It happened. It really happened.
After a three-year marathon pursuit, NVIDIA finally made the official announcement yesterday — acquiring all outstanding shares of Hugging Face for $11.9 billion, plus a $1 billion employee retention package, bringing the total transaction value to $12.9 billion.
This is not the first time the deal has been rumored, but it is the first time it has actually happened.
Let us rewind the timeline to the summer of 2024. At the time, Hugging Face was valued at $4.5 billion, and NVIDIA made its first approach, only to be politely turned down by Clément Delangue. In early 2025, NVIDIA raised its offer to $8 billion and was rejected again. It was not until September 3, 2026, that the two sides finally shook hands at a valuation of $12.9 billion — making this the third-largest acquisition in NVIDIA’s history, behind only its $20 billion acquisition of Groq at the end of last year and its $6.9 billion acquisition of networking equipment company Mellanox in 2024.
Jensen Huang put it plainly in NVIDIA’s official blog: “Hugging Face is the spiritual home of AI developers around the world. Combining NVIDIA’s accelerated computing with Hugging Face’s open-source ecosystem will reshape the way humanity builds artificial intelligence. Together, we will create an open AI development platform across platforms. This is not an acquisition. This is an alliance.”
Clément Delangue, meanwhile, posted a photo of himself wearing a black leather jacket alongside Jensen Huang on social media, with only one line of text: “Together, we are building AI for everyone.”
But the story behind that sentence is far more complicated than a single photo.
From Chatbot to a Million-Model Empire
The story of Hugging Face began in 2016 — as a chatbot app built for teenagers. No one at the time could have predicted that this small Paris-based team would, seven years later, become the shared home of AI developers around the world.
The turning point came in 2019. Hugging Face decisively abandoned its chatbot business and fully embraced the open-source NLP ecosystem, launching the transformers library — a unified API connecting the three major frameworks of TensorFlow, PyTorch, and JAX. The bet paid off. As ChatGPT ignited the large-language-model boom at the end of 2022, Hugging Face quickly became a holy ground for AI developers.
Today, Hugging Face has grown into a massive ecosystem: more than one million open-source models, 200,000 datasets, and 500,000 Spaces applications. More than five million developers worldwide are active on the platform every month. From Llama to Mistral, from Stable Diffusion to Qwen, from Whisper to Phi — for nearly every major open-source model, Hugging Face is the first stop.
Its transformers library records more than 100 million PyPI downloads per month. Its Inference API handles millions of model inference requests every day. It has around 500 paying enterprise customers — from financial giants to medical research institutions — using the platform to manage internal models and deploy MLOps workflows.
This is why NVIDIA was willing to spend $12.9 billion.
The Open-Source Sanctuary Choked by Computing Costs
Why did Hugging Face ultimately choose to sell? The answer lies in a harsh reality that every open-source platform eventually has to confront.
After reaching a valuation of $4.5 billion in 2023, Hugging Face’s valuation never broke through again. Not because the platform was not making money, but because it was burning too much of it. Tens of millions of model downloads every day, petabytes of model-weight storage, free model inference APIs, GPU compute rentals — all of these costs continued to soar as the parameter counts of large models grew exponentially.
In an interview in 2025, Delangue revealed that Hugging Face relied heavily on cloud credits from AWS and GCP to keep its infrastructure running. This state of “depending on handouts from tech giants” left the company in a constant state of financial delicacy. At the same time, venture capital markets were losing patience with the monetization prospects of open-source platforms — after 2025, the fundraising environment for open-source AI companies deteriorated sharply.
Hugging Face’s B2B business was growing, but revenue from 500 enterprise customers was nowhere near enough to cover its compute costs.
NVIDIA’s solution was simple: you no longer have to buy GPUs yourself. DGX Cloud could be connected directly to the Hugging Face backend, compute costs could be internalized, InfiniBand networking could provide direct connectivity, and the infrastructure anxiety surrounding model training and inference could effectively disappear in one stroke.
For NVIDIA, Hugging Face brings more than just brand value and community. More importantly, it provides access to developer behavioral data: which models are most popular, which tasks are most frequently used, and which architectures are gaining momentum. These signals serve as a weather vane for the design of NVIDIA’s next generation of GPU architectures.
A Rapidly Closing Golden Window: Why NVIDIA Had to Buy
There is an even deeper timeline logic behind this acquisition.
From late 2025 into 2026, the AI chip market began showing a trend that was unfavorable to NVIDIA: AMD’s MI400 series started to pose a genuine threat to the market share of the H100/B200. Intel’s Falcon Shores also began making progress in the inference market. More importantly, cloud giants intensified their efforts in custom silicon — AWS Trainium3, Google TPU v7, Microsoft Maia 200. These specialized chips may not be able to fully replace NVIDIA, but they are steadily eating into the incremental market.
NVIDIA’s moat has always been built on the CUDA software ecosystem and developer habits. But if developers gradually become comfortable with AMD’s ROCm, Intel’s OneAPI, or the proprietary SDKs of cloud giants, that moat will slowly begin to dry up.
Hugging Face is the first stop where developers around the world choose frameworks and underlying hardware. Controlling Hugging Face is effectively equivalent to controlling the developer default. When you download a model from Hugging Face in the future, the default hardware backend will be NVIDIA — not through technical coercion, but through ecosystem inertia.
The $12.9 billion is not just buying a platform. It is buying the developer gateway to the next decade of the AI industry.
A New Geopolitical Variable: The Collision Between Open-Source Models and National Security
This acquisition also introduces an entirely new variable that its predecessors never had to face — the sensitivity of open-source AI models in the context of national security.
Since 2025, governments around the world have sharply increased their scrutiny of open-source AI models. Open-source models such as DeepSeek R1, Llama 3, and Qwen have been widely deployed in critical sectors including defense, finance, and healthcare. The U.S. Department of Commerce and Congress have held multiple hearings to discuss whether open-source models should be brought within the scope of export controls.
In its SEC filing, NVIDIA explicitly warned: “Governments around the world may impose new licensing, restriction, or regulatory requirements on open-source models… any such regulatory measures could have a material impact on the Hugging Face platform and our business.”
This means NVIDIA may have to face both antitrust review and national security scrutiny at the same time. The painful lesson of the failed ARM acquisition still looms large — NVIDIA was once willing to pay $40 billion for ARM, only to abandon the deal under global regulatory pressure. The $12.9 billion Hugging Face acquisition is far smaller in scale, but it involves AI infrastructure, now one of the most politically sensitive areas of high technology worldwide. The regulatory difficulty may be even greater, not less.
Margrethe Vestager of the European Commission has already stated publicly that regulators need to “closely monitor acquisitions of open-source AI platforms by large technology companies.” The UK CMA’s newly appointed head, Douglas Fraser, has also identified the open-source AI ecosystem as a priority area for regulatory review.
The Prologue to a New Era
Looking back from this historical moment in September 2026, NVIDIA’s acquisition of Hugging Face may ultimately be defined as a watershed moment for the AI industry.
Before this, the power structure of the AI industry was a three-way balance between “compute providers vs. model developers vs. the open-source community.” After this deal, the strongest compute provider and the strongest open-source community become one — creating a “full-stack super-entity” with top-tier chips, a top-tier software ecosystem, and a top-tier developer community.
For developers, this is an unsettling but impossible-to-ignore signal: will AI development in the future become more open, or more dependent on the infrastructure of a handful of giants?
For AMD, Intel, and the cloud giants, this means bypassing CUDA has become an order of magnitude more difficult — because now they not only have to bypass CUDA, they also have to bypass Hugging Face.
Jensen Huang once said in 2024: “We are not just selling chips. We are selling the entire development workflow.” In September 2026, that statement has finally become a complete reality.
The next step of AI will run on the tracks set by Jensen Huang. And the man in the black leather jacket is stepping harder on the accelerator.