经过数周沸沸扬扬的传闻,Nvidia 今日证实,已以 129.3 亿美元收购 Hugging Face。Hugging Face 平台托管着 300 万个模型、100 万个应用,服务超过 1800 万名开发者,并拥有 50 万个数据集。
在一篇博文中,Nvidia 首席执行官黄仁勋表示,Hugging Face 将继续支持开源和开放权重模型,并将致力于扩大开发者的使用渠道。
黄仁勋表示:“Hugging Face 仍将是面向整个 AI 生态系统的开放平台。开发者可以自由选择他们想要的模型、框架、云与推理服务提供商,以及计算平台。在 Hugging Face 上构建或部署,并不强制要求使用 Nvidia 的计算资源。”
黄仁勋盛赞了 Nvidia 的贡献,并表示这家芯片公司已在 Hugging Face 上发布了超过 500 个模型和 250 个开放数据集。他主张,公司构建开放模型,是为了让全球开发者都能使用它们。
对于 Nvidia 这样一家在 AI 开发和推理领域占据主导地位的硬件平台公司而言,一个由其掌控的开放生态系统是有利的,因为它可以打造一个适配自家芯片的平台。此外,正如 TechCrunch 早前报道的那样,Nvidia 将能够把闲置算力与 Hugging Face 的产品打包,销售给企业客户。
据 Crunchbase 数据,Hugging Face 成立于 2016 年,迄今已累计融资超过 3.95 亿美元。该公司最近一轮融资是在 2023 年,由 Salesforce Ventures 领投,筹集了 2.35 亿美元,Google、Amazon、IBM 和 Nvidia 均参与了投资。
Hugging Face 首席执行官 Clem Delangue 在 X 上发帖,感谢社区证明了该公司可以成为闭源 API 的替代选择。
“但要实现更大规模的发展,我们需要更多的算力、更多的支持、更多的协作和更高的可见度。这就是为什么我们去与 Jensen 交谈,而他主动提出要与我们在这方面展开合作,”Delangue 表示。
Hugging Face 的地位日益凸显,每天都有更多模型发布。据《金融时报》报道,去年该公司拒绝了 Nvidia 提出的 5 亿美元收购要约。上个月,The Information 报道称,Hugging Face 的年化收入已达到 1.5 亿美元。
Delangue 在 7 月接受 TechCrunch 采访时表示,其增长速度正帮助该平台“接近盈利”。
Nvidia 的黄仁勋一直是开源模型的坚定支持者。他撰写了一封由多家机构联署的信函,倡导开放权重模型,以强化美国在 AI 领域对抗中国等竞争对手的地位。
该公司正在大力投资模型开发。上个月,《华尔街日报》报道称,Nvidia 与编程初创公司 Poolside 达成了一项 60 亿美元的交易,用于开发开源模型。在最近一次财报电话会议上,该公司表示已向 AI 前沿实验室注资超过 500 亿美元。
黄仁勋在回答一位分析师的问题时力挺开源模型,并表示几乎所有开源模型都运行在 Nvidia 的硬件上。他还强调了这些模型在网络安全领域的重要性。
“前沿模型至关重要的领域之一就是网络安全。你会看到,许多网络安全公司正是借助前沿模型,才能构建大规模分布式、持续运行的自主网络安全系统来进行防御。这些公司正在涌现,其中不乏一些非常出色的企业。没有开源模型,它们根本做不到这一点。因此,开源模型既取得了巨大成功,也终于达到了前沿水平,同时它们对美国经济至关重要,对世界经济也至关重要。”他说。
今年7月,Delangue表示,在专有模型未能保护平台之后,Nvidia的开源模型帮助Hugging Face抵御了网络攻击。而就在那几天前,OpenAI承认其未发布的模型曾突破Hugging Face的防御。
After weeks of swirling rumors, Nvidia confirmed today that it has acquired Hugging Face for $12.93 billion. Hugging Face’s platform hosts three million models, one million applications used by over 18 million developers, and half a million datasets.
In a blog post, Nvidia’s CEO Jensen Huang said that Hugging Face will continue to support open-source and open-weight models and will work on expanding developer access.
“Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want. Nvidia compute will not be required to build on or deploy through Hugging Face,” said Huang.
Huang touted Nvidia’s contributions and said that the chip company has released more than 500 models and 250 open datasets on Hugging Face. He argued that the company builds open models to get developers across the world to use them.
For a company like Nvidia, which is a dominant hardware platform for AI development and inference, an open ecosystem that it controls is beneficial as it can create a platform suited for its chips. Plus, as TechCrunch wrote earlier, Nvidia will be able to sell its unused capacity to enterprise customers packaged with Hugging Face’s offering.
Hugging Face was founded in 2016 and has raised over $395 million in funding to date, according to Crunchbase. The company’s last round was in 2023, when it raised $235 million led by Salesforce Ventures, with investments from Google, Amazon, IBM, and Nvidia.
In a post on X, Hugging Face CEO Clem Delangue thanked the community for showing the company could be an alternative to closed-source APIs.
“But for it to happen at [a] larger scale, it needs more compute, more support, more collaboration, and more visibility. That’s why we went to talk to Jensen, who offered to do exactly that with us,” said Delangue.
Hugging Face has risen in prominence, with more models being released every day. Last year, the company rejected a $500 million deal from Nvidia, according to the Financial Times. Last month, The Information reported that Hugging Face is clocking $150 million in annualized revenue.
In an interview with TechCrunch in July, Delangue said that its growth rate is helping the platform to get “close to profitability.”
Nvidia’s Huang has been a strong proponent of open models. He wrote a letter, co-signed by several other organizations, to advocate for open-weight models to strengthen the US’s position against rivals like China in the AI sector.
The company is heavily investing in model development. Last month, The Wall Street Journal reported that it struck a $6 billion deal with coding startup Poolside to develop open models. During its recent earnings call, the company said it has infused over $50 billion into AI frontier labs.
Huang pitched open models while answering one of the analysts and said that almost all open models run on Nvidia hardware. He also signaled the importance of these models in cybersecurity.
“One of the areas where frontier models are vital is cybersecurity. You see the number of cybersecurity companies that are enabled by frontier models so that they could have massively distributed, continuously running autonomous cybersecurity systems to defend. Those companies are emerging. There are some amazing companies. They couldn’t do it without open models. And so open models is both incredibly successful and have finally reached the frontier, but they’re also vital to the American economy. It’s vital to the world economy,” he said.
In July, Delangue said that Nvidia’s open model helped Hugging Face defend against cyberattacks after proprietary models failed to protect the platform. Days before that, OpenAI admitted that its unreleased model breached Hugging Face.