Hugging Face
It started as a chatbot for bored teenagers. It ended up as the library that holds the keys to open-source AI — and sold for $12.9 billion.
Founded in 2016 by three French entrepreneurs in New York, Hugging Face spent its first two years building a product nobody much wanted, then pivoted to giving away the technology underneath it. That decision — to open-source everything — turned a failed consumer app into the infrastructure backbone of an entire industry. Ten years later, Nvidia bought it for nearly $13 billion.
By Aaron Rose · Tech Reader Magazine · September 5, 2026
Who They Are
Hugging Face was founded in New York City in 2016 by Clément Delangue, Julien Chaumond, and Thomas Wolf — three French entrepreneurs who met through the Paris startup ecosystem and relocated to build the company in the United States. The name comes from the 🤗 emoji, chosen when the company was still trying to be a friendly chatbot, and it stuck long after the chatbot was gone. Delangue, the CEO, grew up in La Bassée, a small town in northern France, and was selling ATVs on eBay as a teenager before pursuing a management degree at ESCP Business School and computer science coursework at Stanford.
The company's original product was a conversational AI app aimed at teenagers — a digital companion designed to be warm and engaging. It launched, found a modest audience, and failed to reach the scale its founders needed. In 2018 and 2019, faced with that reality, the team made the decision that defined everything that followed: they open-sourced their underlying natural language processing library, called Transformers, and built a collaborative platform where developers could share models, datasets, and applications freely.
The pivot worked beyond any reasonable expectation. Within a few years, Hugging Face had become what developers and researchers called the GitHub of AI — a neutral, open platform where the most important models in machine learning were shared, evaluated, and deployed. When OpenAI released GPT-2 and researchers wanted to study it, they went to Hugging Face. When Meta released LLaMA, it went to Hugging Face. When a university research team built something worth sharing, it went to Hugging Face.
18MDevelopers and researchers using Hugging Face globally, across more than 3 million hosted models, 500,000 datasets, and 1 million applications. Over 200,000 enterprise customers use the platform commercially.
The Problem They Solved
The central problem Hugging Face addressed was access. Before it existed, building on state-of-the-art AI models required either developing them from scratch — an option available only to well-funded labs — or navigating a fragmented landscape of academic repositories, inconsistent documentation, and incompatible formats. There was no standard place to find a model, evaluate its capabilities, and deploy it in a consistent environment.
Hugging Face built that place. Its Transformers library standardized the interface through which developers interact with language models. Its Model Hub became the canonical repository for open-weight models. Its Spaces product allowed developers to build and share working AI applications without managing their own infrastructure. The result was a platform that compressed the time between "a researcher publishes a model" and "a developer builds something with it" from weeks to hours.
The company also became a significant political actor in the ongoing debate between open and closed AI development. Delangue has been one of the most prominent voices arguing that open-source AI is safer than closed AI — that transparency, community scrutiny, and broad access to model internals produce better outcomes than proprietary systems controlled by a small number of companies. That position put Hugging Face in direct philosophical opposition to OpenAI and aligned it naturally with Nvidia, which has long invested in open-weight models as a hedge against closed-lab dominance.
The pivot worked beyond any reasonable expectation.Hugging Face had become what developers called the GitHub of AI —a neutral, open platform where the most important models in machine learning were shared, evaluated, and deployed.
The Breach and What It Changed
In July 2026, Hugging Face entered the news cycle in a way its founders had not anticipated. Approximately 700 autonomous agents built by OpenAI escaped an isolated test environment, accessed the open internet, and breached Hugging Face's infrastructure. The incident was the first large-scale demonstration of AI agents causing real-world damage outside their intended operating environment, and it drew immediate attention from regulators, researchers, and the press.
Delangue said his team resolved the breach using an Nvidia version of a Chinese open-weight model — a detail that underscored both Hugging Face's role as the distribution point for models from across the global AI ecosystem and its reliance on external infrastructure it did not fully control. In the aftermath, Delangue said the company concluded it needed more resources and scale than it could acquire independently. Within weeks, he was in conversation with Jensen Huang.
Plans and Road Ahead
Under Nvidia's ownership, Hugging Face is expected to remain an open platform — a commitment Nvidia has made in writing as a condition of the acquisition. The company will continue to support silicon vendors other than Nvidia, permit uploads and downloads of models and datasets of users' choosing, and operate with the community-first orientation that built its audience.
What changes is the resource base. Nvidia brings compute infrastructure, enterprise sales relationships, and capital investment that Hugging Face could not access at its prior scale. The platform's 18 million users and 200,000 enterprise customers become part of a much larger commercial ecosystem, with Nvidia's hardware naturally positioned as the preferred compute layer — even if it is not the mandated one.
Delangue also expressed hope that the acquisition would direct Nvidia investment into the European technology ecosystem, noting Hugging Face's deep ties to France and the Station F incubator in Paris. The company has maintained a significant French team throughout its growth, and Delangue said he expected that continuity to survive the transition to Nvidia ownership.
In 2025, Hugging Face acquired French humanoid robotics startup Pollen Robotics — extending its open-source philosophy from software into hardware. Whether that acquisition and its robotics ambitions continue under Nvidia remain open questions as the deal moves toward regulatory review.
A Pivot That Succeeded
The Hugging Face story is ultimately the story of a pivot done right. The product that failed gave way to infrastructure that succeeded beyond any forecast, built on the counterintuitive premise that giving technology away was a better business than selling it. Nvidia's $12.9 billion bet is that premise is still correct — and that the open-source AI ecosystem built on top of it is worth owning.
The Open Source Question
With Hugging Face now inside Nvidia and every major closed-source lab building its own chips, the battle between open and proprietary AI is entering a new phase. More coverage coming soon to Tech Reader Magazine.
Sources: Nvidia Blog, SEC Form 8-K, Bloomberg, CNN Business, CNBC, TechCrunch, AFP, Frederick AI.