Nvidia Corp is not only tightening its grip on artificial intelligence (AI), it is also absorbing the companies fighting to keep the industry open and decentralized. On the same day the chip giant announced quarterly revenue that had doubled and would continue rising next year, sending its shares up 7 percent, news broke that it has reportedly agreed to spend US$13 billion on Hugging Face, a platform for open-weight AI models.
That name might sound familiar. It is the company that OpenAI’s rogue agents controversially breached during a recent security-testing exercise. Hugging Face handled that incident with diplomacy, as you would expect of a company that sits at the center of a broader open-source AI community. Founded by three French entrepreneurs, it hosts one of the world’s biggest collections of open-weight models, or AI systems that could be changed to suit a user’s needs, and it has long promoted the idea that AI development should be open to everyone and not centrally controlled.
So strongly did it believe in that credo that the start-up rejected a US$500 million investment from Nvidia last year that would have valued it at US$7 billion, on the grounds that it did not want a single dominant investor to sway decisions.
“I said it and will say it again: Concentration of power is the biggest risk in AI!” Hugging Face co-CEO Clem Delangue said in 2024.
Today’s apparent change of heart might be down to the higher price — and the fact that Hugging Face was reportedly being circled by other potential acquirers such as Salesforce Inc.
The deal would create a strange paradox. Nvidia wants open models to thrive, but for a more self-interested reason: to stop OpenAI and other large AI companies from threatening its chip business. OpenAI this week announced remarkably strong results for a custom AI chip called Jalapeno, with early tests suggesting it could outperform some of Nvidia’s most advanced versions.
Which makes the relationship between the two AI heavyweights ever more awkward. Nvidia is one of OpenAI’s biggest investors, an arrangement that has led to frequent accusations of “circular financing” on the chip giant’s part, but it also wants to keep the ChatGPT maker in its place. Owning Hugging Face could help it create a counterweight to closed-source AI developers such as OpenAI and Alphabet Inc’s Google, which are building chips to compete with its own. Developers using open models can choose which chips to run them on, and Nvidia wants a chance to sell them its hardware — rather than seeing them steered toward processors made by its competitors.
That helps explain the reported US$13 billion price tag. Hugging Face recently generated about US$150 million in annual revenue, which would mean Nvidia is paying a remarkably high sales multiple (about 86 times annualized revenue) for the start-up. However, Nvidia does not need much in the way of money. It has forecast chip sales to grow 70 percent in the next fiscal year and continues to enjoy gross margins of about 75 percent.
Instead, buying Hugging Face would give Nvidia a new strategic position in AI and greater influence over developers, thanks to the platform’s links to millions of coders at the point where they download and deploy models. That could rev up Nvidia’s cloud ambitions, giving it another way to rent computing power to AI makers.
There is a bigger irony here. Some supporters of Hugging Face, which now has hubs in Paris and New York, have been celebrating the reported sale as a win for European tech. However, it would also represent a sellout, not just for start-ups in Europe, which frequently capitulate to larger Silicon Valley buyers before they get a chance to build their own empires, but also for the open-source AI movement itself.
For all of Hugging Face’s efforts to make AI less centralized, it might help Nvidia protect its dominance. The potential acquirer controls an estimated 70 percent to 90 percent of AI chip sales, a market that Google has struggled to dent with its own processors. And Hugging Face’s famous neutrality would come into question if it provides the world’s dominant chipmaker with a new route to persuading developers to choose its hardware for their models.
However many makers of chatbot technology there may be, the infrastructure underpinning those services is still primarily controlled by one company. And one of the world’s biggest advocates for decentralizing AI power is about to hand Nvidia even more of it.
Parmy Olson is a Bloomberg Opinion columnist covering technology. A former reporter for the Wall Street Journal and Forbes, she is author of Supremacy: AI, ChatGPT and the Race That Will Change the World. This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
The “quiet revolution” of Taiwan’s democratization took shape during the 12 years that followed Lee Teng-hui (李登輝) simultaneously becoming chairman of the Chinese Nationalist Party (KMT) and president of the Republic of China (ROC). Chen Shui-bian (陳水扁) of the Democratic Progressive Party (DPP) then took the presidential helm in 2000, bringing an end to one-party rule in the ROC, and was re-elected in 2004. He was followed by the KMT’s Ma Ying-jeou (馬英九), who served two four-year terms; the DPP’s Tsai Ing-wen (蔡英文), who also served two terms; and the DPP’s William Lai (賴清德), currently in his first term. Having
Chinese hackers are ramping up attacks after integrating DeepSeek and other open-source artificial intelligence (AI) models into their operations, demonstrating attackers’ ability to leverage basic AI tools to hit targets abroad. State-affiliated cyberwarfare groups more than doubled the amount of attacks they carried out since they began delegating mundane tasks to AI and using it to develop advanced malicious software, according to TeamT5, a Taiwanese research firm. Researchers said it was not always possible to identify the AI model they used, but in general DeepSeek’s offerings are popular among hackers in China because of its high performance and ability to be
In the rush to summer vacation, some readers might have missed the launch of the renamed Diet members group supporting Taiwan. Indeed, the decision by Japan’s largest pro-Taiwan parliamentary organization to rename itself received only modest media attention. Yet historians might eventually view it as one of the most revealing symbolic acts in the evolution of post-World War II Taiwan-Japan relations. For the longest time, the group, founded in 1973, was called the Nikka Giin Kondankai (Japan-ROC Diet Members’ Consultative Council). However, after more than five decades, the organization on June 11 adopted a new name — the Japan-Taiwan Diet Members’
A disturbing incident in Taichung this month has raised difficult questions about safety, inclusion and responsibility in schools. A junior high-school teacher was seriously injured after a student reportedly broke a broom and attacked her with the handle. The severity of her injuries generated concern among teachers and the wider public. However, incidents such as this require careful interpretation. They should not be used to suggest that students with disabilities, autism or other forms of neurodivergence are inherently dangerous. They are not. Nor should a single act of violence become an argument for retreating from inclusive education. The more important question is
RSS