Is Hugging Face too important to ever fall into the hands of NVIDIA? Jensen Huang: I originally wanted it to remain independent, but there are other bidders.
On Thursday local time, NVIDIA officially announced that it has reached a definitive acquisition agreement with Hugging Face, and will acquire one of the world's most important open model platforms for 12.9 billion US dollars. The transaction is expected to be completed in the first half of 2027, but it still needs to obtain regulatory approval and meet other customary closing conditions.
On the surface, the commercial logic of this transaction is not complicated: open foundation models and the applications built on them will continuously drive demand for GPUs, inference software and AI infrastructure; and Hugging Face is exactly at the core of open model discovery, distribution, deployment and community collaboration. Taking over Hugging Face means that NVIDIA is no longer just selling computing power at the bottom of the AI industry, but further extending to model distribution, developer ecology and community entry.
However, some foreign media commented on this acquisition: "Regulators should block this deal."
The Register believes that this may be one of the most serious M&A deals worthy of review in the AI industry so far. "You would not allow a car manufacturer to acquire a major fuel distribution channel, nor would you allow it to buy the main system for training maintenance technicians, but NVIDIA's acquisition of Hugging Face essentially does almost both things at the same time."
NVIDIA has already occupied an extremely strong position through GPUs, CUDA, networks, inference software and AI frameworks. Now it has further mastered one of the most important entrances to the open model ecology, and the degree of vertical integration is significantly deepening. The real question is to what extent a neutral platform that was originally regarded as "Switzerland in the AI world" can maintain true neutrality after it becomes the asset of the most powerful infrastructure company in the industry.
Nevertheless, Hugging Face CEO Clément Delangue still highly praised NVIDIA in an interview on Thursday, and regarded this acquisition and the capital brought by NVIDIA as an opportunity to expand the user scale of Hugging Face. Hugging Face currently has approximately 18 million users, and Delangue hopes to increase this number to more than 100 million in the next few years.
But just a year ago, Hugging Face rejected a $500 million investment from NVIDIA. This at least means that Delangue is likely to be clearly aware that the agreement he has just reached is like a "deal with the devil". However, a full $1 billion of this transaction will be specifically allocated to Hugging Face employees who join NVIDIA, so Delangue and his team will have enough money to "wipe their tears" even if they really regret it in the future.
1
What does NVIDIA get by acquiring Hugging Face?
NVIDIA may prefer to describe this acquisition as a public welfare move: by providing financial support, it ensures that one of the most important AI resources in the Internet era can obtain long-term financial security.
"It is completely reasonable to think that Hugging Face is a key cornerstone supporting the entire AI ecology. Once it runs out of funds and cannot continue to operate, the consequences are likely to be catastrophic." And Hugging Face itself is also a capital-intensive business. Storing AI models and datasets in petabytes is by no means a cheap business, and the network bandwidth required to continuously distribute these models to users on a large scale is also very costly. With NVIDIA's involvement, Hugging Face hardly needs to worry about the financial pressure brought by infrastructure demand in the future.
But the benefits brought by this kind of NVIDIA "funding" are obvious, and the harm it may cause to NVIDIA's competitors is also obvious.
Hugging Face is often compared with GitHub, but it is no longer just a repository for storing and sharing model weights and training data.
For example, Hugging Face has one of the most complete AI development documentation systems on the Internet. For NVIDIA, it is not difficult to use its newly acquired control position to ensure that its own products always have more complete documentation support than competitors. NVIDIA can of course keep emphasizing that the platform remains "open", but openness does not mean that it must be unbiased.
In addition, there are Hugging Face's important contributions to the software ecology. Earlier this year, the popular local AI inference engine llama.cpp has officially joined the Hugging Face system. At the same time, Hugging Face's Transformers Python library is also an important basic component of mainstream inference platforms such as vLLM and SGLang, which directly compete with NVIDIA's own TRT-LLM.
This acquisition puts NVIDIA in an extremely favorable position to give priority to supporting its own hardware and software products. NVIDIA, the "AI arms dealer", does not even need to take obvious and rude measures such as stopping support for competing platforms. It only needs to ensure that the models and frameworks it favors always have more complete documentation and can run on NVIDIA hardware first, which is enough to produce a huge impact.
A way to quietly change the balance of competition is to inject a large amount of cheap, NVIDIA hardware-based computing power into Hugging Face. In the past few years, Hugging Face has provided computing resources to users through forms such as Inference Endpoints, Providers, and Spaces. In order to provide this computing power, Hugging Face cooperates with a large number of different suppliers, covering a variety of hardware platforms, including large cloud service providers, inference-as-a-service vendors, and chip design companies such as NVIDIA, AMD, Cerebras, SambaNova, and Groq.
Once NVIDIA becomes the parent company of Hugging Face, it will have the ability to subsidize computing resources through various partners, and economically incentivize developers to prioritize development for NVIDIA hardware. Of course, this approach must be sufficiently concealed. For example, it can require some emerging cloud service providers to allocate a certain proportion of computing resources to low-cost education and development purposes, so as to promote developers to give priority to using the NVIDIA platform without causing too much suspicion.
No matter from which perspective, no matter how much NVIDIA insists that it will never secretly manipulate the balance of competition, this acquisition of Hugging Face has a fairly high probability of harming market competition. Compared with being part of the most powerful company in the entire AI industry, Hugging Face's independent existence as "Switzerland in the AI world" is obviously more in the interests of the entire industry.
Facing various controversies, Jensen Huang and Clément Delangue specifically responded to this acquisition in an interview with CNBC.
The two revealed that the acquisition was initiated by Hugging Face. This summer, OpenAI's models exposed uncontrollable and untraceable security risks, and the closed-source black box could not defend itself, making the officials fully realize that only open source can solve the AI security problem, and stronger resources must be obtained to accelerate expansion.
For this acquisition, there are multiple bidders in the market. NVIDIA took over at a high price to "protect the world's largest open source AI ecology", and at the same time matched a 1 billion US dollar talent retention plan, retaining all employees and continuing to expand the team, with the goal of increasing the number of developers from 18 million to more than 100 million. Jensen Huang emphasized that NVIDIA does not make money by locking customers, but by the growth of the entire AI industry. Half of NVIDIA's business has long been driven by open source models, and open source is the real growth engine.
Below are the detailed dialogues and responses from the two parties. We have translated them and made deletions and arrangements on the premise of not changing the original meaning, for the benefit of readers.
2
The beginning and end of the acquisition
Host: Today we have the honor to invite Jensen Huang, CEO of NVIDIA, and Clément Delangue, CEO of Hugging Face, to talk with us about this blockbuster acquisition. Welcome to both of you, and it's a pleasure to communicate with you. Jensen, I see a Hugging Face brand ornament on your desk, which is very thoughtful.
Delangue: I also have the same ornament behind me.
Host: Yes, I just saw it. The rumors of this acquisition have been fermenting for more than a week, and people from all walks of life in the market have been talking about it. Can you tell us the whole story? How did you reach a consensus on cooperation?
Delangue: Well, Jensen, you go first.
Jensen Huang: No, Delangue, you go first.
Delangue: Okay. This summer, we clearly perceived that Hugging Face and even the entire open source artificial intelligence industry have reached a critical turning point. The industry urgently needs more resource support, larger development scale and higher industry influence. So we took the initiative to find Jensen, and told him that we wanted to make the open source artificial intelligence industry bigger and stronger. He immediately replied to us: "Let's do it together." That's how the whole thing started, and just a few weeks later, we finalized this acquisition.
Host: Delangue, what exactly happened this summer? When you mentioned that the industry has reached a turning point, the first thing I thought of was the abnormal incident of OpenAI's agents from May to July this year, and your team detected and dealt with related problems in time. Is this the direct trigger for this acquisition? Or are there other reasons that made you realize that the enterprise needs stronger financial support?
Delangue: That incident made us fully see the core value of open source models. At that time, we found that relying only on closed-source proprietary APIs could not do a good job in security protection at all, and we could only rely on open source models to resist risks in the end. This incident fully proves the irreplaceability of open source AI. Looking at the entire industry, Anthropic will also launch an IPO within a few months, and the trend of industry differentiation is already very obvious.
Now the AI industry has formed two completely different development paths: the first is the closed-source proprietary API route, where all walks of life outsource AI capabilities to a few leading enterprises; the second is the open source and open route, where artificial intelligence technology is open to the whole society, and everyone can become the owner and developer of AI technology, rather than simply renting and using others' AI services. The pattern of the two development paths has been determined, and we must fully focus on the open source AI field to make this technology more widely popularized and applied around the world.
Host: Jensen, recently you and many industry leaders have been promoting open source AI, and your views are particularly clear and your attitude is very firm. But we know that OpenAI, Anthropic and other enterprises that focus on closed-source large models have always been NVIDIA's core large customers, and they will continue to contribute revenue to the company in the future. How do you balance the development of the two tracks of open source AI and closed source AI?
Jensen Huang: In fact, all major closed-source model R&D labs also highly recognize and support the development of open source models, and are well aware that open source AI has its unique industry value and living space. NVIDIA is the world's top artificial intelligence computing platform, with business covering public clouds, new cloud service providers and enterprise on-premises deployment scenarios, and its service network spreads all over the world.
In the past year, especially in the past six months, artificial intelligence has officially entered the practical stage. Both closed-source models and open-source models can stably produce content with actual commercial value. For ordinary users and enterprises, I suggest giving priority to using closed-source models. These models are ready to use, with top performance and extremely fast iteration speed. We at NVIDIA also regularly use closed-source AI tools such as Claud Code, Codex, Perplexity, and Cursor. As long as the commercial AI rental service experience is good enough, we will give priority to it, which is also our actual choice.
But many enterprises have differentiated core needs: scientific research institutions hold exclusive domain technologies and core private data, which need to be kept strictly confidential; many enterprises, out of the consideration of compliance operation and data sovereignty, need to fully control core technologies independently, hold intellectual property rights, and continuously iterate self-developed AI models.
At this stage, the performance of cutting-edge open source AI models is mature enough, and the supporting intelligent agent building tools have also greatly lowered the threshold for enterprises to develop AI independently. Both the open source and closed source tracks are growing at a high speed. Cloud service provider business only accounts for 50% of NVIDIA's total business, and the other 50% of the business is basically driven by open source models. Therefore, NVIDIA always adheres to the parallel development of the two tracks, and our core goal is to fully promote the rapid development of the artificial intelligence industry.
At present, open source AI is in the golden period of rapid growth. We will provide Hugging Face with sufficient platform resources and financial support to help it expand its industrial scale and continuously expand the open source model ecology and global developer community.
Host: Jensen, when Delangue took the initiative to discuss cooperation with you, what was your first reaction? How was the acquisition valuation of 12.9 billion US dollars finalized? As early as 2023, Hugging Face completed a large round of financing, with participation from Google, NVIDIA, and Salesforce. At that time, the company's valuation was only 4.5 billion US dollars.
Jensen Huang: Hugging Face has 200,000 enterprise customers and 18 million developers around the world. The platform hosts more than 3 million AI models, covering almost all artificial intelligence sub-fields such as multilingual foundation models, physics, chemistry, biology, and robotics. At the same time, NVIDIA is also the enterprise that contributes the largest number of open source models in the world.
When Delangue discussed the future development plan of the enterprise with me, I was convinced at the first time that this high-quality platform, this enterprise that has been deeply engaged in the industry for ten years, must be handed over to a reliable acquirer to ensure its continuous and stable development. Hugging Face's development vision is very precise, and the entire industry and a large number of developers are highly dependent on the open source model ecology. Now that open source AI is developing rapidly, the industry is full of both opportunities and challenges, and sufficient resource empowerment is urgently needed.
To be honest, I initially hoped that Hugging Face could maintain independent operation, after all, the company's current development momentum is very strong. But there were multiple bidders in the market at that time. To facilitate this transaction, 12.9 billion US dollars is a reasonable transaction price, and every penny is worth it.
As I said before, NVIDIA is the world's top AI computing platform, and open source models are our core strategic layout, which is also the core reason why we have been deeply engaged in the open source field for a long time. In addition to large language models, all walks of life can benefit from open source AI, and we are unswervingly committed to this. Hugging Face needs a high-quality development platform, and NVIDIA is its best destination.
Host: From what you said, the outside world speculates that you are worried that Hugging Face falling into the hands of other enterprises will damage the existing open source ecology, so you spent 12.9 billion US dollars to complete the acquisition. Is that true?
Jensen Huang: No. The reason why we offered a valuation of 12.9 billion US dollars is, on the one hand, to fit the industrial value of Hugging Face itself, and on the other hand, because open source AI is of great strategic significance to NVIDIA. Since Hugging Face wants to start a new development stage and find a high-quality partner, NVIDIA is the most suitable choice.
Delangue communicated with me very frankly about the company's current development status and his idea of promoting enterprise transformation and upgrading. At that time, many enterprises had already thrown acquisition offers to Hugging Face. Open source business is NVIDIA's core growth engine. After this acquisition is completed, we can further expand the global open source developer community and achieve faster development. This transaction is a win-win good thing for Hugging Face, the open source ecological community, and NVIDIA.
3
All employees join NVIDIA, 1 billion US dollar talent retention plan
Host: Delangue, what is your personal plan next? Will you continue to work at Hugging Face after the acquisition is completed?
Delangue: Yes. The three founders of the company and all employees will continue to stay in the team and join NVIDIA. Just a few minutes ago, Jensen just attended our all-hands meeting, and all employees are very encouraged and full of morale. After being integrated into NVIDIA, Hugging Face will continue to operate with an independent and neutral platform positioning. All of us are looking forward to starting a new development stage and releasing greater industry influence.
As Jensen said, our platform currently has 18 million AI developers. In the next few years, our goal is to increase the scale of platform developers to 100 million, empower global users, so that everyone is no longer just a user of AI, but can become a builder of AI and independently control artificial intelligence technology. If this goal can be achieved, it will definitely bring a subversive positive impact to the global AI industry.
Host: It is reported that this acquisition is accompanied by a talent retention plan of up to 1 billion US dollars to retain Hugging Face's core team. Is this news true?
Jensen Huang: It's true.
Delangue: We have no additional comments on this.
Jensen Huang: There is no need to avoid this issue. Talent is the core foundation of an enterprise. All employees of Hugging Face have poured years of effort into building this platform for global developers to collaborate and innovate and jointly promote the progress of AI technology, making great contributions to the industry and global technology development.
Technology iteration under the open source model is more secure and develops faster, which is also the core reason why NVIDIA