Slider

Nvidia Nears Deal for Diffusion-AI Startup Inception

Nvidia nears a $500M+ deal for diffusion-AI startup Inception. Here's what Mercury LLMs offer and why it matters for Nemotron and open AI.
Nvidia Nears Deal for Diffusion-AI Startup Inception

Nvidia is in advanced talks to acquire Inception, a two-year-old startup that builds diffusion-based large language models, according to a report by The Information published on November 19. The reported price is more than $500 million, with additional performance-based earnouts tied to whether new employees hit specific targets. The report cites three people familiar with the matter. All three of Inception's co-founders are expected to join Nvidia, and Microsoft had reportedly shown interest in a similar deal before Nvidia's talks advanced.

Neither Nvidia nor Inception has publicly confirmed the talks. Deals at this stage can be repriced, delayed or fall through, so the details below should be treated as reported rather than final.

What Inception builds

Most chatbots run on autoregressive language models, which write text one token at a time, each chosen after the last. Inception's models take a different route borrowed from image and video generators. They begin with a rough, noisy draft of the whole answer and refine many tokens in parallel until the text resolves. The company calls these diffusion LLMs, or dLLMs, and says the parallelism is what makes them fast and cheap to run.

The company was founded by three academics who had worked together on AI research for more than a decade: Stanford professor Stefano Ermon, who is CEO and a co-inventor of the diffusion methods behind systems such as Stable Diffusion, Midjourney and Sora; UCLA professor Aditya Grover; and Cornell professor Volodymyr Kuleshov. Ermon has said he has pursued the idea of applying diffusion to text for years, and that earlier attempts by others struggled until Inception developed its own training and inference techniques.

A short timeline

  • Early 2025: Inception launched Mercury Coder, which it described as the first commercial-scale diffusion LLM. It reported throughput of about 1,000 tokens per second on Nvidia H100 chips and said the model beat several speed-optimized rivals of the time. Mayfield, which led an early investment in mid-2024, said the models ran roughly 10x faster and at a tenth of the cost of conventional LLMs.
  • November 2025: Inception raised a $50 million seed round led by Menlo Ventures, with Mayfield, Innovation Endeavors, Nvidia's venture arm NVentures, Microsoft's M12, Snowflake Ventures and Databricks participating. Andrew Ng and Andrej Karpathy invested as angels.
  • February 2026: The company released Mercury 2, billed as the first reasoning dLLM and about 5x faster than leading speed-optimized models.
  • October 20, 2026: Inception released Mercury 3, which it says is more than 40 times the size of Mercury 1 and about three times as capable on standard benchmarks. The context window grew to one million tokens, up from 260,000. Priced at $0.20 per million input tokens and $0.70 per million output tokens, with a launch discount of 50%, it is available through Inception's API, OpenRouter, Baseten and Vercel's AI Gateway.
The Next Web noted that Mercury 3's benchmark results, which compare it with OpenAI's GPT-6 Luna, Anthropic's Claude Haiku 5.5 and Zhipu's GLM-5.3 Flash, are self-reported, and that no outside group had published an independent test at launch. Inception also says it used swarms of Mercury 3 agents to improve the model itself, adding about 16% to its speed.

Why Nvidia wants it

The reported plan is to use Inception's technology and researchers to strengthen Nemotron, Nvidia's family of open-weight models, and to explore diffusion in AI for robotics. Several threads make the logic clear.

Open models are central to Nvidia's strategy. Nvidia released the Nemotron 3 family (Nano, Super and Ultra) in December 2025, alongside its acquisition of SchedMD, the developer of the open-source Slurm workload manager. In March 2026 it launched the Nemotron Coalition, a group that includes Mistral AI, Perplexity, Cursor, LangChain, Black Forest Labs, Reflection AI, Sarvam and Thinking Machines Lab, to build an open frontier model that will underpin Nemotron 4. In October it released new Nemotron 3.5 models, the first since it agreed to buy Hugging Face.

Openness supports the hardware business. Analysts at Futurum have argued that Nvidia's open releases are tied to its own hardware, making its integrated stack the most efficient path for developers. A thriving open-model ecosystem also gives customers alternatives to closed labs, several of which are building their own chips, which keeps more of the market dependent on Nvidia GPUs.

Speed sells inference. A model that answers in a fraction of the time is cheaper to serve per query, which matters for agents and coding tools that call a model many times in a row. Owning a team built around inference speed gives Nvidia a showcase for its hardware.

Diffusion is a growing bet. Google DeepMind released its own diffusion text model, DiffusionGemma, in June, and an IBM researcher has predicted many people could move to diffusion models within a few years. Most of the field is still autoregressive, so a deal would give Nvidia an early position in a minority approach.

The wider acquisition pattern

The Inception talks follow a string of Nvidia moves up the AI stack. On September 3, 2026, Nvidia announced an agreement to acquire Hugging Face for about $12.93 billion. Hugging Face had reportedly turned down a $500 million investment from Nvidia earlier, saying it did not want a single dominant investor. Forbes has also reported smaller purchases of Illumex and Kumo AI. Nvidia has additionally open-sourced its Run.ai workload platform.

Against that backdrop, a deal in the hundreds of millions is small. Market commentary from Newsquawk described the price as immaterial relative to Nvidia's cash generation, and read the deal as strategic: an intake of talent and intellectual property rather than revenue. The earnout structure, which is tied to hiring targets, points the same way.

Open questions

  • Verification: Mercury's speed and quality claims come largely from the company itself. Independent benchmarks will show how it compares at scale.
  • Regulation: Nvidia's deal activity has drawn increasing scrutiny from competition authorities, and even modest acquisitions can face longer reviews than before.
  • Independence: Inception sells to customers through multiple cloud and API channels. How an Nvidia-owned Mercury would be offered to rivals and other hardware platforms is unknown.
  • Terms: The full price including earnouts, the closing timeline and whether Mercury would remain a separate product have not been disclosed.

Bottom line

If the deal closes, Nvidia would add a team that pioneered commercial diffusion language models to its open-model effort, at a price that is a rounding error next to its Hugging Face purchase. The significance lies less in the dollar amount than in the direction: Nvidia is assembling model, developer-platform and research capabilities around the chips it already dominates.

Sources

  • The Information report on Nvidia's talks (Nov 19, 2026), as relayed by Seeking Alpha, TipRanks, Newsquawk, FourWeekMBA, AI Weekly and Futubull
  • TechCrunch, "Inception raises $50 million to build diffusion models for code and text" (Nov 6, 2025)
  • Business Wire, Inception funding and Mercury 2 announcements (Nov 2025, Feb 2026)
  • SiliconANGLE, Inception funding coverage (Nov 6, 2025) and Nvidia open-model coverage (Mar 16, 2026)
  • The Next Web, "Inception's Mercury 3" (Oct 20, 2026)
  • Mayfield, "Introducing Inception Labs" (Mar 2025); PYMNTS interview with Ermon; IBM Think on diffusion LLMs
  • Nvidia press release, Nemotron Coalition (Mar 16, 2026)
  • TechCrunch (Dec 15, 2025) and Futurum (Dec 16, 2025) on Nemotron 3 and SchedMD
  • AI Business, Nemotron 3.5 strategy (Oct 16, 2026)
  • Fortune, TechCrunch, HotHardware and eWeek on the Hugging Face deal (Aug–Sept 2026)
Like this content? Sign up for our daily newsletter to get latest updates. or Join Our WhatsApp Channel
0

No comments

both, mystorymag

Market Reports

Market Report & Surveys
IndianWeb2.com © all rights reserved