Nvidia is in late-stage discussions to acquire Hugging Face, the open-source AI model repository and inference platform, for approximately $14 billion, according to market sources. The transaction would mark Nvidia's largest software acquisition and its most direct move into the model hosting and deployment layer that sits atop its H100 and H200 GPU infrastructure. Terms remain undisclosed, but the timing—arriving days after Anthropic's $35 billion Lambda computing commitment backed by Nvidia capital—suggests the chipmaker is accelerating a two-pronged strategy: funding the compute buyers while owning the software rails.
Hugging Face operates the dominant open repository for pre-trained models, hosting over 500,000 models and 250,000 datasets as of March 2025. The platform processed more than 1 billion model downloads in the trailing twelve months. Its inference API, introduced in 2022 and expanded in 2024, now serves enterprises deploying fine-tuned models without managing GPU clusters directly. Nvidia already integrates Hugging Face's Transformers library into its NeMo toolkit, but acquisition would grant ownership of the user relationships, usage telemetry, and the commercial inference contracts that currently run on multi-cloud infrastructure. The purchase price implies a valuation near 35x trailing revenue, assuming Hugging Face's rumored $400 million run rate holds through mid-2025.
The deal matters because it closes Nvidia's exposure gap in the software margin pool. Chips carry hardware economics; model hosting and API access carry software multiples. Anthropic's Lambda agreement—structured as a $35 billion compute commitment with Nvidia holding the data center lease and investing in Lambda's equity—already demonstrated Nvidia's willingness to finance the demand side. Acquiring Hugging Face extends that logic upstream: Nvidia would own both the silicon and the marketplace where enterprises discover, test, and deploy models at scale. The combination also hedges against OpenAI's and Anthropic's vertical integration, both of which increasingly offer managed inference that bypasses third-party platforms. If closed-source frontier labs own their entire stacks, Nvidia's silicon risks becoming a commoditized input. Owning Hugging Face preserves Nvidia's position as the default substrate for the open-weight ecosystem, which generates the majority of enterprise fine-tuning workloads.
Allocators should monitor three near-term events. First, whether Hugging Face's enterprise customers—including Salesforce, Bloomberg, and ServiceNow—retain existing multi-cloud inference agreements or migrate toward Nvidia DGX Cloud under revised commercial terms. Second, Anthropic's Lambda deal includes a clause for capacity expansion tied to model scaling; if Hugging Face inference volumes shift onto Nvidia-backed infrastructure, the Lambda agreement becomes a de facto captive customer for Hugging Face's API business. Third, regulatory filings in the U.S. and EU, expected within 45 days if the deal proceeds, will reveal whether Nvidia structures the transaction as an outright acquisition or a majority investment with earnout provisions tied to revenue milestones. The distinction shapes how quickly Nvidia can bundle Hugging Face into its enterprise AI software suite, currently priced at $4,500 per GPU per year.
The $14 billion figure, if accurate, represents 4.2% of Nvidia's current market capitalization and would be financed from cash, given the company's $28 billion balance as of January 2025.