NVIDIA has announced an agreement to acquire the open-source AI platform Hugging Face for $12.93 billion, marking its largest single acquisition to date. Jensen Huang, CEO of NVIDIA, stated that the company will maintain the openness of the Hugging Face platform, allowing developers to freely upload and download models and datasets without mandating the use of NVIDIA hardware.
$NVIDIA (NVDA.US)$ It is betting on future influence within the open-source AI ecosystem at a cost in the range of $13 billion.
On Thursday, NVIDIA announced an agreement to acquire the open-source AI platform Hugging Face for $12.93 billion, marking its largest single acquisition to date. Jensen Huang, CEO of NVIDIA, stated that the company will maintain the openness of the Hugging Face platform, allowing developers to freely upload and download models and datasets without mandating the use of NVIDIA hardware.
This acquisition comes at the peak of a wave of M&A activity in AI infrastructure. Just days before the announcement, payment giant Stripe acquired OpenRouter, an AI model routing platform, for over $8 billion. The consecutive closure of these two major deals within the same time frame clearly outlines a shared logic: as competition intensifies and margins narrow at the model layer, the "middle-layer" infrastructure situated between models and users is becoming a core target for strategic acquisitions by major corporations.
Behind the Sky-High Valuation: NVIDIA Is Buying More Than Just Revenue
From a financial perspective, the valuation multiple for this transaction is rare.
Hugging Face generates approximately $150 million in annualized revenue. The $12.93 billion acquisition price implies a revenue multiple of roughly 80x, which remains high even by current standards in the technology M&A market.
However, NVIDIA's rationale clearly extends beyond pure financial considerations.
Founded in 2016, Hugging Face hosts over 3 million AI models, 500,000 datasets, and 1 million applications on its platform. It boasts a user base of more than 18 million developers, researchers, and creators, with over 200,000 enterprises using the platform to discover, evaluate, and deploy AI solutions.
In the first half of this year, its paid subscriber base doubled. CEO Clem Delangue publicly stated that the company is "close to profitability," with annualized revenue rising significantly from approximately $100 million to $150 million in just a few months.
What NVIDIA has effectively acquired are the core nodes of the entire open-source AI ecosystem—including distribution channels reaching millions of developers, technical intelligence on model training trends, and a neutral brand reputation that is difficult to replicate within the open-source community.
Jensen Huang stated that NVIDIA is currently the largest contributor of open-source models and data to Hugging Face, having released over 500 models and 250 open-source datasets on the platform.
Strategic Hedging: The Open-Source Ecosystem as NVIDIA’s Moat
This acquisition carries a clear defensive intent at the strategic level. Closed-source AI giants such as OpenAI and Anthropic are actively developing their own AI chips in an effort to reduce reliance on NVIDIA hardware.
This week, OpenAI released benchmarks for its self-developed "Jalapeño" AI chip, claiming its performance surpasses that of NVIDIA’s flagship Blackwell chip.
NVIDIA’s assessment is that the expansion of the open-source model ecosystem serves as the most effective defense against these threats: the larger and more active the community of open-source model developers becomes, the harder it is for any single closed-source entity to cut off demand for NVIDIA GPUs.
As the core distribution hub for open-source models, controlling Hugging Face means controlling the gateway to the entire open-source ecosystem.
In his statement, Jensen Huang emphasized that "NVIDIA compute is not a prerequisite for building or deploying on Hugging Face," and committed to the platform’s continued support for multi-cloud and multi-accelerator development and deployment.
This move aims to alleviate concerns within the developer community regarding the platform’s neutrality following NVIDIA’s acquisition. Previously, NVIDIA had committed billions of dollars to developing its own Nemotron series of open-source models; the acquisition of Hugging Face can be viewed as a direct extension of this strategy.
Cloud Business Realignment: Seeking Outlets for Excess Compute Capacity
This transaction carries additional financial implications.
NVIDIA previously launched its DGX Cloud business, committing to lease $13 billion worth of its proprietary AI chips and sublease them to enterprise clients, but later gradually scaled back the operation.
Hugging Face offers NVIDIA a new entry point to revive its cloud business: if NVIDIA experiences excess computing capacity under its $36 billion cloud services contract, it can directly allocate this compute power to support AI applications for Hugging Face customers, thereby avoiding asset idleness.
In its latest financial report, NVIDIA disclosed that it currently holds $99 billion in equity investments, with an additional commitment of $25 billion in investments as of the end of July.
Previously, NVIDIA also announced a $30 billion equity investment in OpenAI, along with $108 billion in credit support, to fund the latter’s construction of large-scale data centers in Ohio. NVIDIA’s CFO stated that approximately one-quarter of next year’s revenue will come from AI laboratories financially supported by NVIDIA. Meanwhile, performance guidance released by NVIDIA last week indicates that revenue for fiscal year 2028 is projected to grow by approximately 70%.
The Battle for Value in the AI "Middle Layer"
NVIDIA’s deal with Stripe is not an isolated case; rather, it reflects a deep restructuring of the AI industry chain.
OpenRouter provides a unified API interface, aggregating over 400 models from more than 80 suppliers, serving 8 to 10 million developers, and routing trillions of tokens weekly. Since the beginning of this year, token consumption has maintained a weekly compound growth rate of approximately 9%.
Stripe acquired it for over $8 billion, representing a premium of approximately 5.4 times over OpenRouter’s $1.3 billion valuation during its Series B financing round in May of this year.
Both transactions point to the same conclusion: as competition at the model layer intensifies, "middleware" services such as model hosting, routing and scheduling, and inference optimization can theoretically ensure stable returns regardless of market conditions. This is a practical manifestation of the "selling shovels" logic in the AI era.
However, this logic is not without risks. If the future market converges on only a few mainstream models, the value of the routing layer will be significantly diminished. Furthermore, if model distribution shifts from downloading weight files to a pure API-based model, the necessity of model hosting platforms will face fundamental challenges.
There is currently no market consensus on whether these mid-layer assets represent "permanent location value" akin to Cloudflare, or "time-window value" during the period when AI infrastructure has not yet solidified.
Editor/lambor