Signals Inbox·August 25, 2026·AI Infrastructure

Is Hugging Face really worth $13B today?

Hugging Face looks expensive at $13 billion on today’s revenue, but the price becomes more credible if a buyer is paying for control of the open-AI distribution layer rather than the business as it stands today.

Q1What does Hugging Face’s $13B number actually mean?

Hugging Face’s $13 billion figure currently describes a possible acquisition price, not a completed funding valuation.

The latest reporting says Hugging Face has received acquisition approaches at $13 billion or more and has been speaking with banks to evaluate bids. No buyer has been announced and no transaction has closed. So when we ask whether Hugging Face is “worth $13B,” we are really testing the price someone might pay to own the company today.

The last clean financing benchmark is much lower. Hugging Face raised $235 million in 2023 at a $4.5 billion post-money valuation, led by Salesforce Ventures with Google, Amazon, Nvidia, AMD, Intel, Qualcomm, IBM and Sound Ventures also participating. One year earlier, the company had been valued at $2 billion.

A $13 billion sale would therefore represent almost 2.9x Hugging Face’s 2023 valuation in roughly three years.

There is an even fresher benchmark. Earlier this year, the Financial Times reported that Hugging Face rejected an Nvidia proposal to invest $500 million at a $7 billion valuation. The acquisition figure now being discussed is about 86% higher.

Revenue has obviously not jumped 86% in that short period based on anything publicly disclosed. The gap strongly suggests that a potential buyer is assigning a large premium to control of Hugging Face’s platform, community and position in open AI.

Q2How much money is Hugging Face actually making now?

Hugging Face currently generates more than $100 million in annual recurring revenue, putting the proposed acquisition price at roughly 130 times the publicly disclosed revenue floor.

This is one of the stronger numbers in the analysis because it comes directly from CEO Clément Delangue. In a recent a16z interview, Hugging Face was described as having surpassed $100 million in ARR.

We also have a useful historical anchor. Around the 2023 financing, The Information reported annualized revenue of about $30 million, while Forbes put the run rate somewhere between $30 million and $50 million. Those were outside estimates, so we should give the newer company disclosure more weight.

The available evidence suggests that Hugging Face’s annual revenue base has grown roughly two to three times since the last financing. Healthy growth, yes. It does not resemble the revenue explosion we are currently seeing at the fastest-growing AI inference companies.

The multiple also needs one small qualification. Because Delangue said revenue had surpassed the threshold rather than giving an exact figure, 130x is effectively the upper end of the multiple based on what we know. If the real run rate is meaningfully higher, the multiple comes down.

Even with that adjustment, we are dealing with an unusually expensive company.

Q3Is Hugging Face’s current revenue multiple absurd?

Hugging Face’s implied revenue multiple looks extreme even next to today’s most expensive listed infrastructure software companies.

Cloudflare is probably the hardest public benchmark to beat because investors already pay an exceptional premium for its growth and strategic position. Based on recent market capitalization and trailing revenue, Cloudflare trades at roughly 42 times sales while growing revenue around 34% year over year.

Snowflake is closer to 23 times sales with roughly 31% trailing revenue growth. Datadog is around 21 times with roughly 32% growth.

Hugging Face deserves some premium over those companies if investors believe its smaller revenue base can compound much faster. Private companies can also make product bets without the quarterly pressure of public markets.

But the size of the gap is hard to explain through growth alone. Hugging Face’s implied multiple is more than three times Cloudflare’s and roughly six times Snowflake’s or Datadog’s. Meanwhile, Hugging Face has not disclosed a current revenue growth rate that clearly exceeds all three by anything close to that magnitude.

A normal “private company grows faster” argument does not cover the difference. The valuation only starts to make sense if we give substantial value to Hugging Face’s community, distribution and future monetization options.

Hugging Face versus public infrastructure software

Q4Are investors paying anything close to this for other AI infrastructure startups?

Private AI infrastructure is extremely expensive these days, but the closest recent deals still make Hugging Face look unusually rich.

Fireworks gives us the cleanest comparison. The company recently raised $1.505 billion at a $17.5 billion valuation and said it had passed $1 billion in annualized revenue, up fivefold year over year. That works out to less than 17.5 times revenue.

Baseten reached a $13 billion valuation in its latest financing. The company said revenue had increased twentyfold over the previous year, while Sacra estimates that annualized revenue reached roughly $600 million. If that estimate is close, investors paid around 22 times revenue.

OpenRouter is more interesting because Stripe has just agreed to buy it for more than $8 billion. The Financial Times reported annualized revenue of about $140 million, implying a takeout multiple around 57 times. OpenRouter also says inference volume has increased by at least 10x every year since it was founded.

Private AI infrastructure can command multiples that would look ridiculous in ordinary software. But even the OpenRouter acquisition, where Stripe is clearly paying for strategic positioning as well as revenue, lands at less than half Hugging Face’s implied multiple.

Hugging Face versus recent private AI infrastructure deals

Q5Is Hugging Face growing fast enough to justify the price?

Hugging Face is growing very quickly as an ecosystem, but the public data does not show the kind of revenue acceleration that would normally justify a 100x-plus sales multiple.

At the time of the 2023 financing, Hugging Face hosted roughly 500,000 models, 250,000 datasets and 250,000 applications. Its summer 2026 ecosystem report counted 2.96 million public model repositories, one million datasets and 1.44 million Spaces.

That means models increased almost sixfold, datasets roughly fourfold and Spaces almost sixfold in three years.

The expansion is still happening now. Between the start of 2026 and the company’s summer measurement, public model repositories rose from 2.43 million to 2.96 million, datasets from 711,000 to one million, and Spaces from one million to 1.44 million. Those are increases of roughly 22%, 41% and 44% respectively in only part of a year.

As seen above, revenue moved from roughly $30 million to $50 million of annualized revenue around the 2023 financing to above $100 million today. Depending on which historical estimate we use, that is roughly 2x to 3.3x growth.

The platform itself has expanded much faster than the money Hugging Face extracts from it. That’s the catch.

That gap can become a huge advantage if Hugging Face has deliberately prioritized network growth and can monetize the ecosystem later. It becomes a problem if free activity keeps compounding while the valuable production workloads end up generating revenue for somebody else.

Q6Is open-source AI actually winning enough business for Hugging Face?

Open-source AI is winning a lot of usage today, although enterprise spending remains much more mixed than Hugging Face’s developer momentum might suggest.

The usage side is strong. TechCrunch recently reported that Chinese open-weight models accounted for 41% of downloads on Hugging Face during the spring. On Vercel, open models handled nearly one-third of AI requests in June. Hugging Face CEO Clément Delangue also says companies increasingly move toward open models once the cost of running large proprietary APIs at scale becomes painful.

Broader surveys point in the same direction. Linux Foundation research found that 63% of companies use an open model and that 89% of organizations already using AI have some open-source technology in their AI infrastructure. A McKinsey survey found that 60% of respondents saw lower implementation costs from open-source AI than from proprietary alternatives.

The budget data is less bullish. Menlo Ventures estimated that enterprise generative-AI spending reached $37 billion in 2025, more than tripling in one year. Yet open-source models captured only about 11% of enterprise LLM spending in its survey, down from 19% the year before.

Those numbers can coexist. Companies may use open-source tools throughout their stack while still sending a large share of model spending to OpenAI, Anthropic and Google.

For Hugging Face, that distinction is crucial. The market around open AI is clearly big enough. We still need to see whether Hugging Face can turn widespread technical adoption into a much larger share of enterprise budgets.

Q7Has Hugging Face really become the default home for open AI?

Hugging Face is still the closest thing open AI has to a default distribution layer today.

Hugging Face reached 13 million users in 2025, and Delangue recently said a new repository is now created on the platform roughly every seven seconds. The Hub holds nearly three million public models and around one million datasets.

The companies using it matter just as much as the raw count. Meta distributes Llama models through Hugging Face. Nvidia publishes models and robotics work there. Google, Microsoft, OpenAI, DeepSeek, Alibaba’s Qwen team and many other important AI developers maintain active repositories.

Delangue also says roughly half of Fortune 500 companies use Hugging Face to deploy private or open models. We should treat that as a company claim rather than audited customer data, but it gives us some sense of how far the Hub has moved beyond hobbyist model sharing.

The clever part is that Hugging Face does not need one particular model company to win. If developers move from Meta models to Qwen, from large language models to image generation, or from cloud AI to models running locally, Hugging Face can still sit in the middle.

That is a much better position than betting the company on one model family.

Q8What can Hugging Face do that competitors cannot easily copy?

Hugging Face has a real network-effect moat, although its own latest usage data shows that the moat is concentrated in a relatively small part of the Hub.

Model developers publish on Hugging Face because developers already look there. Developers look there because important models appear there first. Frameworks and tools then integrate the Hub because that is where the models are. Over time, the workflow becomes familiar enough that moving the community somewhere else gets harder.

Still, we should not confuse millions of repositories with millions of equally valuable assets.

Hugging Face’s summer analysis found that 85.6% of models have fewer than 200 lifetime downloads. Just 1.5% of repositories generate 99.2% of all downloads. A competitor therefore would not need to recreate every obscure repository to attack the economically important part of the network.

Model files are also portable. Developers can download weights, host them themselves or move them onto AWS, Azure, Google Cloud or specialist inference platforms. Hugging Face does not have the kind of hard lock-in that comes from proprietary data formats or deeply embedded transaction systems.

The defensibility comes from habit, discovery, collaboration, integrations and community density. That can be extremely durable. GitHub proved as much in software development. But it gives Hugging Face more power over where AI developers congregate than over where every dollar of AI infrastructure spending eventually goes.

Q9Can Hugging Face make much more money without ruining what made it popular?

Hugging Face has plenty of room to make more money from enterprise infrastructure while keeping the open Hub easy to use, and its recent product moves show that this is already happening.

Storage is becoming one of the clearest examples. Hugging Face now advertises large-scale AI storage at roughly $8 to $12 per terabyte per month at volume, versus a listed $23 for AWS S3, with CDN and egress included. Its Xet technology also deduplicates model and dataset files at the chunk level, which is particularly useful when companies store many versions of huge models.

The first serious commercial proof arrived recently. Arcee AI signed a multi-million-dollar agreement to move its public and private models, datasets and agent traces onto Hugging Face, replacing AWS S3 for that workload.

That is a much bigger deal than selling a few extra enterprise seats. If AI labs begin treating Hugging Face as the system of record for private model assets, storage can turn community distribution into recurring infrastructure revenue.

Hugging Face is still being careful about extraction. Its Inference Providers product routes users to more than 200 models from outside providers and explicitly says Hugging Face adds no markup to the provider price. That makes the platform more attractive, while leaving much of the inference economics with companies such as Fireworks, Baseten and others.

We would watch storage, private repositories, governance, dedicated endpoints and enterprise contracts much more closely than consumer subscription pricing. Those products can raise revenue per organization without putting a paywall around the community that made Hugging Face valuable.

Q10Why did Hugging Face turn down Nvidia at a $7B valuation?

Hugging Face turned down Nvidia’s $500 million proposal at a $7 billion valuation because the company had little financial pressure and a lot to lose from looking controlled by one AI giant.

The Financial Times reported that Hugging Face was profitable in 2025 and still held roughly half of the approximately $400 million it had raised. Delangue has since said the company only recently began touching capital raised three years earlier and remains close to profitability.

That gave Hugging Face something most AI startups do not have: the ability to say no to Nvidia.

Strategic independence also has real economic value here. Hugging Face’s 2023 round included Nvidia and AMD, Google and Amazon, Intel, Qualcomm, IBM and Salesforce. Several of those companies compete directly with one another.

A platform where Nvidia, AMD, Google, Amazon and independent AI labs all feel comfortable publishing models is more valuable than a platform perceived as belonging to one of them.

The rejected proposal tells us more than a financing price normally would. Hugging Face had a credible investor willing to mark the company much higher than its previous round, yet management still preferred independence.

That strengthens the case for a premium valuation. It also makes an outright acquisition harder, because whoever buys Hugging Face would need to convince the rest of the ecosystem that the platform would remain genuinely neutral.

Q11Could a buyer justify paying more for Hugging Face than investors would?

A strategic buyer could rationally pay more for Hugging Face than a financial investor would, and GitHub gives us a useful precedent for how that can work.

Microsoft paid $7.5 billion for GitHub in 2018 when GitHub had 28 million developers. Contemporary estimates put GitHub’s ARR at roughly $200 million to $300 million, meaning Microsoft paid somewhere around 25 to 38 times revenue.

Four years later, Microsoft disclosed that GitHub had passed $1 billion of ARR and 90 million users. GitHub also became a central part of Microsoft’s developer strategy, eventually giving the company distribution for products such as GitHub Copilot.

Hugging Face has a similar strategic feature: its importance to developers is larger than the revenue directly captured from those developers.

The analogy only goes so far. Microsoft’s GitHub purchase was expensive, but the multiple was nowhere near Hugging Face’s current implied level.

The very recent OpenRouter deal moves the benchmark closer. Stripe is paying more than $8 billion for a model-routing platform whose reported annualized revenue is around $140 million. Stripe clearly sees strategic value in sitting between developers and many AI models rather than simply buying a stream of current profits.

There are two useful precedents for paying ahead of revenue: GitHub shows the long-term developer-platform logic, while OpenRouter shows how aggressive AI infrastructure takeout multiples have become right now.

Hugging Face would still be the more extreme bet.

Q12What revenue would Hugging Face need for $13B to look normal?

Hugging Face needs roughly $430 million to $1.3 billion in annual revenue for a $13 billion valuation to fit within a 10x to 30x revenue range.

Thirty times revenue would still be a very expensive valuation by normal public-market standards. At that multiple, Hugging Face would need around $433 million in annual revenue, more than four times its currently disclosed revenue floor.

At 20x, the requirement rises to $650 million. At 15x, it reaches roughly $867 million. A 10x multiple requires $1.3 billion.

This is probably the cleanest way to frame what investors or a buyer are underwriting. Hugging Face does not need to become the next AWS for the valuation to work, but it does need several hundred million dollars of additional monetization.

If revenue can compound around 50% annually for several years, the upper end of these thresholds becomes reachable. Public evidence has not yet established that Hugging Face is growing revenue at that pace today.

Revenue required for a $13B valuation

Q13What has to go right for Hugging Face to be worth $13B?

Hugging Face can justify the price if it turns its open-model distribution advantage into the default enterprise system for storing, governing and deploying AI assets.

The bull case cannot just be “more people will download models.”

Hugging Face needs the free Hub to keep attracting model developers while enterprises gradually move their private models, datasets, evaluation traces and internal AI workflows onto the same infrastructure. The recent Arcee agreement gives us one concrete example of that transition already happening.

The product expansion also fits this strategy. Hugging Face acquired XetHub to improve large-model storage, built out private and enterprise infrastructure, expanded model routing and inference products, and acquired Pollen Robotics after developing the LeRobot ecosystem.

Those moves give Hugging Face exposure to several versions of the AI market at once: cloud inference, local models, enterprise customization, datasets and eventually robotics.

A multi-model world would help enormously. The more companies mix OpenAI, Anthropic, Qwen, Llama, DeepSeek and specialized internal models, the more useful a neutral platform for finding, storing and managing those assets becomes.

If Hugging Face can turn that position into several hundred million dollars of recurring enterprise revenue while remaining the place developers naturally publish open models, the current acquisition price stops looking crazy surprisingly quickly.

Q14What would make the $13B price look obviously wrong?

Hugging Face looks badly overpriced if its community stays huge while the expensive production workloads keep flowing to inference specialists and hyperscalers.

That is the bear case we take most seriously. Pretty simple.

Fireworks and Baseten are currently converting open-model demand into production revenue much faster. Cloud providers already control enormous enterprise budgets. Hugging Face itself routes inference requests to outside providers without adding a markup, which is excellent for neutrality but much less exciting for value capture.

The enterprise market also remains more proprietary than developer activity might make it appear. Menlo’s research found that open-source models captured only 11% of enterprise LLM spending in 2025 despite much broader open-source usage elsewhere in the AI stack.

Hugging Face could therefore become indispensable without becoming proportionately profitable. Plenty of open-source infrastructure has created far more economic value for users than for the company maintaining it.

The concentration inside the Hub adds another risk. Most repositories receive almost no downloads, while a tiny percentage accounts for nearly all activity. If major model publishers increasingly distribute directly through their own APIs, clouds or other gateways, the valuable part of the network could prove easier to attack than the headline repository count suggests.

A buyer also faces a uniquely awkward problem: paying billions to control Hugging Face could weaken the neutrality that helped create the premium in the first place.

If monetization stays modest while competitors capture inference, compute and enterprise deployment spending, this valuation will be very difficult to defend.

Q15So, is Hugging Face really worth $13B today?

Hugging Face looks stretched at $13 billion as a standalone financial valuation, but a strategic buyer can make the number work if it is paying for control of the open-AI distribution layer.

The financial comparisons are harsh. Public infrastructure leaders currently trade around 20 to 40 times revenue. The recent private AI infrastructure benchmarks we examined range from roughly 17 times to the high-50s. Hugging Face sits far above that range based on its disclosed revenue base.

We would not pay this price simply because Hugging Face is growing or because open-source AI is popular.

The premium comes from something harder to reproduce. Hugging Face has become the common meeting point for model creators, developers, enterprises and competing AI infrastructure companies. Its ecosystem is still expanding quickly, open models are gaining real usage, and the company has managed to build that position without burning through the enormous amounts of capital required by frontier-model labs.

There is also a credible path toward much stronger monetization. Private AI storage, enterprise governance, dedicated infrastructure and model management can turn the Hub from a popular distribution platform into something companies depend on for production AI.

The evidence still leaves a large gap between that future and the business today.

Our judgment is fairly sharp: $13 billion is an aggressive but believable strategic acquisition price, while it remains clearly stretched as a standalone valuation based on current financial performance.

For the price to age well, Hugging Face needs to preserve its neutrality, keep its position as the default open-model hub and push annual revenue into the several-hundred-million-dollar range without damaging the free ecosystem.

If that happens, a buyer may eventually look smart for paying ahead of the numbers. If the community keeps growing while other companies capture most of the production spending, the same $13 billion will look like a very expensive way to own the internet’s favorite AI repository.