Updated 5 Sep 2026: Replaces the reported/unconfirmed acquisition framing with NVIDIA's definitive agreement and SEC filing: $12.9303B announced value, roughly $11.9B to stockholders plus up to $1.0B retention, H1 2027 expected close, regulatory conditions, and explicit commitments to keep Hugging Face open and support other silicon vendors.
NVIDIA formally agrees to acquire Hugging Face for $12.93 billion — and promises to keep it open across rival hardware
NVIDIA has formally agreed to acquire Hugging Face in a deal it values at $12.9303 billion. Its SEC filing says roughly $11.9 billion goes to stockholders and up to $1.0 billion is reserved for employee retention, with closing targeted for the first half of 2027 and explicit commitments to keep the platform open to other models, clouds and silicon vendors.
NVIDIA publicly announced the acquisition on September 3, 2026 after entering a definitive agreement on September 2.
NVIDIA states a total announced transaction value of $12,930,300,000.
The SEC 8-K says approximately $11.9 billion is payable to Hugging Face stockholders, subject to adjustments.
Up to approximately $1.0 billion is reserved for an equity-based employee retention program.
The deal is expected to close in the first half of 2027, subject to customary conditions and required regulatory approvals.
NVIDIA says Hugging Face will remain open to developers' choice of models, frameworks, clouds, inference providers and computing platforms.
NVIDIA says its compute will not be required to build on or deploy through Hugging Face.
The SEC filing specifically says Hugging Face will continue to support other silicon vendors.
NVIDIA reports more than 18 million users and more than 200,000 companies on Hugging Face; those are first-party figures.
What builders should take away
If Hugging Face is a production dependency, the official multi-silicon commitment reduces the need for an immediate migration, but keep export, mirroring and reproducible-deployment paths tested until post-close behavior is observable.
Track defaults and optimization effort, not only formal compatibility. A platform can remain technically open while becoming commercially or ergonomically biased toward one hardware ecosystem.
For model publishers, keep canonical artifacts and metadata portable across more than one distribution surface where the operational cost is reasonable.
If you depend on non-NVIDIA accelerators, watch Hugging Face's integrations, benchmarks and hosted inference options after closing for evidence that the explicit other-silicon commitment is being honored in practice.
Treat the first-half-2027 date as a target, not a certainty; regulatory review or other closing conditions can alter timing or terms.
What changed
NVIDIA formally announced on September 3, 2026 that it has agreed to acquire Hugging Face for $12,930,300,000, resolving the central uncertainty in BTN's earlier dossier, which was based on reported but unconfirmed deal talks. NVIDIA's September 2 Form 8-K says the definitive agreement includes approximately $11.9 billion payable to Hugging Face stockholders, subject to adjustments, plus an equity-based retention program of up to approximately $1.0 billion for Hugging Face employees joining NVIDIA. The transaction is expected to close in the first half of 2027, subject to customary closing conditions including required regulatory approvals. NVIDIA says Hugging Face will remain open to the broader AI ecosystem: developers can keep choosing their models, frameworks, clouds, inference providers and computing platforms, NVIDIA compute will not be required, and the SEC filing specifically says Hugging Face will continue supporting other silicon vendors.
Why it matters
The acquisition puts a central open-model distribution and tooling layer under the dominant AI-accelerator supplier, but the definitive agreement now gives builders concrete commitments to evaluate rather than only speculate about. NVIDIA is publicly promising that Hugging Face will remain multi-model, multi-cloud and multi-silicon, which lowers the immediate risk of an abrupt NVIDIA-only turn. It does not eliminate platform risk: the real test will be post-close product behavior, default integrations, pricing, optimization priorities and governance. The transaction structure also shows NVIDIA is paying not simply for current software revenue but for ecosystem distribution, developer reach and strategic position across the open-model stack.
The acquisition is now a definitive agreement
NVIDIA's September 2 SEC filing says it entered into a definitive agreement to acquire Hugging Face, and NVIDIA publicly announced the transaction the following day. That materially changes the dossier from a credible reported deal to a company-confirmed transaction subject to closing conditions.
The headline price includes a large employee-retention component
NVIDIA describes the transaction as worth $12.9303 billion. The Form 8-K provides the more useful breakdown: approximately $11.9 billion is payable to Hugging Face stockholders, subject to adjustments, while up to approximately $1.0 billion is an equity-based retention program for Hugging Face employees who join NVIDIA. Builders and investors should therefore avoid treating the entire headline figure as a conventional purchase price paid to existing shareholders.
NVIDIA has made openness part of the deal's public commitment
Jensen Huang says Hugging Face will remain an open platform for the whole AI ecosystem and that developers will be able to choose models, frameworks, clouds, inference providers and computing platforms. NVIDIA explicitly says its own compute will not be required to build on or deploy through Hugging Face. The SEC filing separately commits to continued model/dataset choice and support for other silicon vendors.
The commitment reduces immediate risk but does not settle neutrality
The promise to support rival hardware is meaningful because Hugging Face sits between model authors, developers, clouds and accelerator vendors. But neutrality is ultimately observable in product decisions: default runtimes, benchmark emphasis, optimization investment, commercial integrations and pricing can change while formal compatibility remains. AMD, Intel, Google, AWS and other partners therefore remain important signals after closing.
Closing is targeted for the first half of 2027
The transaction is expected to close in the first half of 2027, subject to customary conditions and required regulatory approvals. Until then, the companies remain separate and the acquisition can still be delayed, modified or fail to close. NVIDIA's SEC filing also highlights regulatory risks around open models and global model distribution, including restrictions that could affect models originating in different jurisdictions.
The strategic value is developer distribution
NVIDIA says more than 18 million developers, researchers and creators use Hugging Face, with more than 3 million models, 500,000 datasets and 1 million applications on the platform, and more than 200,000 companies using it. Those vendor-reported figures help explain why ownership of Hugging Face is strategically different from acquiring a single model vendor: the asset is a distribution and collaboration layer across much of the open-model ecosystem.
What to watch next
Regulatory review and whether the first-half-2027 closing target changes.
Any Hugging Face or NVIDIA governance commitments beyond the current openness and multi-silicon statements.
Changes to Hub pricing, Inference Endpoints, model discovery, benchmark presentation and preferred deployment paths.
Whether AMD, Intel, Google, AWS and other ecosystem partners retain equivalent integration and optimization support after closing.
Leadership retention and whether the up-to-$1.0B employee program keeps key Hugging Face teams intact.
Any restrictions on geographically sensitive open models or datasets as NVIDIA navigates regulatory obligations.
Still unclear
The transaction has been signed but has not closed and remains subject to regulatory approvals and other customary conditions.
NVIDIA's openness commitments are explicit, but future neutrality will depend on implementation and product behavior after closing.
The $12.9303 billion headline value is not identical to cash or consideration paid to existing shareholders because the SEC filing separates roughly $11.9 billion of purchase consideration from up to $1.0 billion of employee retention equity.
NVIDIA's user, repository and company counts are first-party figures rather than independently audited metrics.
U.S. Securities and Exchange Commissionregulatory filing
Definitive-agreement confirmation, approximately $11.9B purchase consideration, up to approximately $1.0B employee retention program, H1 2027 expected close, regulatory conditions and commitment to support other silicon vendors.
K2 Horizon is notable less for another benchmark claim than for reproducibility: IFM is publishing model weights, architecture, training code, data or construction recipes, evaluation resources and intermediate training material instead of stopping at a final checkpoint.
Funes treats agent memory as user-owned data rather than a hosted account feature: retrieval and reranking run locally, provenance stays attached to recalled passages, and cross-machine sharing is optional. The main risk is that publishing session-derived memory can still expose secrets if redaction misses them.
Hugging Face has released 207 Apache-2.0 WebGPU kernels, a JavaScript loader and Fleet, a browser benchmarking service. The package makes kernel contracts and correctness evidence inspectable, but performance remains device- and workload-dependent.