What is reported, and what is not

Start with the status of the story itself. The Information's report, published on November 19 and summarized by Dealroom, TipRanks and others, describes advanced talks, three unnamed sources, a price floor of $500 million and earnouts on top. As of that report, no agreement had been signed or announced, and neither company had said anything publicly. Acquisition talks at this stage fail regularly, and the terms reported mid-negotiation are the ones most likely to change.

What the reporting does establish is direction. One source told The Information that Inception's technology and team could help Nvidia build out its Nemotron family of open models. Buying the most prominent commercial diffusion team would be a way to staff that line of work in one move.

The talks also reportedly began before Jensen Huang's resignation as Nvidia's chief executive took effect on November 12. The reporting does not say who initiated them.

The price needs its context

At least $500 million is a floor, not a total: the earnouts sit on top of it, tied to performance targets for the employees who join, so the eventual figure, if a deal closes, would be higher than the floor by an amount nobody outside the talks can size. Set it against the earlier reporting anyway, with care. In May, Reuters reported, as relayed by Techmeme and The Neuron, that Inception was asking above $1 billion, with Microsoft in discussions and SpaceX having shown interest earlier. An asking price and a reported bid floor plus earnouts are not the same kind of number, and the sources do not support reading the gap between them as a settled discount.

The confirmed part: the API is ending

Inception has emailed its customers that it is winding down the Mercury API. The notice, quoted in full by a customer, says the API will remain fully available through December 28, 2026. After that date, customers will no longer have access to the service or any data submitted to the platform, and the company encourages them to retrieve any necessary data before then. Mercury models stopped being served through OpenRouter on November 8, per a Discord message from an Inception team member. Inception has published no blog post explaining the decision. The notice gives no reason and names no buyer.

So the sequence is checkable without believing a single anonymous source. A startup whose models were its product has set a date after which access to them ends. The practical consequence for anyone building on Mercury is already dated, whatever the reason turns out to be.

What to do with it

If your product calls the Mercury API, treat December 28 as a migration deadline, not a news event. Retrieve the data you need from the platform now, while access remains, and line up a replacement model before the holiday week, not during it.

If you are watching the deal itself, wait for a signature. The useful signal will be an announcement from either company. A diffusion team inside Nvidia would matter for Nemotron. None of that has happened yet.

Sources

  1. [1] Dealroom, "Nvidia is in advanced talks to buy diffusion-model startup Inception for at least $500M," November 19, 2026 (summary of The Information's paywalled report, which was not read directly)Read source
  2. [2] TipRanks / TheFly summary, "Nvidia Nears $500M Deal for AI Startup Inception to Strengthen AI Models," November 2026Read source
  3. [3] Seeking Alpha, "Nvidia in advanced talks to buy diffusion AI startup Inception for over $500M — report," November 2026Read source
  4. [4] The Neuron, "Nvidia's Potential Inception Deal: Why Diffusion AI Matters," November 20, 2026 (May asking price and Microsoft/SpaceX interest attributed there to Reuters reporting as summarized by Techmeme; Mercury 3 figure attributed to Inception)Read source
  5. [5] Inception Mercury API wind-down notice (customer email, November 2026), quoted in full on Hacker NewsRead source
  6. [6] Daily AI Brief, "Inception tells customers it will shut down its Mercury diffusion-model API on 28 December," November 2026Read source
  7. [7] ALFW, "Mercury 3 Claims 1,068 Tokens a Second. Its Quality Lead Is Less Clear," October 19, 2026Read source