NVIDIA buys Hugging Face for $12.93 billion: who controls the warehouse of open models

The figure in the official announcement is written out in full: $12,930,300,000. The post on NVIDIA's blog, signed by chief executive Jensen Huang, opens with the line "I'm excited to announce that NVIDIA has agreed to acquire Hugging Face for $12,930,300,000". The same figure is independently confirmed by TechCrunch, The Register and Quartz, all dated 3 September 2026. According to The Register, closing is expected in the first half of 2027, subject to regulatory approval and the usual conditions. How much the acquired platform weighs is quantified by NVIDIA itself: 18 million developers, researchers and creators, more than 3 million shared models, 500,000 datasets, one million applications and over 200,000 corporate users. The buyer is no stranger to the place: it says it has already contributed more than 500 models and 250 open datasets.

The interesting part is not the price, it is the position. Hugging Face is the leading public repository of open models and datasets, the point through which much of the distribution of weights passes — including the weights of NVIDIA's direct competitors in inference software. The buyer is the dominant supplier of the hardware on which those same models are trained and run. NVIDIA answered the obvious question in advance, with explicit commitments in the announcement: developers remain free to choose models, frameworks, cloud providers and compute platforms; the use of NVIDIA hardware will not be required; the platform will keep supporting multi-cloud and multi-accelerator development. "Hugging Face will remain an open platform for the entire AI ecosystem", Huang writes. The commitments come with no deadlines, no metrics and no stated verification mechanisms.

On the corporate numbers, TechCrunch reports that Hugging Face was founded in 2016, has raised more than $395 million in total — the last round in 2023, $235 million led by Salesforce Ventures — and has annualised revenue of around $150 million. TechCrunch, the only outlet to report it, also writes that the company had turned down a $500 million NVIDIA offer in the past. On the structure of the deal the information comes from CNBC, whose page could not be opened directly (HTTP 403 response): roughly $11.9 billion to shareholders and up to $1 billion in retention equity for the employees moving to NVIDIA, with Hugging Face's chief executive said to have reached out to Huang over the summer and the three co-founders set to join NVIDIA. The official post gives only the total figure and does not confirm this breakdown. What is missing should be said too: no official announcement was found on Hugging Face's own blog, and the words of its chief executive — "Open source AI is at an inflection point... it needs more compute, more support, more collaboration and more visibility" — all arrive by way of the press. NVIDIA does not specify which jurisdictions will have to clear the deal, which filings have already been made, whether Hugging Face will remain a separate entity or how it will be governed.

Quartz collects an outside assessment: analysts urge companies to watch for any drift away from the declared openness, in an ecosystem described as circular, with capital circulating among a small group of large firms. Charlie Dai, an analyst at Forrester Research, notes that "Nvidia has already built an extended ecosystem through GPUs, CUDA, networking, inference software, and AI frameworks". What stays with me is a question of measurement, not of intent: a neutrality commitment with no deadline, no metric and no referee is not false, it is simply unverifiable. Whether the platform stays open is something we will learn by watching what happens to a model that runs better on someone else's hardware — and so far nobody has said who will be keeping score.

— Olya