Nvidia Corp. today confirmed that it has agreed to buy Hugging Face Inc. for just over $12.93 billion.

CNBC reported that the acquisition talks began a few weeks ago. Rumors that a deal was in the works surfaced last Monday, when sources tipped off Business Insider about the discussions. Two days later, The Information reported that Nvidia had clinched the deal.

Some of the $12.9 billion will go to the chipmaker’s top competitors. Advanced Micro Devices Inc., Intel Corp. and Qualcomm Inc. backed Hugging Face’s most recent funding round. They were joined by Salesforce Inc., which reportedly sought to buy the company before Nvidia swopped in.

Hugging Face operates a GitHub-like platform that researchers use to share open-source artificial intelligence models. It also hosts related technical assets such as training datasets. Nvidia Chief Executive Officer Jensen Huang disclosed in a blog post today that the platform hosts more than 500,000 datasets and four times as many models.

Hugging Face has ventured into a few other markets over the years. Last week, it introduced a rollerblade-equipped robotic duck that can pick up items with its beak. Sales topped $5 million in five days. The company earlier teamed up with a group of more than 1,000 researchers to develop its own large language model.

Hugging Face’s main source of revenue is a set of cloud services for AI development teams. One offering, the Hugging Face Hub, makes it possible to create private versions of its AI project hosting platform. The company’s other services enable developers to host their inference models on managed cloud infrastructure.

Hugging Face’s paid products likely aren’t the main reason Nvidia is spending $19.3 billion on the acquisition.

“Nvidia gains visibility into customer’s preferences and the AI models they use,” said Forrester principal analyst Naveen Chhabra. “They can see which models are trending, what datasets customer are downloading, and the architectures that are gaining traction weeks before they hit mainstream tech news.”

Huang wrote in today’s blog post that Hugging Face will continue to support AI projects capable of running across multiple AI chips and clouds. He added that Nvidia won’t require users of the platform to deploy their workloads its silicon.

“Nvidia also gets data-driven foresight into what next-generation chip architectures they need to build based on developer habits,” Chhabra said. “They are securing the software layer because hardware-only companies get commoditized over time.”

Roman Stanek, CEO of GoodData.AI, told SiliconANGLE in an email that the deal is less about Nvidia consolidating AI and more about the market moving toward smaller, more specialized models running alongside frontier models.

“Nvidia’s acquisition of Hugging Face supports our belief that smaller, specialized models, for which Hugging Face has become the leading hub, will ultimately power 80% of AI applications,” he said. “Frontier models will remain important, but most real-world AI workloads will favor models that are faster, cheaper, more specialized, and can run close to the application and its data.”

The acquisition is Nvidia’s second largest startup deal after its $20 billion licensing agreement with Grok Inc. last year. That transaction bought the company access to an inference-optimized chip architecture and a significant amount of related talent. Nvidia used the technology to develop the Groq 3 LPU, a processor optimized to run latency-sensitive LLMs.

Photo: Nvidia