AI 平台與基礎設施
NVIDIA Acquires Hugging Face for $12.93 Billion, Pledges to Maintain Multi-Cloud and Multi-Accelerator Support
NVIDIA has agreed to acquire Hugging Face, gaining control of a key distribution hub for models, datasets, applications, and inference services. The company has pledged that the platform will not be tied to NVIDIA hardware, but its governance, ranking practices, and cross-hardware optimization after the deal will require continued scrutiny.
NVIDIA announced on September 3 that it would acquire Hugging Face for $12.9303 billion. This is more than the acquisition of a model company: Hugging Face Hub hosts more than 3 million models, 500,000 datasets, and 1 million applications. It also connects Transformers, model cards, evaluations, training services, and third-party inference providers, effectively serving as the registry and distribution layer of the open AI software supply chain.
NVIDIA explicitly pledged that developers will remain free to choose their models, frameworks, clouds, inference services, and computing platforms. Using the Hub or deploying assets hosted on it will not require NVIDIA GPUs, and the platform will continue to support multiple clouds and accelerator architectures. This commitment matters because toolchains for AMD, Intel, Apple Silicon, Google TPUs, and various domestically produced chips all depend on whether model formats, converters, containers, and examples receive equivalent maintenance.
One potential technical benefit is that NVIDIA could directly apply its computing resources and inference engineering expertise to the Hub’s reliability, security scanning, model evaluation, and deployment services. The path from model upload to quantization, compilation, and endpoint deployment could become shorter, while popular models might receive TensorRT, CUDA kernel, or multi-node inference optimizations more quickly. At the same time, however, this creates vertical integration: a single company would control a leading accelerator platform, its software stack, and a major gateway for discovering and deploying models.
Engineering teams should look beyond assurances that the platform will “remain open.” They should monitor whether search and recommendations favor NVIDIA assets, whether test coverage for non-CUDA backends is maintained, whether the ranking and pricing of Inference Providers remain neutral, and whether the terms governing model and dataset downloads, mirrors, and APIs change. Organizations that rely heavily on the Hub should also preserve model weights, model cards, dataset revisions, and build environments, while validating self-hosted mirrors and alternative registry workflows to prevent changes in a single platform’s governance from becoming a supply-chain risk.