]. CFO Colette Kress was appointed interim CEO while the board searches for a permanent replacement [7]. Microsoft's M12 participated in Inception's seed round, and The Information says Microsoft also expressed acquisition interest [2][3][5].

Outstanding questions

No confirmation has come from either Nvidia or Inception. All details stem from a single paywalled report relayed through secondary outlets [3]. Inception notified customers on November 12 that its Mercury API will shut down December 28, 2026, with no mention of a buyer [3][4]; it is unclear how this relates to the reported talks. The report does not say whether Mercury will remain available, be folded into Nemotron, or be discontinued. A half-billion-dollar acquisition by Nvidia could draw antitrust attention; no filings or comments have surfaced. Whether Microsoft made a formal offer is unconfirmed [3].

Implications for users and builders

If you use Mercury via API, plan for the December 28 shutdown regardless of the acquisition outcome [3]. Export data and line up alternatives now. If you are evaluating diffusion LLMs for latency-sensitive work, test them on your own prompts and hardware—vendor speed claims are not a substitute for independent measurement [3].

The reported deal, if it closes, would give Nvidia a distinct model architecture (diffusion) alongside its autoregressive Nemotron line, plus a research team with roots in the invention of diffusion methods. Whether that translates into a shipping advantage remains to be seen.

Sources

  1. Nvidia in advanced talks to acquire Stanford-led diffusion AI startup ... (techfundingnews.com)
  2. Nvidia Nears $500M Deal for AI Startup Inception to Strengthen AI Models (tipranks.com)
  3. Nvidia Inception $500M Deal: Reported vs Confirmed (explainx.ai)
  4. Nvidia in advanced talks to buy diffusion-LLM startup Inception for ... (companieshistory.com)
  5. Nvidia In Talks To Buy Inception For $500M To Boost AI Models (ascendants.in)
  6. Inception Raises $50M to Power Diffusion LLMs, Increasing LLM Speed and Efficiency by up to 10X and Unlocking Real-Time, Accessible AI Applications (businesswire.com)
  7. nvda-20261111 (sec.gov)