AI

Nvidia is paying $12.9 billion for Hugging Face. Its reported revenue is $150 million.

The Information reported the agreement on Wednesday; Reuters, CNBC and Forbes followed. Neither company has confirmed it. The price is about 2.9 times the $4.5 billion valuation Hugging Face carried in 2023 — a round Nvidia invested in — and roughly 86 times reported annualised revenue.

N Noah · The Sharp Brief · August 27, 2026 · 3 min read

Nvidia has agreed to buy Hugging Face for $12.9 billion, The Information reported, with Reuters, CNBC and Forbes all carrying the figure. Neither company had confirmed the agreement publicly when the reports landed.

Five days ago we wrote that Hugging Face was testing a sale and that the buyer list was the problem — every plausible bidder had a reason to want the neutral distribution layer for open models to stop being neutral. The list has now resolved. The buyer is the company whose chips run essentially everything hosted on the platform.

Set the price against the two numbers that give it shape. Hugging Face was valued at $4.5 billion in a $235 million round in 2023 — a round Nvidia itself invested in, alongside Google and Salesforce. Reuters reported the company’s annualised revenue at about $150 million. At $12.9 billion, that is roughly 2.9 times the 2023 valuation and roughly 86 times revenue.

Nobody is paying 86x for the revenue

They are paying for position. Hugging Face is where open and open-weight models get published, versioned, benchmarked and pulled — the default first stop for a developer who wants to try something that did not come from OpenAI or Anthropic. It is a distribution chokepoint dressed as a package registry.

Nvidia’s logic is not complicated, and the company has been fairly open about it: the longer open-source models stay competitive, the longer the industry keeps buying general-purpose accelerators instead of consolidating around a handful of closed labs running custom silicon. Owning the shop window for open models is a hedge on the thing that keeps Nvidia’s market structure intact.

It also fits the pattern. This is the same company that agreed to pay $6 billion for Poolside’s model factory without buying Poolside, and that has been reported in talks over a stake in Perplexity. Nvidia is spending its earnings on the layers above the chip.

Our take: Open weights were supposed to be the counterweight to closed labs. The counterweight’s distribution layer now belongs to the one company that profits whichever side wins. That is not a conspiracy — Nvidia has every commercial reason to keep the Hub open, cheap and vendor-neutral, because a neutral Hub is precisely what makes it valuable. The risk is subtler: a registry that is technically open but strategically owned tends to develop opinions about which runtimes work best.

What to watch

It has been a loud week for one company. Nvidia reported $96.2 billion in quarterly revenue, with $89 billion of that from data centre, and guided to about $108 billion for the current quarter. Against those figures, $12.9 billion for the front door to open-source AI is roughly six weeks of sales.

Advertisement

Get the day, decoded — at 7 PM ET

The Sharp Brief: AI, money, business & performance in five sharp minutes. Free.

Free bonus: subscribe today and The 2026 AI Playbook (PDF) lands with your welcome email.

Recommended by 5+ newsletters across AI, markets & business.