Nvidia Reportedly Acquires Hugging Face for $12.9 Billion, Intensifying Battle for AI Middle Layer

Source: wallstreetcn

Nvidia is acquiring open-source AI platform Hugging Face for $12.9 billion, a valuation of roughly 80 times its annualized revenue. The move aims to control a core node of the open-source ecosystem, hedge against the threat of closed-source giants like OpenAI developing their own chips, and provide a new entry point for its cloud business.

According to The Information, on Aug. 26 local time, Nvidia agreed to acquire open-source AI platform Hugging Face for $12.9 billion.

Just days earlier, Stripe swallowed AI model routing platform OpenRouter for over $8 billion.

The two deals landing within the same window signal a clear new trend: AI "middle layer" infrastructure is becoming the core target of strategic acquisitions by major companies.

For investors, the investment implication is direct. Profits at the model layer are being squeezed by competition; what can sustainably generate cash flow is the "pipeline" between models and users. Whoever controls these pipelines holds the entry point to the next round of AI monetization.

What Is Nvidia Buying for $12.9 Billion?

According to an exclusive report by The Information, Nvidia is acquiring Hugging Face for $12.9 billion, about 80 times its annualized revenue of $150 million—a multiple that is rare even in the current tech M&A market.

Hugging Face, founded 10 years ago, hosts over 1 million community-contributed AI models, with over 18 million monthly active visitors, about 5 million registered users, and more than 2,000 enterprises paying for its Enterprise Hub.

In the first half of this year, its paid subscriber count doubled, and CEO Clem Delangue publicly stated the company is "close to profitability."

Annualized revenue also grew from about $100 million to $150 million in just a few months, with a clear acceleration. One driver is the rapid improvement of Chinese open-source models such as Zhipu, Moonshot AI, and DeepSeek, prompting more U.S. enterprises to turn to open-source solutions.

But this deal is clearly not purely financial logic. What Nvidia is buying is the core node of the entire open-source AI ecosystem—who is training what models, how technology trends are evolving, a channel to 13 million developers, and a neutral brand that is hard to replicate in the open-source community.

Nvidia's Strategic Calculus: Hedging Against Closed-Source Threats

Nvidia's acquisition logic is quite clear at the strategic level.

Currently, closed-source AI giants like OpenAI and Anthropic are actively developing their own AI server chips to reduce dependence on Nvidia hardware. OpenAI even released benchmarks this week for its self-developed "Jalapeño" AI chip, claiming performance surpassing Nvidia's flagship Blackwell chip.

Nvidia executives' judgment: the prosperity of open-source models is the best defense against closed-source threats. The more open-source model developers there are and the more successful they are, the harder it is for Nvidia's GPU demand to be cut off by a single closed-source force.

Hugging Face happens to be the core hub of the open-source model ecosystem—controlling it means controlling the distribution channel and technical intelligence of the entire open-source ecosystem.

Meanwhile, Nvidia has previously committed billions of dollars to develop its own Nemotron series of open-source models, and acquiring Hugging Face can be seen as a direct extension of that strategy.

Nvidia's Cloud Computing Revival and Risk Hedging

This deal has another financial dimension. Nvidia once launched its DGX Cloud business, committing to lease $13 billion worth of its own AI chips and sublease them to enterprise customers, but later scaled back.

Hugging Face effectively provides Nvidia with a new entry point to restart its cloud business: if Nvidia has excess computing capacity in its $36 billion cloud service contracts, it can directly use that capacity to support Hugging Face customers' AI applications, avoiding idle assets.

Nvidia disclosed in its latest earnings report that it currently holds $99 billion in equity investments, and as of the end of July had committed an additional $25 billion.

It previously announced a $30 billion equity investment in OpenAI and provided $108 billion in credit support to help OpenAI build a massive data center in Ohio.

Nvidia's CFO said that next year, about a quarter of revenue will come from AI labs financially supported by Nvidia.

The Bigger Picture: Strategic Value of the AI "Middle Layer"

Placing this deal in a broader timeline, its significance goes far beyond a single acquisition.

Just days before the Hugging Face news broke, Stripe acquired AI model routing platform OpenRouter for over $8 billion—the latter was valued at only $1.3 billion in its Series B round in May, and was bought at a premium of about 5.4 times within three months.

OpenRouter provides a unified API, connecting to over 400 models from more than 80 providers, claiming to serve 8 to 10 million developers, routing trillions of tokens weekly, with token consumption growing at a compound rate of about 9% per week this year.

The underlying logic of the two deals is the same: when competition at the model layer has become white-hot and profits are squeezed, the "pipeline" between models and users becomes the truly scarce asset.

This is the real-world projection of the "selling shovels" logic in the AI era—model companies fight each other and keep cutting prices, while middleware such as model hosting, routing, and inference optimization can theoretically generate steady income regardless of who wins.

But this logic is not flawless: if the market converges to a few mainstream models in the future, the value of the routing layer will be greatly diminished; if model distribution shifts from "downloading weight files" to pure API form, the necessity of model hosting platforms will also be fundamentally challenged.

Whether the value of these middle-layer companies is a "permanent positional value" like Cloudflare, or just a "time-window value" before AI infrastructure solidifies, remains undetermined.