Issue 002 · September 2026For founders, marketers & owner-operators · monthlyAll issues
FoundersEngine Room
Join the Crew
Issue 002 · September 2026 · p.06 · AI & Business WireAwaiting edit

Nvidia is buying Hugging Face for $12.9B, and promises it stays open

The platform hosting three million models is joining the biggest AI chipmaker. Jensen Huang says Nvidia hardware won't be required. The deal isn't expected to close until 2027.

FER Research Agent · awaiting editUpdated

A vast dark hall of shelves lined with small glowing boxes, a single empty desk in the middle
Image generated by AI, screened in two inks

Nvidia confirmed on 3 September that it will acquire Hugging Face for about $12.93 billion, made up of roughly $11.9 billion to shareholders and up to $1 billion in retention equity for staff. The agreement was signed on 2 September and disclosed in an SEC filing. Closing is expected in the first half of 2027, subject to regulatory approval.

Hugging Face is where most of the open-source AI world lives: more than 18 million developers, over 3 million models, 500,000 datasets and a million apps, used by more than 200,000 companies. TechCrunch reported its annualised revenue at about $150 million, which puts the price at roughly 86 times revenue (a FER calculation).

Nvidia's pitch is continuity. "Hugging Face will remain an open platform for the entire AI ecosystem," Jensen Huang said, adding that "Nvidia compute will not be required to build on or deploy through Hugging Face." Hugging Face CEO Clem Delangue said the company needed "more compute, more support, more collaboration, and more visibility."

Why it matters if you're not an ML engineer

You may use Hugging Face without knowing it. Plenty of AI features in everyday tools, and plenty of agency-built prototypes, quietly pull an open model or dataset from the Hub. If someone on your team or a contractor built something with an open model, there's a good chance it came from there.

Nothing changes before 2027. The things worth watching are the ones that can shift without anyone changing a policy: which hardware gets the best default inference endpoints, which models get featured, and how quickly non-Nvidia chips get first-class support. Defaults steer most people, and defaults are easy to change quietly.

What to do this month

Cheap insurance, mostly:

  1. List the open models and datasets your products or client projects depend on. Ask whoever built them if you're not sure.
  2. Pin each one to a specific version rather than "latest", so an update can't change behaviour under you.
  3. Keep a copy of the model files you rely on in your own storage. It's usually a one-off download.
  4. If you use Hugging Face's hosted inference, note what it runs on and what it costs today, so you'll spot a change.
  5. Check the licence of every model you use commercially. An acquisition is a good prompt for the check you should have done anyway.

None of this assumes Nvidia will break its promise. It just means you won't have to rely on it.

Also in this issue

Two rows of server racks side by side in a dark aisle, one lit amber and one lit cool blue
AI & Business Wire

Two new models, one Monday, lower prices

GPT-6 Sol and Luna arrived at half the price of the models they replace. Claude Opus 5.5 costs about 40% less to run than Opus 5. Here's what changed for builders.

p.04
Four glass jars of different sizes on a dark shelf, each filled with glowing orange light
GTM Signal

Same summer, a 32x gap in what revenue is worth

Four deals between June and September, four very different prices per dollar of revenue. The spread tells founders what buyers are paying for.

p.08
A streak of warm light across a dark city at night, seen through a window in a long exposure
AI & Business Wire

Gemini's cheap price has an expiry date

The new Flash model beats the one it replaces at the same price, but that price is introductory. On 1 January it doubles.

p.10