Author | Wildcard

Edited by Jingyu

$12.9 billion—NVIDIA is set to acquire Hugging Face.

On August 26 local time, foreign media broke the news that, just two days earlier, reports had only mentioned Hugging Face "exploring a sale"—suddenly, it was announced that an agreement had already been reached. The speed was astonishing.

This is not an ordinary acquisition of an AI company—it’s the first time the king of chips has reached out to seize the "distribution rights" in the open-source AI world.

Hugging Face is known as the "GitHub" of AI, hosting over 3 million model repositories and serving 13 million developers. Open-source models from Meta’s Llama, Alibaba’s Qwen, and Mistral all reside on this platform. It is the de facto default distribution layer for the entire open-source AI ecosystem.

NVIDIA, on the other hand, has already monopolized the GPU hardware for training and inference.

Now, it also wants to own the path through which the model circulates.

01 Rolling around in AI

Let’s start with the money. Hugging Face’s last funding round was the Series D in August 2023, led by Salesforce with $2.35 billion, valuing the company at $4.5 billion. Google, NVIDIA, Amazon, IBM, Intel, AMD, and Qualcomm all participated in that round. Three years later, a $12.9 billion valuation is roughly three times the previous valuation.

At first glance, a threefold increase seems staggering. But within the context of the AI industry’s valuation bubble, this premium is actually restrained. Perplexity is currently negotiating funding at a valuation exceeding $30 billion, while Anthropic and OpenAI are valued at levels orders of magnitude higher than their revenues.

According to Sacra, Hugging Face's ARR (Annual Recurring Revenue) reached $150 million by August 2026, a fivefold increase from approximately $30 million in 2023. The company serves 50,000 enterprise customers and has 769 employees; CEO Clément Delangue recently stated on a podcast that the company is "close to profitability."

But at a price of $12.9 billion, you're not buying Hugging Face's revenue—you're buying its position in the AI world.

This logic is more akin to IBM’s 2019 acquisition of Red Hat for $34 billion—where a strategic buyer pays to secure a position in the developer ecosystem, rather than for short-term cash flow.

02 NVIDIA, Greater Ambitions

If you view this acquisition merely as "a chip company buying a model hosting platform," you're missing the bigger picture.

Over the past year, NVIDIA's investment in open-source AI has been surprisingly substantial.

At the GTC conference in March this year, Jensen Huang announced the formation of the "Nemotron Alliance," a global collaborative organization comprising eight AI labs—including Mistral AI, Perplexity, Cursor, and LangChain—aimed at jointly developing state-of-the-art open-source foundation models. NVIDIA has committed $26 billion over five years to this initiative, as disclosed in SEC filings.

This is the largest funding commitment ever made to open-source AI.

The first model developed by the alliance will serve as the foundation for the Nemotron 4 series. According to foreign media reports in early August, the largest version of Nemotron 4 is expected to reach a trillion-parameter scale, directly competing with the world’s most powerful open-source and proprietary models. Prior to this, NVIDIA had already released the Nemotron 3 series, including the Nemotron 3 Super (120 billion parameters, 12 billion activated parameters), released in March, and the recently launched Nemotron 3.5 Lightning, a lightweight open-source model capable of running on a single GPU.

Jensen Huang personally defended open-source models.

In July this year, he posted his first tweet on X, publicly supporting open-source AI, at a time when U.S. political circles were fiercely debating national security concerns surrounding Moonshot AI’s Kimi K3 model. His stance was clear: “Free AI is good for hardware. Free AI is good for chips.”

In simple terms, NVIDIA’s logic for open-sourcing is exactly the same as Google’s for Android—make the software free and drive hardware sales through the software ecosystem. Every developer using open-source models to build products needs NVIDIA’s GPUs for training and inference. The more models, the more users, the greater the demand for chips.

Now, by acquiring Hugging Face, NVIDIA is no longer just providing the hardware to train open-source models—it directly owns the home where these open-source models reside.

03 Embodied Intelligence, a Bigger Game

But NVIDIA's ambitions extend beyond language models and code generation.

This year at GTC, Jensen Huang unveiled Isaac GR00T N1, billed as the world’s first open-source general-purpose humanoid robot foundation model. At GTC Taipei in June, he introduced the Isaac GR00T reference humanoid robot—a complete open-source hardware reference design based on the Unitree H2 Plus humanoid chassis, the Sharpa five-finger dexterous hand, the Jetson Thor compute platform, and the full Isaac GR00T software stack. Stanford, ETH Zurich, the Allen AI Institute, and UCSD are among the首批合作机构.

Meanwhile, Cosmos 3, released at the end of May, is NVIDIA’s first fully open-source “multimodal model,” specifically designed for “physical AI”—that is, controlling robots, autonomous vehicles, and other machines that operate in the physical world. The autonomous driving inference model Alpamayo, unveiled at CES in January, is also open-source.

Looking at these points together, NVIDIA is building a complete open-source ecosystem for "Physical AI," covering everything from world models (Cosmos) and robot foundation models (GR00T N1) to simulation engines (Isaac Sim), hardware reference designs, and edge computing chips (Jetson Thor)—all fully open or openly available.

Huang Renxun said in Taipei: “Humanoid robots will bring physical AI to the world’s largest industry, unlocking a multi-trillion-dollar economic opportunity.”

Hugging Face, meanwhile, is the default platform where robotic researchers worldwide share models, datasets, and tools. With NVIDIA now owning this distribution layer, its open-source ecosystem for embodied intelligence has transformed from a fragmented collection into a fully integrated end-to-end闭环. Training occurs on NVIDIA GPUs, models are hosted on NVIDIA-owned platforms, inference runs on NVIDIA chips, and robots controlled by NVIDIA-powered computing systems.

04 "Neutral Platform" Controversy

The most pressing concern for everyone is Hugging Face's "neutrality."

In January of this year, it was reported that Hugging Face rejected NVIDIA's $5 billion investment, primarily because the founding team feared that becoming too closely tied to one chip giant would compromise the platform's neutrality among competitors like AMD and Intel.

But seven months later, they chose to accept the $12.9 billion acquisition.

The concerns of the developer community are real. One user on X bluntly stated: “One way to suppress the open-source path is to have Hugging Face acquired by a large corporation.” The platform hosts a large number of open-source models from Chinese companies, and Chinese models such as DeepSeek and Alibaba’s Qwen are already among the most popular options on Hugging Face by download volume across multiple categories. Can a platform owned by a U.S. chip giant continue to serve as a globally neutral hub for AI model distribution?

History provides two reference points. In 2018, Microsoft acquired GitHub for $7.5 billion, causing initial panic among developers, but Microsoft ultimately preserved GitHub’s independent operations and open ecosystem, making it even better. Similarly, after IBM acquired Red Hat for $34 billion in 2019, it pledged to maintain its open-source culture. These acquisitions ultimately demonstrated that "absorbing" an open-source platform does not necessarily mean "closing" it.

But NVIDIA's situation is fundamentally different.

When Microsoft acquired GitHub, Microsoft was not a monopolist in the developer tools market; when IBM acquired Red Hat, IBM was not the absolute dominant force in the enterprise computing market.

NVIDIA holds over 80% of the AI GPU market.

A player that monopolizes hardware while also owning the largest distribution platform for open-source software must be carefully scrutinized by regulators.

The ratios of cash to stock, arrangements for retaining the core R&D team, and how the platform’s open-source commitment will be continued have not yet been disclosed. These details will determine whether this acquisition represents a normal expansion of the ecosystem or a vertical integration that shakes the industry.

From investing $26 billion to build the Nemotron alliance and develop trillion-parameter models, to GR00T robots and the Cosmos world model, and finally acquiring Hugging Face for $12.9 billion—NVIDIA’s actions over the past six months have traced a clear path.

It no longer wants to merely sell shovels to gold miners—it wants to own the gold mine itself. Or more accurately, it wants to own the entire supply chain, from the mine to the marketplace.

What does this mean for the AI industry? When the world’s most powerful AI infrastructure company begins to control hardware, models, distribution platforms, and robotics development stacks simultaneously, the very meaning of “open source” may need to be redefined.