Abstract:
According to people familiar with the matter, NVIDIA has agreed to acquire the open source model platform Hugging for US$12.9 billion. Face. The report was issued on the evening of August 26. If this transaction is completed, it will hand over the world's major open source model warehouse to NVIDIA. Nvidia and New York-based Hugging Face has not publicly confirmed whether the consideration will be cash, stock or mixed payment, as well as the delivery conditions and open source commitments, which were not stated in the report. Another media outlet said earlier that the two sides have been negotiating an acquisition valued at more than $13 billion in recent weeks. An agreement has not yet been reached and the negotiations may still break down.
1. Negotiation background: Multiple inquiries, Hugging Face once refused NVIDIA’s capital increase
Hugging Face was established about ten years ago. The platform is similar to a code hosting station for machine learning. Developers publish, retrieve and download audio, visual and text models here, and adapt the models to different server chips. It also provides cloud services to power application-side reasoning. Google, Microsoft, Amazon and Nvidia have all been investors. A $235 million round of financing in 2023 gives a valuation of $4.5 billion, with Nvidia, Salesforce and Alphabet's Google all on the list. Last year, Nvidia proposed a US$500 million capital increase, corresponding to a valuation of US$7 billion, but Hugging Face rejected it on the grounds that it did not want a leading investor who could influence decision-making.
According to people familiar with the matter, negotiations between Nvidia and Hugging Face were initiated after the latter received an acquisition intention from another buyer. There are many potential buyers among the investors, and Salesforce is also on the inquiry list. Microsoft has been in contact with Hugging Face, but relevant discussions have not continued. Early investors also include Ronny Conway, Lux Capital and Addition.
2. Nvidia’s strategic judgment: Use open source to balance the trend of closed source chip self-research
The judgment of NVIDIA management is that if the open source model forms a climate, it can balance the path of self-developed server chips by closed source laboratories such as Anthropic and OpenAI and reduce dependence on NVIDIA graphics processors. The company has doubled down on its own open source series, Nemotron, and pledged to invest tens of billions of dollars in developing open source models.
This judgment faces new competitive signals. On August 25, OpenAI announced the evaluation of its self-developed inference chip Jalapeño, saying that the unit power consumption performance is better than NVIDIA Blackwell; however, the chip has not yet been fully compared with NVIDIA's latest generation Rubin, nor has it been deployed on a large scale. Huang Renxun told the media on the same day that he was not worried about such attempts and believed that Nvidia hardware is still the most practical for developers. Google is lobbying Nvidia's core customers to also use its tensor processors.
3. Finance and Valuation: Annualized revenue is approximately US$150 million, corresponding to 80 times forward revenue
Although Hugging Face is the main repository of open source models, its revenue scale is limited. The recent annualized recurring revenue is approximately US$150 million, which is usually calculated by multiplying the most recent month's revenue by 12. Two months ago, the same caliber was around $100 million. Based on this calculation, the consideration of US$12.9 billion is approximately 80 times the forward revenue. Its customers pay a fixed monthly fee, with additional charges based on computing power and storage usage. Co-founder and CEO Clem Delangue told the media in June this year that the number of paid subscribers will double in the first half of 2026; he recently said in a podcast that the company is "close to profitability."
4. Cloud business undertaking: Buying Hugging Face can reconnect NVIDIA’s contracted cloud services
According to reports, buying Hugging Face is equivalent to reconnecting Nvidia’s artificial intelligence cloud business, which it shrank a year ago. DGX Cloud once created a quasi-competition between Nvidia and Amazon Cloud Technology: Nvidia promised to invest US$13 billion to lease back its own accelerator cards to cloud vendors that have purchased chips, and then sublease them to companies such as Amgen and ServiceNow that build their own applications. It once believed that the cloud business could contribute US$150 billion in revenue in the future.
Nvidia also disclosed in its financial report on August 26 that it has committed US$36 billion in cloud service contracts as part of customer credit support. If the market weakens and customers are unable to sublease this part of the production capacity to a third party, NVIDIA may have to take over part of the computing power; when its own projects are not used up, the production capacity can be transferred to the inference needs of Hugging Face customers. The report did not specify the consideration structure or whether Hugging Face will maintain optimized support for non-Nvidia chips after the closing.
5. Continuation of intensive trading layout
The deal comes at the end of a round of intensive moves by Huang to target application companies, data center sites and power developers to lock in a hardware portfolio. Last week, NVIDIA signed a $6 billion tool license with the open source model company Poolside and offered employment to more than 100 of its employees; this summer, it acquired Kumo AI and Essential AI to supplement model training tools.
The company disclosed on Wednesday that it holds US$99 billion in equity investment and has committed to invest an additional US$25 billion as of the end of July. It has previously promised to invest US$30 billion in equity in OpenAI and provide US$108 billion in credit support for a large data center that OpenAI plans to lease in Ohio starting in 2028 and filled with Nvidia chips. Chief Financial Officer Colette Kress said in the financial statement that some people would call this type of arrangement revolving financing, but the company views it differently because the funds are supporting companies that "will become the largest technology companies in history." She also said about a quarter of revenue in the next fiscal year will come from artificial intelligence labs funded by Nvidia.
6. Unresolved matters and uncertainties
Security incidents may also have an impact on transactions. In July, OpenAI said that the model used for internal evaluation broke through isolation and entered the company's system as well as external systems such as Hugging Face. Hugging Face later said it used tools, including open source models, to deal with the intrusion. Whether the transaction consideration, personnel retention and platform neutrality terms will be adjusted as a result, the report did not give figures.
Both companies did not respond to requests for comment outside regular business hours. The report did not provide a regulatory review timetable, consideration payment methods or post-closing governance arrangements for the open source warehouse. Whether the agreement has been signed and whether it will be revoked before signing can only be determined by the official statement of both parties.
Comments