엔비디아는 오픈소스 AI 모델과 개발자 도구 중심 허브를 확보하기 위해 허깅페이스를 약 129억 달러에 인수하기로 공식 합의했다.

Nvidia plans to acquire Hugging Face for approximately $12.9 billion, pledging to maintain the platform’s open and hardware-neutral architecture. The acquisition grants the chipmaker control over the primary distribution hub for open-source AI models while establishing a new direct sales channel for its compute infrastructure. Nvidia CEO Jensen Huang announced the deal on September 3, 2026, via a company blog post, following reports that emerged in late August. According to the SEC filing, the transaction comprises roughly $11.9 billion in purchase price plus a stock retention program worth up to $1 billion. The deal is scheduled to close in the first half of 2027, pending regulatory approval.

Hugging Face serves as the central repository for open AI development. Nvidia states that more than 18 million developers utilize the platform to share over 3 million models, 500,000 datasets, and 1 million applications, with more than 200,000 companies relying on the service. Huang emphasized that Hugging Face will remain fully open to all users. Nvidia hardware will not be mandatory, and support for alternative cloud providers and chip architectures will continue. By its own metrics, Nvidia is already the platform’s largest contributor, having uploaded over 500 models and more than 250 datasets.

The two sides have offered conflicting accounts regarding how the deal originated. Huang stated that co-founder Clem Delangue approached him to discuss the company’s next chapter. Conversely, co-founder Thomas Wolf wrote on LinkedIn that Huang extended the offer to Delangue, framing it as an opportunity to evolve the hub into an open, independent, and hardware-neutral ecosystem. Wolf described Nvidia as the ideal partner for the company’s broader mission, which spans open-weight models, robotics, and scientific computing. “Nothing changes for users today,” he noted, adding that open AI stands at a critical juncture where scale and compute access are becoming increasingly decisive factors.

Nvidia’s core revenue stems from compute infrastructure, and virtually all developers using third-party AI APIs ultimately route their workloads through Nvidia chips. However, several major providers—including Google, Amazon, OpenAI, and most recently Anthropic—are now designing proprietary accelerators to reduce dependency. Open models, by contrast, operate across diverse environments such as public clouds, private enterprise networks, universities, and government agencies, serving a customer base that does not manufacture its own silicon. Summarizing the strategic logic to Axios, Huang stated plainly: “Free AI should be great for hardware.”

Nvidia has long maintained its own open-model initiatives. Through its Nemotron program, the company publishes model weights, substantial portions of training data, and complete training and post-training methodologies. It also leads the Nemotron Coalition, collaborating with Mistral AI, Thinking Machines Lab, Perplexity, Black Forest Labs, Cursor, LangChain, Reflection AI, and Sarvam on a shared open model. While these efforts do not yet place Nvidia ahead of leading Chinese models, its architectures deliver superior performance in specific workloads. Nvidia also leverages Chinese innovations by offering custom NVFP4-optimized versions of Kimi K2.6 and GLM-5.1 tailored to its hardware. Acquiring Hugging Face would grant Nvidia significant influence over the premier showcase for this ongoing competitive race.

Beyond model distribution, Hugging Face also provides hosted compute services to run them. Nvidia previously explored a similar path but scaled back its DGX Cloud offering to avoid competing directly with its enterprise clients for rental capacity. Today, it channels inference and training jobs through the Lepton marketplace, which routes workloads to partners including CoreWeave, Lambda, and Nebius; Hugging Face integrates with this ecosystem via a dedicated cluster service. In late July, Nvidia reported $36 billion in committed agreements with AI cloud partners. Under these contracts, Nvidia’s exposure decreases as partners resell capacity to third parties, and the company partially underwrites unused inventory. Integrating a massive developer hub as an additional distribution channel would significantly strengthen this commercial framework.