Jensen Huang didn’t earn the nickname “Godfather of AI” for nothing.
With demand for chips made by Nvidia Corp. greatly outstripping supply, the company’s chief executive officer has been able to name his price for them. Through a web of investments, AI companies big and small are firmly in his pocket. Even to the world’s wealthiest men, Huang’s the don. “I would describe the dinner as me and Elon begging Jensen for GPUs,” Oracle Corp. founder Larry Ellison said of one meeting in 2024.
That’s why news of Google’s discussions to sell “billions of dollars” of its own AI chips to Meta Platforms Inc. had the feel of a plot; not to ambush Huang at the grocer but to demonstrate his grip on the AI hardware sector could be under threat.
A deal with Meta could lead Mark Zuckerberg’s company to install Google’s chips in its own data centers by 2027, The Information reported on Monday, and rent them from Google Cloud as soon as next year. Google’s coup would follow a similar move to provide up to 1 million of its chips — known as Tensor Processing Units — to leading AI startup Anthropic, creator of the Claude chatbot.
The following day’s stock reaction — a pullback of 2.6% for Nvidia and a 1.6% gain for Google parent Alphabet Inc., which is already up almost 70% this year — was recognition of AI builders’ eagerness to find an alternative to Nvidia’s chips and the sky-high prices that allowed it to post gross margins of 76% in the last quarter.
The central shift in sentiment is this: Last week, Nvidia investors worried most that demand for AI hardware might fall. Now they also worry some of that could be poached.
Meta may be only one of Nvidia’s customers, but any shift in client behavior is a significant warning sign for the company’s growth. Just four hyperscalers (Google and Meta likely among them, though this isn’t confirmed) make up 61% of Nvidia’s total revenue, according to its most recent quarterly earnings report.
Several of Nvidia’s biggest clients are trying to reduce their reliance on its chips by designing their own. After Google, Amazon.com Inc. is the next furthest along: It’s Trainium and Inferentia chips are vying for business on the same terms as a cheaper and more readily available alternative.
Where Google has the upper hand on Amazon, however, is that it is walking the walk: Its Gemini 3 model, which was trained and runs on Google’s own chips, topped most industry benchmarks when it was released this month. “The reason why everyone believes Google’s chips are comparable to Nvidia,” Bloomberg Intelligence’s Mandeep Singh explained to me, “is because they have an LLM that’s comparable to OpenAI and Anthropic in performance.”
Amazon lacks a flagship AI model of its own, though Anthropic is already building some of its models on Amazon-designed chips. Amazon’s re:Invent conference, the showcase for its AWS cloud unit, is being held in Las Vegas next week. Any announcements with more big names could place even more pressure on Nvidia stock as we head toward the end of a dizzying year.
In 2026, Huang will be forced to do something he hasn’t had to worry so much about for a while: jockey for business. He’s been preparing. Much has been written about the “circular” investments Nvidia has made to lock down future demand, such as a $100 billion investment in OpenAI. After Google announced its Anthropic deal, Nvidia followed up with its own to have Anthropic develop on Nvidia chips. The cut and thrust underlines that no one expects this to be a winner-takes-all scenario. Every piece of hardware sold will count.
As its stock price took a tumble on Tueday, Nvidia was moved to post on X that it was “delighted by Google’s success” but called its own chips “a generation ahead of the industry.” It argued that its hardware was more versatile than TPUs and other similar chips designed only for specific tasks.
But at risk of declaring it prematurely, the power dynamic has shifted in the AI trade. The AI builders have clamored for Nvidia chips. They still do and will for some time. But the Google-Meta talks speak to the battles to come for chipmakers, which must fight to make sure the top models are made on their infrastructure. For Nvidia, that’s no longer a given.
A message from Advisor Perspectives and VettaFi: Interested in learning more about how bond ETFs can help diversify your portfolio? Click here to read more.
Bloomberg News provided this article. For more articles like this please visit
bloomberg.com.
Read more articles by Dave Lee
