英伟达以129.3亿美元收购Hugging Face，承诺保持开源平台性质

英伟达宣布以129.3亿美元收购开源AI平台Hugging Face 1345。这笔交易包括约119亿美元的投资者付款和最高10亿美元的员工留任计划 6。

Hugging Face成立于2016年 1，已成为AI领域重要的基础设施平台，拥有超过1800万开发者、300万个模型、100万个应用和50万个数据集 35，服务超过20万家企业 5。英伟达承诺将Hugging Face保持为开放平台 24，确保开发者可自由选择模型、框架和计算平台，并非必须使用英伟达芯片 35。英伟达首席执行官詹森·黄表示："我们将共同扩展Hugging Face平台、加强基础设施并为全球开发者和机构扩大AI访问权限" 4。

值得注意的是，Hugging Face去年曾拒绝过英伟达的5亿美元收购报价 3。根据之前的报道，Hugging Face年化收入达1.5亿美元 3，其最近一轮融资为2023年由Salesforce Ventures主导的2.35亿美元 3。英伟达已贡献了500多个开源模型和250个开放数据集到该平台 5，是该平台最大的模型和数据贡献者 5。

Nvidia has announced its agreement to acquire Hugging Face, a leading open-source artificial intelligence platform, in a deal valued at approximately $12.93 billion 134. The acquisition represents one of Nvidia's largest transactions and underscores the chipmaker's strategic push into the broader AI software and developer ecosystem.

Hugging Face, founded in 2016 1, has grown into a central hub for AI development, hosting three million models, one million applications, and 500,000 datasets that serve over 18 million developers, researchers, and creators 35. The platform also counts more than 200,000 companies as enterprise users 5. Nvidia committed to maintaining Hugging Face as an open platform where developers retain the freedom to choose their preferred models, frameworks, and computing infrastructure 35. According to Nvidia CEO Jensen Huang, "Together, we will scale Hugging Face's platform, strengthen its infrastructure and expand access to AI for developers and institutions worldwide" 4.

The transaction structure includes approximately $119 billion in investor payments and up to $1 billion in employee retention arrangements 6. Hugging Face had previously rejected a $500 million acquisition offer from Nvidia 3 and raised $235 million in its most recent funding round in 2023, led by Salesforce Ventures 3. The platform reported annualized revenue of $150 million as of the previous month 3. Nvidia noted that it is already the largest contributor of open models and data to Hugging Face, with over 500 open-source models and 250 open datasets on the platform 5.