The deal is signed. It has not closed. The product you actually run is the Hub: more than three million models, a default pull path, and a neutrality story attached to an acquisition expected in the first half of 2027. Jensen Huang posted $12,930,300,000 on 3 September 2026. The 8-K filed for the 2 September agreement is where close timing and the consideration mix live. The stays-open sentences are NVIDIA's current public promise. They are not a regulatory docket.

What it does

Hugging Face operates a platform for developing, sharing, and deploying open-source models, datasets, and applications, per NVIDIA's 8-K. Huang's post gives NVIDIA's scale figures: more than 18 million developers, more than 3 million models, 500,000 datasets, 1 million applications, more than 200,000 companies. NVIDIA's self-reported footprint on the same post: more than 500 models and more than 250 open datasets; "largest contributor of open models and data."

The operator feature in the post is the neutrality clause. Hugging Face "will remain an open platform." Developers choose models, frameworks, clouds, inference providers, and compute platforms. "NVIDIA compute will not be required to build on or deploy through Hugging Face." Open-source and open-weight models from every model builder stay supported. Multi-cloud and multi-accelerator development and deployment stay supported. Huang ties that to an open-weights letter he coauthored; the letter is cited as background, not as a schedule to the merger agreement.

TechCrunch the same day repeated the $12.93 billion figure and the open-source / open-weight support line. Reuters the same day called it a $12.93 billion buy and a bet on open-source models that can nearly match closed frontier labs at lower cost. That is a newsroom restatement of NVIDIA's number, not a second independent valuation.

Why it is moving now

NVIDIA is proposing to own the default directory where everyone else publishes weights. These documents do not say what share of training clusters that implies. The operator question is whether the Hub remains the accelerator-neutral town square after the first half of 2027, and what you do with mirrors until then.

Close is expected in the first half of 2027, subject to regulatory approvals. Cash to stockholders is about $11.9 billion, subject to adjustments. Retention is up to about $1.0 billion in equity for joining employees. The 8-K itself does not give formal opening dates for U.S. or EU antitrust reviews.

License + activity + maturity (demo | usable | production-shaped)

License: the Hub's terms of service plus each model's own card (MIT, Apache, custom). NVIDIA buying the company does not rewrite those files. Activity: definitive agreement on 2 September 2026; Huang post on 3 September; expected close in the first half of 2027. Maturity: usable today as the same Hub you already pull. Production-shaped as a vendor-neutral registry only if you treat NVIDIA's multi-accelerator sentence as a current policy, pin weight digests, and keep a second mirror. Demo: any slide that says "stays open" without the 8-K close window.

When to pick it vs the default alternative

Keep pulling from Hugging Face when your pipeline already depends on it and you can pin commit hashes, weight SHAs, and license text in your registry.

Stand up a second mirror (an internal registry, object storage, or another host) when Hugging Face is your only copy of a production checkpoint. The post promises NVIDIA compute will not be required. It does not promise unchanged terms of service, rate limits, or inference pricing through close.

Pick a cloud-native model garden (Bedrock, Vertex, Azure) when procurement wants one vendor invoice and you do not want Hub change-of-control risk in the critical path.

Pick direct publisher Git plus object storage when the model card already allows it and you need to stop treating Hugging Face as source of truth.

The regulatory calendar is not a runtime plan. The 8-K already told you close is a 2027 event.

Failure modes

Confusing Huang's $12,930,300,000 with the 8-K's $11.9 billion plus retention. They are the same deal described two ways.

Treating "will not be required" as a structural remedy. The 8-K lists regulatory approvals as a closing condition. It does not print a behavioral consent decree.

Assuming model-card licenses change at signing. They do not, on these documents.

Assuming August's unconfirmed talks reporting was the agreement. This is the signed deal plus the neutrality pledge.

Links

The 8-K does not say whether U.S. or EU review has formally opened, or on which docket numbers.