COMPANY PROFILE / AI & MACHINE LEARNING
Hugging Face
The sharing and deployment platform where over 3 million models gather; NVIDIA announced an agreement to acquire it in September 2026, with closing expected in the first half of 2027
Hugging Face matters in AI and machine learning because it is the common place to publish, find and run models, datasets and apps. On 15 October 2026 its site shows about 3.15 million public models and about 1.09 million public datasets. It also runs open-source libraries such as Transformers and Diffusers (Apache 2.0). But Hugging Face is expected to stop being an independent private company. NVIDIA signed a definitive agreement on 2 September 2026 and, in its SEC filing, describes a purchase price of about US$11.9 billion payable to stockholders plus an equity retention program of up to about US$1.0 billion for employees, with closing expected in the first half of 2027 subject to regulatory approvals and other conditions. As of 15 October 2026 the acquisition has not closed. The company's own revenue, profit and headcount could not be confirmed from official material and are not disclosed. This profile uses only official material available by 15 October 2026 (the company's blog, documentation and pricing page, and NVIDIA's announcement and SEC filing).
- Hugging Face in 30 seconds, and where it stands in AI and machine learning
- The logo and how the name is written
- Where the company sits in AI and machine learning
- Scale
- Resources committed to AI and machine learning
- What it has done so far
- What comes next
- What the company is aiming for
- Partnership map
- Technologies, products and services
- The risks this company carries
- Glossary, references and claim-to-source audit
1. Hugging Face in 30 seconds, and where it stands in AI and machine learning
Hugging Face, SAS (Paris)Terms are with the U.S. corporation, governed by New York law; the main EU establishment is the Paris SAS
NVIDIA agreed to acquireDefinitive agreement on 2 September 2026; closing expected in the first half of 2027 (NVIDIA Form 8-K)
(15 October 2026)As shown on the site; NVIDIA's announcement says "3 million+ / 500,000", counted differently
(NVIDIA announcement)The company's own material says 13 million in 2025; years and definitions differ
Hugging Face is where other companies' models live. It is not itself competing at the model frontier; it runs a common place where models from many developers, such as Llama, DeepSeek, Mistral and NVIDIA, sit side by side, and it layers libraries, an inference service and enterprise features on top. NVIDIA's September 2026 announcement of the acquisition (about US$12.9 billion) therefore changes the picture: a company that controls AI compute would own the common place for open models. In its blog and Form 8-K NVIDIA promises that the platform will stay open, that NVIDIA compute will not be required, and that support for other silicon vendors will continue Sourced. These are the acquirer's commitments and pre-closing plans; how the platform will actually run after closing cannot be confirmed.
2. The logo and how the name is written
A note on naming: the company name is Hugging Face, and the logo is a smiling yellow face with open hands (resembling the "🤗" emoji) plus the company name. The contracting party in the terms is Hugging Face, Inc. (a Delaware corporation), and the main EU establishment is Hugging Face, SAS (Paris register number 822 168 043), per the privacy policy. NVIDIA's announcement says the company will keep "its same iconic 🤗 brand" after the acquisition. Products include the Hub for models and data, Spaces for apps, Inference Providers / Inference Endpoints for inference, Buckets for storage and libraries such as Transformers. The logo image is "hf-logo-with-title.svg", which is distributed on the official brand page.
Official site (brand): https://huggingface.co/brand3. Where the company sits in AI and machine learning
As Fig. 1 shows, Hugging Face is a platform & adopters type. It runs the Hub, where anyone can publish models, datasets and apps, and offers inference through "Inference Providers", which lets users call outside providers such as Cerebras, Groq, Together AI, fal and Replicate through one API (official documentation). Its open-source libraries (Transformers, Diffusers, PEFT, TRL and others, all Apache 2.0) have become standard tools for defining, fine-tuning and deploying models. The company's own analysis shows that use is concentrated in a few models (the top 200 models account for 49.6% of downloads, and half of models have under 200 downloads in total) and that Chinese models took about 41% of downloads in 2025 (company analysis). See also our explainers on post-training and fine-tuning, model efficiency, retrieval-augmented generation and context windows, and edge and on-device AI. Sourced
4. Scale
Hugging Face is unlisted and does not publish revenue, profit or headcount in official material. The indicators of scale that could be confirmed are platform usage, cumulative funding and the acquisition price. Sourced
| As of | Users | Models | Datasets | Other | Source |
|---|---|---|---|---|---|
| May 2022 | ― | 100,000 | 10,000 | Over 10,000 companies | Ref. 5 |
| April 2025 | Over 7 million | ― | ― | Over 500,000 Spaces | Ref. 4 |
| 2025 (year) | 13 million | Over 2 million | Over 500,000 | Over 30% of the Fortune 500 hold verified accounts | Ref. 3 |
| 3 September 2026 | Over 18 million | Over 3 million | 500,000 | 1 million apps; over 200,000 companies | Ref. 1 |
| 15 October 2026 | ― | 3,153,478 | 1,090,025 | Counts shown on the site's listing pages | Ref. 11, 12 |
The figures in the table come from the company and the acquirer, with different dates and definitions. For example, for datasets in September 2026 NVIDIA's announcement says 500,000 while the site listing shows about 1.09 million (the listing may include evaluation data and environments, which I did not check). This article therefore does not compute growth rates from them.
| Funding and acquisition | What could be confirmed | Source |
|---|---|---|
| Series C | US$100 million (9 May 2022), led by Lux Capital with major participation from Sequoia and Coatue | Ref. 5 |
| Total raised | "Over US$395 million", with investors including Google, Amazon, NVIDIA, Salesforce, Sequoia, Coatue, Lux Capital and Addition (company statement of April 2025). Round-by-round amounts and valuations are not in the company material I checked | Ref. 4 |
| Acquisition by NVIDIA | Definitive agreement on 2 September 2026. About US$11.9 billion payable to stockholders (subject to adjustments) and an equity retention program of up to about US$1.0 billion for employees; NVIDIA's blog states a total of US$12,930,300,000. Closing is expected in the first half of 2027, subject to regulatory approvals | Ref. 1, 2 |
| Revenue, profit, headcount | Not disclosed | Not confirmed |
The purchase price is the amount NVIDIA agreed to pay, not a measure of Hugging Face's revenue or profit. The acquisition has also not closed. NVIDIA's Form 8-K states that the price is subject to "adjustments" and that closing requires "regulatory approvals" and other conditions. Whether and when it closes is expected for the first half of 2027 and not settled Not yet confirmed.
5. Resources committed to AI and machine learning
For each of the four resources, the table shows what could and could not be confirmed.
| Resource | What primary information confirms | Type |
|---|---|---|
| People | Headcount is not disclosed. The founders are Clément Delangue (CEO), Julien Chaumond (CTO) and Thomas Wolf (Chief Science Officer). NVIDIA says it will set up an equity retention program of up to about US$1.0 billion for employees joining it. The company says robotics researcher Rémi Cadene (formerly of Tesla) joined in March 2024 | Sourced / headcount not disclosed |
| Money | Cumulative funding is over US$395 million, with a US$100 million Series C in 2022. R&D spend, capex and revenue are not disclosed. Prices are public: PRO US$9 a month, Team US$20 per user a month, Enterprise US$50 per user a month, and storage US$8-18 per TB a month (egress and CDN included) | Sourced / spending and revenue not disclosed |
| Investments, acquisitions, JVs, divestitures | Acquisitions and integrations: Gradio (21 December 2021), XetHub (8 August 2024), Pollen Robotics (14 April 2025, called "the fifth acquisition" by the company). Projects joining: Sentence Transformers (22 October 2025), GGML / llama.cpp (20 February 2026). On 22 September 2026 the creator of oMLX joined to support the MLX community. The company is now also the target of an acquisition (NVIDIA) | Sourced |
| Universities, research institutes, public funds | An official blog post (22 September 2026) describes UK AISI and EvalEval working to make benchmark results reproducible. In the EU, the French data protection authority (CNIL) is the lead authority (privacy policy). Public grants could not be confirmed | Sourced / grants not confirmed |
Next, the status of the main services and facilities. Figures follow the company's own wording.
| Service / facility | Details | Stage | Date | Source |
|---|---|---|---|---|
| Hub | About 3.15 million models and about 1.09 million datasets (site display). Storage Buckets offered as "new". The storage backend is Xet (a chapter in the official documentation) | Operating | 15 October 2026 | Ref. 10, 11, 12 |
| Inference Providers | Baseten, Cerebras, Cohere, DeepInfra, fal, Featherless AI, Fireworks, Groq, HF Inference, Novita, Nscale, OVHcloud, Public AI, Replicate, Scaleway, Together, WaveSpeedAI, Z.ai and others through one API (pay-as-you-go; PRO, Team and Enterprise include monthly credits) | Operating | 15 October 2026 | Ref. 13 |
| Spaces / ZeroGPU | The runtime for apps. ZeroGPU allocates NVIDIA RTX Pro 6000 Blackwell GPUs (up to 96 GB) dynamically (free tier available). Paid hardware starts at US$0.40 an hour for a T4, US$2.50 for an A100 (80 GB) and US$20.00 for eight A100s | Operating | 15 October 2026 | Ref. 10 |
| LeRobot / Reachy | An open-source robotics library (released May 2024) and Pollen Robotics' humanoid Reachy 2 (US$70,000). LeRobot has about 28,000 GitHub stars | Offered | April 2025, October 2026 | Ref. 4, 18 |
6. What it has done so far
| When | What happened | Meaning |
|---|---|---|
| 2018 | Open-sourced a PyTorch version of BERT (company statement) | Started from a library |
| 21 December 2021 | Acquired Gradio | Brought in a tool for building ML demos and apps (the base of Spaces) |
| 9 May 2022 | Series C (US$100 million, led by Lux Capital); at the time 100,000 models and 10,000 datasets | Described by the company as the "Home of Machine Learning" |
| March-May 2024 | A robotics researcher joined and LeRobot was released | Widening into robotics |
| 8 August 2024 | Acquired XetHub | Storage infrastructure (Xet) for huge models and data |
| 14 April 2025 | Acquired Pollen Robotics (the fifth acquisition) and offered the humanoid Reachy 2 | From software to hardware |
| 22 October 2025 | Sentence Transformers joined | The standard library for embedding models |
| 2025 (year) | 13 million users, over 2 million models and over 500,000 datasets (company analysis) | Usage roughly doubled |
| 20 February 2026 | GGML / llama.cpp joined; the company says it stays "100% open-source and community driven" | Brought in a base for local inference |
| 2-3 September 2026 | NVIDIA signed and announced a definitive agreement to acquire the company | Closing expected in the first half of 2027 |
| 22-28 September 2026 | The oMLX creator joined, work with UK AISI and EvalEval on reproducibility, and RL environments added to the Hub | Platform expansion continued after the acquisition announcement |
7. What comes next
first half of 2027Subject to regulatory approvals and other conditions (NVIDIA Form 8-K)
The above are the acquirer's plans and commitments, made before closing Not yet confirmed. NVIDIA's Form 8-K adds a risk factor saying that government restrictions may negatively impact the Hugging Face platform and NVIDIA's business (chapter 11).
Revenue, profit and headcount / round-by-round funding and valuations (official company material) / the status of regulatory review and any actual closing / post-deal governance (board, operational independence) and pricing / the company's own official statement on the acquisition (not found on the company's blog) / the difference in how datasets are counted. None of these is stated in the official material I checked.
8. What the company is aiming for in AI and machine learning
This separates what the company says directly from what we read into it. Inference marks our own reading of primary information.
| What can be read | Primary information behind it | Type |
|---|---|---|
| It aims to remain the open common place for AI | The company's statements: "the most used platform for AI builders", "keep future AI open" when GGML joined, and openness bringing trust in the Series C post | Sourced (company statement) |
| It is assembling standard tools from model definition to inference | The plan to combine Transformers and GGML / llama.cpp (model definition and local inference), Inference Providers and Xet storage | Sourced / Inference |
| It is widening into robotics (physical devices) | LeRobot, the Pollen Robotics acquisition, Reachy 2 and RL environments on the Hub | Sourced / Inference |
| Revenue seems centered on paid plans, inference and storage | The pricing page: PRO, Team, Enterprise, Inference Providers (pay-as-you-go), storage (per-TB price) and Spaces hardware (hourly price) | Sourced / Inference |
| Even under NVIDIA it asserts neutrality | NVIDIA's blog and Form 8-K: stays open, NVIDIA compute not required, other silicon supported | Sourced (acquirer's commitment) |
The basis for reading it as assembling standard tools including local inference is that the GGML announcement spells out a division of roles. The company writes that llama.cpp is the fundamental building block for local inference and transformers for model definition, and says it will make it almost "single-click" to ship new models from the transformers definition to llama.cpp. A September 2026 official blog post titled "Transformers now runs llama.cpp quants" suggests the plan is reaching the product. The company does not, however, describe this as a revenue strategy Not yet confirmed
9. Partnership map in AI and machine learning
As the figure shows, NVIDIA is both the acquirer and the largest contributor to the Hub. NVIDIA's blog says Hugging Face will remain free to choose "the clouds and inference service providers, and computing platforms" after the acquisition. On the other hand, Spaces' ZeroGPU runs on NVIDIA GPUs (RTX Pro 6000 Blackwell), so part of its compute already depends on NVIDIA (pricing page). Inference providers such as Cerebras and Groq include companies that use non-NVIDIA chips, but I did not check their configurations Sourced.
10. Technologies, products and services
Hugging Face is not in the business of ranking models itself; it provides a place where other companies' models sit. This article shows the company's products, prices and libraries within what its official material states.
| Item | Official statement | What could not be confirmed |
|---|---|---|
| Prices (checked 15 October 2026) | PRO: US$9 a month (individual; 10x private storage, US$2 of inference credits a month, 8x ZeroGPU quota). Team: US$20 per user a month (SSO, audit logs, storage regions). Enterprise: US$50 per user a month (SCIM, legal and compliance processes, dedicated support). Storage: base US$12/TB a month for public and US$18/TB for private repositories, with discounts up to 33% (US$8-12) at volume; the company compares with AWS S3 at US$23 and Backblaze at US$15 (company table) | Pricing changes after the acquisition |
| Open-source libraries (GitHub) | Transformers (about 168,000 stars), Diffusers (35,000), smolagents (30,000), LeRobot (28,000), Datasets (22,000), PEFT (22,000), TRL (20,000) and Text Generation Inference (11,000), all Apache 2.0 (retrieved 15 October 2026) | Actual user numbers for each library |
| Inference | Inference Providers (outside providers through one API), Inference Endpoints (dedicated) and HF Inference. A Responses API (beta) and integration guides for Codex, Claude Code and VS Code with GitHub Copilot | Performance and cost track records per provider |
| Safety and security | A malware scan runs on every repository file smaller than 2 GB at each commit; pickle scanning and secrets scanning are also provided (official documentation). The safetensors format is offered as well | Handling of files of 2 GB or more; detection record |
| Evaluation and community | Daily Papers, leaderboards, the Open TTS Leaderboard (30 September 2026), and reproducibility work with UK AISI and EvalEval. The official blog shows use concentrated in a few models (the top 200 models at 49.6%) | Operating criteria of the leaderboards |
Hugging Face's main costs appear to lie in storing and distributing very large files, models and data. Its pricing page quotes storage per TB and states that egress and CDN are included. For huge models (a trillion-parameter-class model can run to several TB for the weights alone Our calculation; because the same weights are uploaded and downloaded many times) the company acquired XetHub in 2024 and adopted Xet as its storage backend (a "Storage Backend (Xet)" chapter in the documentation). On formats, safetensors (a weight format that loads safely and quickly) and GGUF (a quantized format used by llama.cpp; Transformers added support for llama.cpp quants in September 2026) are shared formats for running models under memory and bandwidth limits. On compute, ZeroGPU allocates RTX Pro 6000 Blackwell GPUs (up to 96 GB of memory) on a shared basis, so the GPU's memory size caps the size of models that can run free Inference. See our explainers on model efficiency, inference optimization and serving, HBM, and edge and on-device AI.
11. The risks this company carries
The following is arranged neutrally, within what the official material of the company and the acquirer says.
| Risk | Details | Type |
|---|---|---|
| The acquisition is pending and uncertain | Closing is expected in the first half of 2027 subject to regulatory approvals and other conditions, and the price is subject to adjustments. Before closing the company appears to run independently, but post-deal organization, governance and pricing could not be confirmed | Sourced / Inference |
| Expectations of neutrality versus the acquirer's commitments | NVIDIA commits to "stay open", not to require NVIDIA compute and to support other silicon vendors, but these are pre-closing plans. NVIDIA is also a GPU supplier and the largest contributor to the Hub | Sourced (acquirer's commitment) |
| Government restrictions (acquirer's statement) | NVIDIA's Form 8-K adds a risk factor: "other parties are actively lobbying" governments for measures that would restrict or disadvantage open-source models, "many of the world's most popular and successful open-source models originated in China", and "any regulatory control or other restriction" limiting support for models derived from any region "could have a material impact" on the Hugging Face platform and on NVIDIA's business | Sourced (acquirer's statement) |
| Concentration of use | The top 200 models (0.01% of models) account for 49.6% of downloads and half of models have under 200 downloads in total (company analysis). In 2025 Chinese models were the largest share at about 41% of downloads (same) | Sourced |
| Malicious files and provenance | The malware scan covers files under 2 GB, and the documentation says a file may also show neither "ok" nor "infected" while scanning or after an error. Model weights also carry varied licenses (for example a conditional MIT or CC BY-NC), so users must check each model's terms | Sourced |
| Reliance on other companies' compute | Inference depends on outside providers (Inference Providers) and ZeroGPU runs on NVIDIA GPUs. The scale of Hugging Face's own compute could not be confirmed | Sourced / not confirmed |
| Limits of disclosure | Revenue, profit, headcount and round-by-round valuations are undisclosed. The company's and the acquirer's figures, such as dataset counts, use different definitions | Not yet confirmed |
This article summarises a company and is not legal, safety or procurement advice. Use of models and data on the Hub is subject to each repository's license and to national and regional law.
12. Glossary
- Hub
- Hugging Face's shared place for models, datasets, Spaces and Buckets.
- Spaces
- A place to run and share apps and demos.
- ZeroGPU
- A scheme in Spaces that allocates GPUs dynamically only when needed.
- Inference Providers
- A scheme that calls outside inference providers' model serving through one API.
- Transformers
- The standard library for defining and using models (Apache 2.0).
- safetensors
- A format that loads weights safely and quickly.
- GGUF
- A quantized model format used by llama.cpp.
- Xet
- A storage backend for large files that came from XetHub.
- Form 8-K
- A current report that listed U.S. companies file with the SEC; NVIDIA used it to report the agreement.
13. References
- NVIDIA Blog, "NVIDIA to Acquire Hugging Face" (Jensen Huang, 3 September 2026). The agreement, total of US$12,930,300,000, Hugging Face usage figures and commitments to open operation. https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/
- NVIDIA Form 8-K (filed 3 September 2026, for the definitive agreement of 2 September 2026). Purchase price of about US$11.9 billion, equity retention program of up to about US$1.0 billion, expected closing, and risk factor. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000078/nvda-20260902.htm
- Hugging Face Blog, "State of Open Source on Hugging Face: Spring 2026". 2025 usage figures, concentration of use, and share of Chinese models. https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026
- Hugging Face Blog, "Hugging Face to sell open-source robots thanks to Pollen Robotics acquisition" (14 April 2025). The fifth acquisition, Reachy 2, cumulative funding and founders. https://huggingface.co/blog/hugging-face-pollen-robotics-acquisition
- Hugging Face Blog, "We Raised $100 Million for Open & Collaborative Machine Learning" (9 May 2022). The Series C. https://huggingface.co/blog/series-c
- Hugging Face Blog, "XetHub is joining Hugging Face!" (8 August 2024). https://huggingface.co/blog/xethub-joins-hf
- Hugging Face Blog, "Gradio is joining Hugging Face!" (21 December 2021). https://huggingface.co/blog/gradio-joins-hf
- Hugging Face Blog, "Sentence Transformers is joining Hugging Face!" (22 October 2025). https://huggingface.co/blog/sentence-transformers-joins-hf
- Hugging Face Blog, "GGML and llama.cpp join HF to ensure the long-term progress of Local AI" (20 February 2026). https://huggingface.co/blog/ggml-joins-hf
- Hugging Face Pricing page (checked 15 October 2026). Plans, storage and Spaces hardware. https://huggingface.co/pricing
- Hugging Face Hub models listing page (checked 15 October 2026). Displayed count of public models. https://huggingface.co/models
- Hugging Face Hub datasets listing page (checked 15 October 2026). Displayed count of public datasets. https://huggingface.co/datasets
- Hugging Face Documentation, "Inference Providers". Provider list and mechanism. https://huggingface.co/docs/inference-providers/index
- Hugging Face Documentation, "Malware Scanning". Scope of scanning. https://huggingface.co/docs/hub/security-malware
- Hugging Face Terms of Service. Contracting party (Hugging Face, Inc., a Delaware corporation) and governing law. https://huggingface.co/terms-of-service
- Hugging Face Privacy Policy. Main EU establishment (Hugging Face, SAS, Paris) and CNIL. https://huggingface.co/privacy
- Hugging Face Official blog index (checked 15 October 2026). Recent announcements. https://huggingface.co/blog
- Hugging Face GitHub organization "huggingface" (retrieved 15 October 2026). Library stars and licenses; location listing. https://github.com/huggingface
- Hugging Face Brand page (logo source: hf-logo-with-title.svg). https://huggingface.co/brand
- Hugging Face Spaces listing page (checked 15 October 2026). https://huggingface.co/spaces
14. Claim-to-source audit
| Claim in the text | Type | Source |
|---|---|---|
| NVIDIA's agreement to acquire (2 September 2026), the purchase price of about US$11.9 billion, the equity retention program of up to about US$1.0 billion, expected closing (first half of 2027), and regulatory approvals as a condition | Sourced | Reference 1 https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/ Reference 2 https://www.sec.gov/Archives/edgar/data/1045810/000104581026000078/nvda-20260902.htm |
| Total of US$12,930,300,000, over 18 million users, over 3 million models, 500,000 datasets, 1 million apps, over 200,000 companies, and NVIDIA's contributions (over 500 models, over 250 datasets) | Sourced | Reference 1 https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/ (NVIDIA's announcement) |
| Commitments to open operation (NVIDIA compute not required; support for other silicon) | Sourced | Reference 1 https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/ Reference 2 https://www.sec.gov/Archives/edgar/data/1045810/000104581026000078/nvda-20260902.htm (acquirer's commitment) |
| The risk statement on government restrictions (open-source models, models that originated in China) | Sourced | Reference 2 https://www.sec.gov/Archives/edgar/data/1045810/000104581026000078/nvda-20260902.htm |
| 2025 usage (13 million, over 2 million, over 500,000), concentration of use (49.6%), Chinese models at about 41% of downloads, over 30% of the Fortune 500 | Sourced | Reference 3 https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026 (company analysis) |
| April 2025 users over 7 million, Spaces over 500,000, cumulative funding over US$395 million, founders, the fifth acquisition, Reachy 2 price, robotics timeline | Sourced | Reference 4 https://huggingface.co/blog/hugging-face-pollen-robotics-acquisition |
| The Series C (US$100 million, 9 May 2022, led by Lux Capital) and scale at the time | Sourced | Reference 5 https://huggingface.co/blog/series-c |
| The joining of XetHub, Gradio, Sentence Transformers and GGML / llama.cpp, and the dates | Sourced | Reference 6 https://huggingface.co/blog/xethub-joins-hf Reference 7 https://huggingface.co/blog/gradio-joins-hf Reference 8 https://huggingface.co/blog/sentence-transformers-joins-hf Reference 9 https://huggingface.co/blog/ggml-joins-hf |
| Prices (PRO, Team, Enterprise, storage, Spaces hardware, ZeroGPU) | Sourced | Reference 10 https://huggingface.co/pricing (checked 15 October 2026; may change) |
| About 3.15 million public models and about 1.09 million datasets (listing page display) | Sourced | Reference 11 https://huggingface.co/models Reference 12 https://huggingface.co/datasets |
| Inference Providers' providers and mechanism | Sourced | Reference 13 https://huggingface.co/docs/inference-providers/index |
| Malware scanning (files under 2 GB, at each commit) | Sourced | Reference 14 https://huggingface.co/docs/hub/security-malware |
| Legal entities (Hugging Face, Inc.; Hugging Face, SAS), governing law, CNIL, register number | Sourced | Reference 15 https://huggingface.co/terms-of-service Reference 16 https://huggingface.co/privacy |
| Library stars and licenses (Transformers, Diffusers, LeRobot and others) and the location listing | Sourced | Reference 18 https://github.com/huggingface |
| Logo image (hf-logo-with-title.svg from the official brand page) | Sourced | Reference 19 https://huggingface.co/brand |
| Recent announcements (UK AISI and EvalEval, Open TTS Leaderboard, RL environments, the oMLX creator, Transformers support for llama.cpp quants) | Sourced | Reference 17 https://huggingface.co/blog |
| Revenue, profit, headcount, round-by-round funding and valuations, the review status of the acquisition, post-deal governance and pricing, and the company's own statement on the acquisition | Not yet confirmed | Not stated in the official material checked. Not estimated in this article (items not confirmed) |
| Closing of the acquisition, the commitment to keep the platform open, NVIDIA's stated aims, and brand continuity | Not yet confirmed | Reference 1 https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/ Reference 2 https://www.sec.gov/Archives/edgar/data/1045810/000104581026000078/nvda-20260902.htm The acquirer's plans and commitments, not results |
| The readings on remaining the open common place, tools from model definition to inference, robotics, and the center of revenue | Inference | Reference 3 https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026 Reference 4 https://huggingface.co/blog/hugging-face-pollen-robotics-acquisition Reference 9 https://huggingface.co/blog/ggml-joins-hf Reference 10 https://huggingface.co/pricing our own interpretation |
| The explanation linking data volume, movement and memory to storage and formats, and the size of a huge model's weights | Not yet confirmed | our own explanation (general commentary, not a company disclosure) |
Last updated: 15 October 2026 / Troy Technical
The figures in this article come from Hugging Face's official blog, documentation, pricing page and terms, NVIDIA's announcement and Form 8-K, and public GitHub information. Model names, prices, contracts and corporate structure were checked on 15 October 2026 and may change. Market-research estimates and press-report-only figures are not used. This is not an assessment of the company's financial condition or of the relative merit of its products. Passages marked "Inference" are our own reading of primary information and are not statements by the company.