Nvidia 接近以 129 亿美元收购 Hugging Face

据 The Information 周三晚间报道，知情人士透露，Nvidia 已同意以 129 亿美元收购 Hugging Face。Business Insider 此前在周末最先报道称，Hugging Face 正收到潜在收购方的兴趣；该媒体周三晚间又报道称，这项对公司估值超过 130 亿美元的谈判尚未形成已签署协议，且仍可能告吹。

TechCrunch 此前已向 Nvidia 和 Hugging Face 双方寻求置评，但两家公司均尚未回应。（Nvidia 的沉默尤其值得注意，因为该公司过去通常会迅速回应其认为不准确的报道。）

或许这从一开始就是注定的。Hugging Face 成立于 2016 年，是开发者分享和下载开源 AI 模型的最受欢迎平台之一。收购它将使 Nvidia 在开源 AI 世界中获得一个强大的立足点；而此时，开源开发者正竭尽全力追赶 Anthropic 和 OpenAI 等公司推出的闭源 AI 系统。

Nvidia 为什么会想要这么做？最明显的原因，是保护其在 AI 芯片领域的主导地位。从外部看，即便 Nvidia 正积极推进芯片发布节奏，这一地位似乎仍面临越来越大的风险。几乎所有最大的闭源 AI 实验室——OpenAI、Google、Amazon 和 Anthropic——如今都在开发自己的 AI 芯片，以降低对 Nvidia 的依赖。一个繁荣的开源 AI 模型生态，能为客户提供更多替代这些闭源实验室的选择，进而让更大一部分市场继续依赖 Nvidia 的硬件。这也是 Nvidia 已经投入数百亿美元开发自有开源 AI 模型的原因。

我们是否应该对 Hugging Face 作为独立公司的日子似乎所剩不多感到意外？其实不必。围绕华盛顿官员据称正在考虑限制开放权重模型的争论已持续数月，而 Hugging Face CEO Clem Delangue 在今年大部分时间里都公开支持 Nvidia 推动开源的立场。（中国实验室，例如 Moonshot AI，已发布了 Kimi K3 等系统；这些系统在基准测试中可与领先的美国模型匹敌，但运行成本低得多。因此，华盛顿对竞争和国家安全的担忧持续升温。一些闭源实验室的批评者，例如白宫顾问 David Sacks，认为这些担忧是由 Anthropic 和 OpenAI 这一“二元垄断”所煽动的。）

例如，Delangue 本月早些时候在 CBS 节目《Face the Nation》中表示，Hugging Face 在遭遇网络攻击后，使用了一款经 Nvidia 修改的中国开源模型进行防御；他还提及一封近期公开信。该信由 Nvidia CEO Jensen Huang 与包括 Hugging Face 在内的另外 24 家公司签署，呼吁美国政府支持开放模型，而不是对其加以限制。7 月下旬，在另一场 CNBC 采访中，Delangue 也援引同一封信表达了类似观点，并警告称，中国“显然正在主导”开源 AI。

这笔交易还将标志着 Nvidia 在云计算领域的某种回归。据报道，Nvidia 大约一年前缩减了其自有云业务 DGX Cloud。但据 The Information 称，收购 Hugging Face——该公司已经帮助开发者利用租用的计算资源运行 AI 模型——可以让 Nvidia 无须从零开始便重返这一市场。

其中还存在一层财务安全保障。Nvidia 曾承诺协助客户承担数百亿美元云计算交易的成本。如果这些客户最终没有用完其签约的全部算力，Nvidia 可能会被迫接手这些资源。拥有 Hugging Face 后，Nvidia 将能够把这些闲置算力出售给 Hugging Face 的客户。

这一价格较 Hugging Face 上一次已知估值实现了巨大跃升。该公司在 2023 年的一轮融资中筹集了 2.35 亿美元，当时估值为 45 亿美元。这轮融资由 Salesforce Ventures 领投，Alphabet 旗下 GV、IBM Ventures 以及 Nvidia 本身等也参与了投资。

这也并非 Hugging Face 首次面对 Nvidia 的报价。英国《金融时报》此前报道称，Hugging Face 在去年年底拒绝了 Nvidia 提出的5 亿美元投资要约，该要约将使公司估值达到 70 亿美元。Hugging Face 当时表示，它不希望有一家能够影响其决策的主导性投资者。

至于它如今为何可能答应，有人或许会认为，被收购不同于接受单一巨型支持者的投资——后一种情形通常意味着在持续增长压力下让渡控制权。

在 AI 行业中，按营收衡量，Hugging Face 仍是一家相对较小的企业。The Information 报道称，该公司近期的年化营收约为 1.5 亿美元，高于仅两个月前的约 1 亿美元。

正如 Delangue 上个月对 TechCrunch 所说，这种增长已让公司“接近盈利”。但对于一家这一规模的公司而言，接近 130 亿美元的价格意味着极高的估值倍数，也很难拒绝。

最后，这笔交易将让 Hugging Face 获得 Nvidia 雄厚得多的资金支持；与此同时，其他 AI 基础设施竞争者也开始被纳入更大型企业，Stripe 最近收购 OpenRouter 的交易就是一个例子。OpenRouter 成立于 2023 年初，帮助客户根据自身需求和预算，选择不同的 AI 模型来完成不同任务。

OpenRouter 在 5 月进行 B 轮融资时的估值仅为 13 亿美元。据报道，Stripe 在本月早些时候支付了超过70 亿美元，将其纳入旗下。

Nvidia has agreed to buy Hugging Face for $12.9 billion, The Information reported Wednesday night, citing a source familiar with the matter. Business Insider, which first reported over the weekend that Hugging Face was fielding takeover interest, reported Wednesday night that the talks — which would value the company at more than $13 billion — had not yet produced a signed agreement and could still atomize.

TechCrunch reached out earlier to both Nvidia and Hugging Face for comment, and neither has yet responded. (Nvidia’s silence is particularly noteworthy here as the company has moved quickly in the past to address reports it considers inaccurate.)

Maybe it was destined from the start. Hugging Face, founded in 2016, is one of the most popular hubs where developers share and download open-source AI models. Buying it would give Nvidia a strong foothold in the world of open-source AI, right as open-source developers are doing their level best to catch up to closed AI systems from companies like Anthropic and OpenAI.

Why would Nvidia want that? Most obviously, it comes down to protecting its dominance in AI chips, which, from the outside at least, appears increasingly at risk, even with Nvidia’s aggressive chip-release schedule. Pretty much all of the biggest closed-source AI labs (OpenAI, Google, Amazon, and Anthropic) are now in the process of building their own AI chips to lessen their reliance on Nvidia. A thriving ecosystem of open-source AI models gives customers more alternatives to those closed labs, which in turn keeps more of the market dependent on Nvidia’s hardware. That’s also why Nvidia has already poured tens of billions of dollars into building its own open-source AI models.

Should we be surprised that Hugging Face’s days as an independent outfit appear numbered? Not really. Hugging Face CEO Clem Delangue has spent much of this year publicly aligned with Nvidia’s open-source push, amid a debate that has been building for months, as Washington officials reportedly weighed restrictions on open-weight models. (Chinese labs like Moonshot AI had released systems — like its Kimi K3 model — that matched leading U.S. models on benchmarks while costing a lot less to run, and talk of competitive and national-security concerns grew in Washington as a result. Some critics of closed labs, like White House advisor David Sacks, suggested those fears were being fanned by the “duopoly” of Anthropic and OpenAI.)

In an appearance on CBS’s “Face the Nation” earlier this month, for example, Delangue said Hugging Face used an Nvidia-modified version of a Chinese open-source model to defend itself after a cyberattack and pointed to a recent letter — signed by Nvidia CEO Jensen Huang and 24 other companies, including Hugging Face — urging the U.S. government to support open models rather than restrict them. In a separate CNBC interview in late July, Delangue made similar points, citing that same letter while warning that China is “clearly dominating” open-source AI.

The deal would also mark something of a comeback for Nvidia in cloud computing. Nvidia reportedly scaled back its own cloud business, called DGX Cloud, about a year ago. But according to The Information, owning Hugging Face — which already helps developers run their AI models using rented computing power — could give Nvidia a way back into that market without starting from scratch.

There’s also a financial safety net at play. Nvidia has promised to help cover the cost of tens of billions of dollars in cloud computing deals for its customers. If those customers end up not using all the computing power they signed up for, Nvidia could get stuck with it. Owning Hugging Face would give Nvidia the ability to sell that unused capacity to Hugging Face’s customers.

The price marks a huge jump from Hugging Face’s last known value. The company raised $235 million in 2023 in a funding round that valued it at $4.5 billion. That round was led by Salesforce Ventures, with money also coming from Alphabet’s GV, IBM Ventures, and Nvidia itself, among others.

This wouldn’t be Hugging Face’s first brush with an Nvidia offer, either. Hugging Face turned down a $500 million investment offer from Nvidia late last year that would have valued it at $7 billion, the Financial Times previously reported. Hugging Face said at the time it didn’t want a dominant investor that could sway its decisions.

As for why it would say yes now, one could argue that a buyout is different from taking on one giant backer — a scenario that often means ceding control while being pressured to continue growing.

Hugging Face is also still a comparatively small business by revenue in the world of AI. The Information reported it was recently generating about $150 million a year in revenue, up from roughly $100 million just two months earlier.

That growth has enabled the company to get “close to profitability,” as Delangue told TechCrunch last month. Still, a price near $13 billion would be a massive multiple for a company this size and hard to resist.

Not last, the deal would give Hugging Face access to Nvidia’s much deeper pockets just as other, AI infrastructure competitors start to get pulled into other outfits, as suggested by Stripe’s recent deal to acquire OpenRouter, a startup founded in early 2023 that helps customers select different AI models to perform different tasks depending on their needs and budget.

OpenRouter was valued at just $1.3 billion back in May during its Series B round. Stripe reportedly paid more than $7 billion to make it its own earlier this month.