On August 27, 2026, The Information reported that NVIDIA had agreed to acquire Hugging Face for $12.9 billion. The exclusive report was based on a single source familiar with the deal, and neither NVIDIA nor Hugging Face has made an official announcement. Hugging Face has not become a subsidiary of NVIDIA.
Shortly before that, Business Insider reported that the two companies were "in talks over a valuation exceeding $13 billion, but had not yet reached an agreement." Both reports describe an intermediate stage—either ongoing negotiations or an agreement—and neither indicates that the acquisition has been completed. The most recent claim is that the deal "was reportedly agreed to," but a signed definitive agreement and deal completion remain unconfirmed.
From 'In Talks' to 'Reportedly Agreed' in About 52 Minutes
The apparent inconsistency between the reports comes down to the verbs used and the publication times. Laying out the two original reports in chronological order shows how the claims shifted:
| Publication Time (UTC) | Outlet | Reported Status | What Remains Unconfirmed |
|---|---|---|---|
| Aug 27, 00:34 | Business Insider | In talks for several weeks; valuation exceeds $13 billion; deal not yet agreed and could still fall through | Formal offer, signatures, closing conditions |
| Aug 27, 01:26 | The Information | Agreed to acquire for $12.9 billion, according to a source familiar with the deal | Official confirmation from either company, signing date, payment method, expected closing date |
It's possible that negotiations advanced within that roughly 52-minute window. It's also possible that the two sources simply had knowledge of different stages of the process. Public information alone can't determine which explanation is correct, but if we go with the later report from The Information, the accurate phrasing is that the acquisition "was reportedly agreed to."
The Information's English headline also used "Agrees to Buy," not "Completed," which would indicate the acquisition had closed. Even if the agreement is accurate, it cannot be read as meaning that control has already changed hands.
The $12.9 Billion Price Tag Is for a Model Distribution Pipeline
According to The Information, Hugging Face's most recent annualized revenue run rate was $150 million. The reported $12.9 billion price is 86 times that figure. An annualized run rate is a metric derived by multiplying recent monthly revenue (or similar) by 12—it is neither actual full-year revenue nor profit. The reported price is difficult to explain based on current business scale alone.
The valuation reported during Hugging Face's 2023 funding round was $4.5 billion. The current reported figure of $12.9 billion is roughly 2.87 times that amount. There is a large gap between the reported price and the company's current revenue scale, but neither the method used to calculate the acquisition price nor a breakdown by business segment has been disclosed.
Hugging Face's value lies in controlling the entry point through which AI developers find, download, and share derivative versions of models. According to official documentation, the Hub hosts more than 2 million public models, 1.5 million public datasets, and 1.5 million public AI apps called "Spaces." Separately, enterprises can also use private repositories and access controls, along with pathways to run models through external inference services. The ability to handle both public and internal assets through the same development workflow is what gives the enterprise offering its value.
The connection with NVIDIA is already in place. In 2023, the two companies announced a partnership enabling users to connect from Hugging Face to DGX Cloud to train and fine-tune models. At the time, the companies cited more than 250,000 models and 50,000 datasets. Today, the figures stand at more than 2 million public models and 1.5 million public datasets. However, because the model counts at both points in time are stated as lower bounds ("more than"), the exact growth multiple cannot be calculated.
The same joint announcement also noted that more than 15,000 organizations were using Hugging Face at the time. Since current official documentation doesn't provide a comparable organization count, it's not possible to calculate how much enterprise usage has grown. Still, the increase in models and datasets suggests that the pool of development projects NVIDIA could connect to has expanded.
If NVIDIA is indeed targeting something specific, it's likely the pipeline from selecting a model to deploying it in a runtime environment—rather than storage capacity itself. This would let NVIDIA link its GPU and cloud computing resources to the Hub that developers already interact with daily. However, no product plans or priorities for after the acquisition have been announced. This strategic rationale is inferred from the existing partnership and the Hub's functionality—it is not a confirmed integration plan.
Owning the Company Isn't the Same as Owning 2 Million Models
Acquiring Hugging Face the company is not the same as acquiring outright ownership of the more than 2 million models hosted on the Hub. The terms of service leave ownership of user-created content with the users themselves. What Hugging Face receives is a worldwide, royalty-free, non-exclusive license to use that content for providing the service and for purposes permitted under its terms and privacy policy. If a repository displays a license, those terms continue to apply.
The rights that Hugging Face itself holds are a separate matter. The same terms of service stipulate that Hugging Face owns the intellectual property related to the design, code, and trademarks associated with its website and services. The scope of the acquisition agreement hasn't been disclosed, but legally, the company's own assets—which NVIDIA presumably seeks to acquire—are distinct from the content that users have entrusted to the platform.
The official figure of more than 2 million refers to public models. Separately, the Hub also supports collaboration for internal, private teams. Public repositories each carry their own usage terms, and whether commercial use is permitted or redistribution is allowed depends on those individual licenses. Even if NVIDIA acquires Hugging Face, the copyrights held by contributors and model-developing companies do not automatically transfer along with it.
Control over the platform itself, however, is significant. Decisions about what's easier to discover through search and recommendations, and how inference services are configured by default, shape how models get used in practice. Changes to enterprise pricing and access controls would also affect the choices of customers who store internal assets on the platform. Business Insider noted that Hugging Face's neutrality—supporting industry players including AMD and Intel—has been one of its core strengths. If the acquisition officially proceeds, whether non-NVIDIA hardware and external services continue to be treated on equal terms will be a key consideration for developers.
The Arm Precedent: The Gap Between 'Agreement' and 'Completion'
NVIDIA itself has a precedent in which a deal reached a definitive agreement but was never completed. NVIDIA and SoftBank signed a definitive agreement for NVIDIA's acquisition of Arm in 2020, but they were unable to resolve serious regulatory obstacles and terminated the deal in February 2022. Signing an agreement and an actual transfer of control over the target company are two separate events.
In the case of Hugging Face, neither company has even confirmed that a definitive agreement exists. Because it's still soon after the report, the mere absence of an official announcement doesn't necessarily disprove that an agreement was reached. At this point, all that can be confirmed is that a report—based on an anonymous source—claims an agreement was reached.
The deal's status can be considered to have advanced only once both companies announce a signed agreement and its terms. Such an announcement could reveal how the $12.9 billion would be paid, the closing conditions, and the expected completion timeline. The acquisition could only be described as completed after the necessary reviews and conditions are met and control actually changes hands. The fate of Hugging Face's management team also cannot be determined unless the companies announce it.
Hugging Face's value has come from its role as a neutral gateway connecting diverse models and computing infrastructure, drawing users precisely because of that neutrality. Whether NVIDIA can deepen the connection to its own infrastructure while preserving that value will ultimately be judged not by the language of any official agreement, but by how it treats non-NVIDIA hardware, external inference services, and individual licenses after the acquisition.