As long as open-source AI models dominate, Nvidia's custom chip competitors won't be able to erode its share of the semiconductor market, an analyst says

CEO Jensen Huang announced the Nvidia acquisition of Hugging Face on Thursday.

Nvidia has a history of knowing where to invest to maintain its chip dominance, and one analyst sees its acquisition of Hugging Face as the latest example.

Consider the chip maker's (NVDA) $13 billion purchase of the open-source artificial-intelligence repository as "health insurance" for a company facing increasing competition from custom chips, said Richard Windsor, founder of independent research firm Radio Free Mobile.

In his view, the deal "is about ensuring that no one becomes dominant in AI models and the services that are built on them." For Nvidia, that "will go a long way to maintaining its competitive edge in silicon," Windsor wrote in a Friday note to clients.

Nvidia's vice president of enterprise AI, Justin Boitano, said on a call following the announcement on Thursday that the company is "incentivized" and "motivated" to keep Hugging Face open and neutral so that its more than 18 million AI developers can run models and access data sets on any platform.

Windsor said it's in the chip maker's "best self-interest" to keep Hugging Face independent, as that's how it will be able to avoid losing its share of the AI chip market to AI service providers that are making their own chips.

Nvidia has been able to maintain its lead in the AI chip market through its Compute Unified Device Architecture, or CUDA, platform, Windsor said. That platform allows developers to do general computing tasks with Nvidia's graphics processing units. The chip maker's annual product cadence has also kept it at least one chip generation ahead of competitors, he added.

As the AI industry shifts to inference, or the process of AI models making predictions based on what they've learned, however, Windsor argued that Nvidia is losing its power with CUDA, which has been a sort of moat for the company in the model-training era, and is only influential on the chip layer of the full AI stack that spans hardware to services.

Windsor said that if Hugging Face can remain the preferred platform for open-source AI models, "no one will be able to corner the market for AI services," where Nvidia has less influence.

He pointed to the risk posed by companies like Alphabet (GOOGL) (GOOG) unit Google and Anthropic, which are developing custom chips to power their AI offerings. If one of the companies were to become "the dominant enabler of AI services," Windsor said, the entire AI ecosystem could then "easily become optimized" to run on their in-house chips.

That would not only reduce Nvidia's pricing power, he said, but cause a "vast amount" of market-share loss.

Therefore, if open-source AI models can remain "a viable and thriving alternative" to closed-source models from frontier labs and major cloud providers, Nvidia will likely be able to fend off competition from other layers of the AI stack where it is not as strong, Windsor said.

Now, he added, with its massive investment from Nvidia, Hugging Face can focus on investing and growing its role as a major open-source provider.

Don't miss: Nvidia just made a $3.5 billion move that could expand its AI empire

William Blair analyst Sebastien Naji shared a similar view about Nvidia's strategy with the Hugging Face acquisition. "By extending its reach into the model layer, Nvidia widens the CUDA funnel," Naji said in a note to clients.

He added that Nvidia also "gains better visibility into where open development is heading," which will help the company in developing its product roadmap to remain the standard for AI infrastructure.

"Visibility into model development is a hedge against rising ASIC share, with custom silicon increasingly running closed frontier models at the AI labs," Naji said, referring to application-specific integrated circuits like OpenAI's Jalapeño, which was co-developed with Broadcom (AVGO).

To him, Nvidia's support for the open-source ecosystem ensures demand for AI hardware is "broad-based and drives more value towards the merchant GPU ecosystem."

-Britney Nguyen