Hugging Face spent eight years building the one thing almost every AI lab, researcher, and enterprise now depends on and pays nothing for: the default place where open-source models get published, discovered, and downloaded. Then, in September 2026, Nvidia agreed to pay $12.93 billion to own that chokepoint outright.
That deal is still unclosed. A Justice Department antitrust inquiry confirmed on October 28, 2026 now sits between Hugging Face and Nvidia's expected first-half-2027 close — a reminder that owning the default distribution layer for open-source AI is exactly the kind of chokepoint regulators are trained to worry about. This deep dive covers how Hugging Face actually makes money, its funding history before the deal, its quieter push into physical AI through Pollen Robotics and LeRobot, and what the DOJ review means for the deal Nvidia itself has called its largest acquisition to date, according to its own announcement.
Company Snapshot
Figures compiled from TechCrunch, Nvidia's own announcement and 8-K filing, and Sacra estimates, as of November 2026. Hugging Face is privately held and does not publish audited financials.
Founders
Clément Delangue (CEO), Julien Chaumond (CTO), Thomas Wolf (Chief Science Officer)
HQ
New York, NY
Headcount
Roughly 650-770 by third-party workforce trackers; not disclosed by the company
Pending acquirer
Nvidia, agreed Sep 3, 2026
Key investors
Salesforce, Google, Amazon, Nvidia, Intel, AMD, Qualcomm, IBM, Lux Capital, Addition
Scale (Sep 2026)
3M+ models, 500K+ datasets, 1M Spaces, 18M+ users
Named users
Google, Meta, Microsoft, Bloomberg, Grammarly, most major AI research labs
Sector
AI Infrastructure / Open-Source Machine Learning
What Hugging Face Actually Does
Hugging Face runs the Hub: a free repository where anyone can publish, search, and download AI models, datasets, and demo applications. Meta publishes Llama there, Google publishes Gemma, Stability AI publishes its image models — and researchers, students, and companies pull them down without paying a cent. The company's open-source Transformers library, first released in 2018, became the standard way developers load and fine-tune those models in code, which is how Hugging Face ended up as infrastructure for the entire field rather than just one more AI startup.
The company did not start this way. It launched in New York in 2016 as a chatbot app aimed at teenagers. The team kept open-sourcing pieces of the natural-language-processing tooling they built along the way, the reaction from other developers was strong enough that the side project became the product, and by 2018 Hugging Face had abandoned the chatbot entirely to build the open model hub it's known for today.
How Hugging Face Makes Money
Inference Endpoints
Customers pick any model from the Hub, choose a GPU tier starting at $0.50/hour for a T4, and Hugging Face deploys it as a managed, autoscaling production API — billed by compute time, not seats.
Enterprise Hub
$50 per user per month (a lighter Team plan runs $20/user/month) for private repositories, SSO, granular access controls, audit logs, and higher-rate-limit inference, sold to organizations like Google and Bloomberg.
Pro subscriptions
$9 per month for individual developers who want persistent Spaces, higher API rate limits, and early access to new features on top of the free public hub.
Training and consulting
Paid engagements helping large enterprises fine-tune and deploy open-source models at scale, the smallest and least standardized of the four revenue lines.
The free hub is the growth engine, not a loss leader to apologize for. Every model a lab uploads makes the platform more valuable to every other visitor, and that network effect produces the 18 million users and 200,000-plus companies Nvidia cited when it announced its acquisition offer in September 2026 — a tiny fraction of whom convert to one of the four paid products above, which is still enough to produce roughly $150 million in annualized revenue.
Hugging Face's Estimated Revenue Mix, 2026
Inference Endpoints and Enterprise Hub together account for roughly three-quarters of revenue
Value Add VC estimate based on Hugging Face's public product lineup; the company does not disclose revenue by product line.
Data
| Item | Value |
|---|---|
| Inference Endpoints | ~40% |
| Enterprise Hub | ~35% |
| Pro subscriptions | ~15% |
| Training & consulting | ~10% |
Funding History
Hugging Face raised about $395 million across four disclosed rounds in seven years before Nvidia's acquisition offer, with nearly every major cloud and chip company eventually becoming an investor rather than risk the default open-source hub tilting toward a rival.
| Round / Event | Date | Amount | Lead Investor(s) | Valuation |
|---|---|---|---|---|
| Series B | Mar 11, 2021 | $40M | Addition | n/d |
| Series C | May 9, 2022 | $100M | Lux Capital | $2B |
| Series D | Aug 24, 2023 | $235M | Salesforce Ventures | $4.5B |
| Nvidia acquisition (pending) | Sep 3, 2026 | $12.93B total | Nvidia | $12.93B |
Sources: TechCrunch, Series B, Mar 11, 2021, SiliconANGLE, Series C, May 9, 2022, TechCrunch, Series D, Aug 24, 2023, and Nvidia's 8-K, Sep 2, 2026.
Hugging Face Valuation Timeline, 2021-2026 ($B, log scale of milestones)
Hugging Face went from a $40M Series B with no disclosed valuation to a $12.93B acquisition price in five and a half years
TechCrunch, SiliconANGLE, and Nvidia's 8-K filing, 2021-2026
Data
| Item | Value |
|---|---|
| Series B (Mar '21) | $40M raised |
| Series C (May '22) | $2B valuation |
| Series D (Aug '23) | $4.5B valuation |
| Nvidia deal (Sep '26) | $12.93B deal price |
The investor roster tells its own story: Salesforce, Google, Amazon, Nvidia, Intel, AMD, and Qualcomm all bought into the 2023 Series D simultaneously, an unusual lineup of chip and cloud rivals who otherwise compete fiercely. Each had the same reason — keeping the default open-source AI hub neutral and interoperable with their own infrastructure, rather than letting it become captive to one of their competitors. That is also precisely the dynamic the DOJ is now examining in reverse: whether Nvidia, having lost the argument for staying a minority investor, buying the whole thing outright hands it the leverage those other 2023 investors paid to prevent.
Product Portfolio
The Hub (core)
A free repository of 3M+ models, 500K+ datasets, and 1M Spaces (hosted demo apps), searchable and downloadable by anyone.
Transformers library
The open-source Python library, first released in 2018, that became the standard interface for loading and fine-tuning machine learning models in code.
Inference Endpoints
Managed, autoscaling GPU hosting that turns any Hub model into a production API, billed by compute time starting at $0.50/hour.
Enterprise Hub
Private repositories, SSO, access controls, and audit logs for organizations, at $20-$50 per user per month depending on tier.
LeRobot and Pollen Robotics
An open-source robotics stack built around Pollen Robotics' Reachy platform, acquired in April 2025, extended in May 2026 with a $2,500 3D-printable humanoid robot, LeRobot Humanoid.
AutoTrain and Spaces
No-code fine-tuning tools and a hosting layer for interactive ML demo apps, used by both individual developers and the research community to showcase models publicly.
Inside Nvidia's $12.93B Bet — and the Antitrust Question It Raised
Nvidia's offer, confirmed September 3, 2026, structures as roughly $11.9 billion payable to Hugging Face shareholders plus up to $1 billion in retention equity for employees who join Nvidia, per Nvidia's own 8-K filing. CEO Clem Delangue said all staff are staying and there will be no changes to the Hugging Face team, and Nvidia CEO Jensen Huang pledged the platform stays "open" — developers can keep choosing their own models, frameworks, clouds, and chips, and Nvidia compute will not be required to build on or deploy through Hugging Face.
That openness pledge exists because the deal's biggest risk is exactly the opposite outcome. Reuters reported on October 28, 2026 that the Justice Department is asking how the acquisition would affect AI competition — specifically, whether owning Hugging Face would let Nvidia control a key distribution channel for AI models and use that position against rivals' chips and clouds, according to reporting on the inquiry. The questions sit inside a broader, multiyear DOJ review of Nvidia's conduct that also covers its December 2025 SchedMD purchase and its roughly $20 billion Groq licensing-and-talent arrangement — meaning Hugging Face is now one data point in a pattern regulators are already tracking, not an isolated review.
Context explains why Nvidia moved at all. The company reportedly offered about $500 million for a stake in Hugging Face in 2025 at a $7 billion valuation, which Hugging Face turned down, according to the Financial Times. CNBC reported on September 28, 2026 that AMD and Salesforce had separately explored acquiring the company, and that OpenAI had proposed a $100 million investment, before Nvidia's $12.93 billion offer closed out the process in early September. Losing the company to a rival chipmaker or an AI lab it supplies would have cost Nvidia far more than the premium it ultimately paid.
No formal challenge or second request had been publicly detailed as of the October 28 report, and a DOJ spokesperson said government investigations "aren't evidence of wrongdoing." Nvidia still expects the deal to close in the first half of 2027. But a staff-led antitrust inquiry at the DOJ can run for years without resolving, and the gap between "agreed" and "closed" is exactly where this acquisition now sits.
Revenue and Key Metrics
End-2023 ARR
~$35M
Jun 2026 ARR
~$100M
Aug 2026 ARR
~$150M
up 50% in 2 months
Models on Hub
3M+
500K+ datasets, 1M Spaces
Hugging Face has never disclosed audited financials, but Delangue told TechCrunch in August 2026 that the company was "close to profitability" — a nearly identical claim to one he made in May 2024, meaning the company has described itself as near-breakeven for at least two years without confirming it has crossed the line. He also said the company had only begun spending, in 2026, the $235 million it raised back in 2023, a sign of how capital-light the business has run relative to most AI infrastructure companies.
Headcount estimates vary widely by tracker — Revelio Labs puts it at roughly 650 as of December 2025, while other workforce-data firms report figures as high as 1,000-plus, likely conflating staff with the much larger contractor and community base. The company has not published an official headcount.
Competitive Landscape
Hugging Face occupies a position no single rival fully matches: part model registry, part inference platform, part collaboration tool. The closer comparisons are increasingly companies that got absorbed into bigger balance sheets rather than independent peers — a pattern Hugging Face itself is now part of.
Hugging Face vs Together AI vs Replicate: Valuation and Independence
Nvidia 8-K, Dealroom/The Information, and Cloudflare's press release, 2025-2026
Replicate's deal size with Cloudflare was not disclosed; shown at $0 to indicate no public figure, not zero value.
Together AI is the closest remaining independent peer, having raised $800 million at an $8.3 billion valuation in July 2026 on roughly $100 million in annualized revenue, led by Aramco Ventures with Nvidia also participating. The two companies are more complementary than competitive in practice — Hugging Face is the discovery and community layer where models get published, while Together AI focuses on production inference speed and cost — but both now count Nvidia as a direct financial stakeholder in the open-model ecosystem. Replicate, a popular model-inference API, lost its independence first: Cloudflare agreed to acquire it in November 2025, folding its 50,000-plus model catalog into Cloudflare's Workers AI rather than continuing to compete as a standalone hub.
Leadership Team
Clément "Clem" Delangue — Co-Founder & CEO
Co-founded Hugging Face in New York in 2016 and has been its public face through the pivot from chatbot app to open-source AI hub, two separate "close to profitability" claims (2024 and 2026), and the negotiation of the Nvidia acquisition. He has publicly voiced caution about an AI funding bubble even while running a company whose own valuation rose nearly 3x in three years.
Julien Chaumond — Co-Founder & CTO
Met Delangue in 2016 and co-founded the company with him, overseeing the technical build-out of the Hub and its infrastructure as it scaled from a chatbot backend to the platform serving 18 million-plus users.
Thomas Wolf — Co-Founder & Chief Science Officer
Joined as the third co-founder and leads Hugging Face's research direction, including the open-source Transformers library that became the company's technical anchor in the AI research community.
Bull Case / Bear Case
The Bull Case
- +ARR nearly doubled in two months (Jun to Aug 2026), with Enterprise Hub and Inference Endpoints adoption accelerating as more companies move from experimenting with open models to running them in production.
- +The free Hub's network effects compound without Hugging Face spending on content creation — every model a lab publishes makes the platform more valuable to the next visitor.
- +Nvidia's $12.93 billion offer, if it closes, gives the company the chip maker's balance sheet and distribution while CEO Delangue says the team and product stay unchanged.
- +A near-universal strategic investor base (Salesforce, Google, Amazon, Nvidia, Intel, AMD, Qualcomm, IBM) signals the industry needs this hub to stay neutral and well-funded.
- +The LeRobot and Pollen Robotics push gives Hugging Face an early foothold in physical AI and robotics, a second growth vector beyond the core software hub.
The Bear Case
- -A confirmed DOJ antitrust inquiry now sits between the signed deal and its first-half-2027 expected close, and staff-led federal investigations can run for years without resolving.
- -Hugging Face has claimed to be 'close to profitability' twice, two years apart, without ever confirming it crossed the line — a pattern that leaves its actual margin structure opaque.
- -The company's entire monetization model depends on converting a small fraction of a free user base; if DOJ scrutiny forces behavioral commitments or a divestiture, that funnel could be disrupted.
- -Nvidia's own openness pledge is the thing regulators are specifically testing — the deal's central risk is that owning the default model-distribution layer gives one chip maker leverage every other chip and cloud company spent 2023 trying to prevent.
- -If the deal is delayed or blocked, Hugging Face has no disclosed fallback financing plan or restated independent valuation to fall back on.
IPO Outlook
Hugging Face is not on an IPO track. Instead of filing an S-1, the company agreed to a full acquisition by Nvidia — a far more common exit path for infrastructure companies this strategically important to a single dominant buyer than a public listing would have been. No underwriters, no roadshow, and no IPO timeline have ever been reported for Hugging Face as an independent company.
The one scenario that revives an IPO conversation is regulatory failure: if the DOJ's inquiry escalates into a formal challenge that blocks or unwinds the Nvidia deal, Hugging Face would be back to square one as a private company that last priced at $4.5 billion in 2023, now generating roughly $150 million in ARR — numbers that would support a very different, and likely higher, independent valuation than its last round, but with no public-market process currently underway to test it.
Bottom line: Hugging Face built the free infrastructure the entire open-source AI community runs on, and it worked well enough that four years of industry giants bankrolling its independence ended with one of them buying it outright for $12.93 billion. The business itself is in good shape — ARR nearly doubled in two months, the product lineup is diversifying into robotics, and the team is staying intact under Nvidia's ownership pledge. What's unresolved isn't the company's performance; it's whether regulators let the deal that values that performance actually close. Until the DOJ's inquiry resolves one way or the other, Hugging Face remains, on paper, an independent company worth $4.5 billion and, on a signed term sheet, a $12.93 billion acquisition waiting to complete.
All financial figures are based on publicly reported data, Nvidia's own disclosures, and credible media reporting as of November 2026. Hugging Face is a private company and does not publicly disclose audited financials; the Nvidia acquisition is a signed agreement, not a closed transaction, as of publication.
Latest from the Pulse
Get VC data most people never see
— free to subscribe
Trace's notes on venture, AI and startups, a few times a week. Join 5,000+ subscribers. No spam.
No spam, unsubscribe anytime. Sent via Substack · Privacy · or subscribe on Substack ↗