Hugging Face 传出要卖 130 亿：现金流会骗人，位置不会Hugging Face Is Reportedly Exploring a $13B Sale: Cash Flow Lies, Position Doesn't

The setup

Hugging Face is reportedly exploring a sale, asking $13 billion. The Lakers just traded at $12.5 billion, only 17 months after the last time they changed hands. The buyers: a legendary Disney CEO, and the younger brother of Trump's son-in-law.

Read the two together: both are assets, and neither one is priced on its ability to make money.

Why rich people keep buying sports teams

Run the Lakers on profit and it takes decades to earn the price back. Profit isn't what the buyer is looking at. What they're buying is a franchise: scarcity itself, and the fact that it doesn't break in a downturn. Sport has become an asset class, in the US and globally. Ten years ago that sentence wouldn't have been true. And historically, teams hold up extraordinarily well through recessions; their values don't fall off a cliff. Zoom out fifty years and you can count on your fingers the teams that sold for less than they cost. That's what scarcity looks like on a chart.

Over those eighteen years the S&P 500 returned 3.17x with dividends. Mean franchise value: MLS 15.65x, NBA 10.79x, MLB 7.08x, NHL 6.32x, NFL 6.10x, and not one league lost to the index.

There are 32 NFL teams and there will not be more. There are new billionaires every year. The NBA has 30, and Adam Silver is actively pushing expansion, and two more is the likely outcome. But expansion is rare, especially across North America's big four.

Nail down what's actually known

As of today, not one named buyer. All we know is that HF hired a bank to test interest. The thing worth reading is the 2023 cap table: Salesforce led, with Google, Amazon, Nvidia, Intel, IBM, Qualcomm and AMD alongside. A buyer probably comes off that table, because they have the data, the relationships, and the right to look first.

My ranking

Sorted by who can write the check and build their own Copilot behind the door:

Nvidia: the cleanest logic

Nearly all three million models on HF run on Jensen's silicon. Three million open models and a million datasets, in one place, nowhere else. That's the inventory. Thirteen million users whose first instinct, when publishing or hunting for a model, is to go there. That's the position.

The most valuable part is that every download passes through its hands. Who's running what, which tasks are scaling, which dataset suddenly gets pulled in bulk by enterprises, it knows before anyone else. It's a live meter on the whole open-source AI economy. Earnings arrive a quarter late; this meter is spinning every day.

Buying it welds the default distribution point for models into the CUDA estate, so every model download leads back to your own compute. And then there's neutrality. Hugging Face is worth what it's worth because it's Switzerland, and every lab is willing to park its models there. Nvidia is the only buyer that can own it and keep it Swiss, because Jensen was already selling shovels to everybody. Hand it to a cloud or a model company and Meta and Google start packing the next morning.

Google / Amazon: tied for second

The CSP case is compute adjacency: wherever the models live is where the inference burn happens. Both are existing shareholders. Both have the Switzerland problem: does your competitor still want his models sitting on your shelf?

Salesforce: third

Led the last round. The enterprise AI story needs it, and buying this would feel like an answer to the "SaaS is dead" line. Benioff is also famously fond of shopping. But $13 billion against its current capital discipline is a real hurdle.

Microsoft: cleanest playbook, biggest obstacle

If it bought, the GitHub script could be run again word for word. But precisely because it already owns GitHub, buying the GitHub of models would trigger an antitrust review on contact.

Nobody buys, also possible

The ask works out to roughly 100x revenue against 2024 third-party ARR estimates of ~$130M, or 185x against 2023's ~$70M. Either way it is far outside normal software territory, and the number may itself be the probe. Run the probe, discover that selling Switzerland to anyone destroys Switzerland, and take the IPO route instead.

Nobody ever made the ticket price back on tickets

So if it sells, it means the public infrastructure of open-source AI gets an owner for the first time. The position doesn't make much money running on its own. Slot it into a giant's P&L and every model download leads to compute, to cloud, to enterprise services. Microsoft paying $7.5B for GitHub was the same arithmetic. 25x revenue, everyone said it was expensive, and then Copilot started billing a few billion a year.

One asset never issues new shares; the other only just became infrastructure. The pricing logic is identical: cash flow lies, position doesn't.

Appendix: How Hugging Face actually got here

Everything above assumes it already occupies the position. But the position wasn't designed. It was salvaged from a failure.

A teen chatbot that didn't work (2016–2018)

Hugging Face was founded in New York in 2016 by three Frenchmen: Clément Delangue, Julien Chaumond and Thomas Wolf. The first product was a chatbot app for teenagers, named straight after the 🤗 emoji.

It raised twice: $1.2M in 2017, with SV Angel, Betaworks and NBA star Kevin Durant on the cap table; then a $4M seed in May 2018. $5.2M in total, all of it spent on a product that never worked. The chatbot topped out around 100,000 daily actives and stalled there. The team has been blunt about why: the technology kept improving, and none of that improvement converted into growth or retention.

One week that changed the company (late 2018)

In late 2018, Google released BERT. The Hugging Face team built and open-sourced a PyTorch implementation of it in a week. That week was the hinge. They were still a chatbot company when they did it, and the model was meant for their own product. But once it was out, the community response dwarfed anything their actual product had ever produced. In 2019 the company formally turned: the chatbot was shut down, and the Transformers library became the main line.

This is the part worth remembering: they didn't reason their way into the pivot. They shipped something, and the reaction told them what to be.

From library to platform (2019–2023)

The funding cadence after the pivot is really just the market repricing that position, round by round:

The interesting shift happens between C and D: the investors stop being financial funds and start being the industry. Everyone at the Series D table was someone who actually uses it: clouds, chip makers, enterprise software. At that point it stopped being a startup raising money and became something closer to shared industry ownership.

That same table is why today's buyer list is essentially pre-assembled.

The business model is: deliberately don't charge

Hugging Face's core move is not charging. Models, datasets and Spaces, unlimited, public or private, free. Revenue comes in three layers: individual Pro subscriptions, Enterprise Hub (SaaS and on-prem), and Inference Endpoints (managed inference). ARR ran roughly $70M in 2023 and about $130M in 2024. By 2025 the platform had around 13 million users and more than 500,000 organizations, with verified accounts at over 30% of the Fortune 500.

The model in one line: get every model in the world onto the shelf, then charge the companies that can't operate without it. Same script as early GitHub, and the consensus is the same too: it charges too little. The free surface is far larger than the paid one.

Was it ever in danger

Yes, but not in the way people assume. The real danger was 2018: a consumer product with 100,000 daily actives, flat growth, and $5.2M spent. The normal ending is a quiet shutdown. What saved it was an open-source side action with nothing to do with the core business. After the pivot the question changed from can it survive to how do you turn a position into money. Hosting costs scale linearly with model size and download volume; revenue depends on a thin layer of enterprise customers. That scissor never killed it, but it has never gone away either.

Which reframes "should it sell": the question is whether monetizing this position requires a bigger balance sheet than its own.

So what multiple is $13B

Against 2023 ARR of ~$70M, $13B is 185x. Against 2024 ARR of ~$130M, it's 100x.

Either way, this price isn't set on revenue. It's set on the position.

On sourcing: round sizes, valuations and investor lists come from contemporaneous coverage of each round (TechCrunch, Axios, SiliconANGLE, PYMNTS). ARR and user counts are third-party estimates (Contrary Research, Sacra, Latka), not company disclosures, so treat them as estimates. As of publication Hugging Face has not confirmed any intent to sell, and no buyer has been named.