Meta Cuts Dependency on Nvidia: Google’s AI Chips Disrupt Monopoly in the AI Race

Nvidia experienced a drop in stock price during pre-market trading on NASDAQ earlier this week after it was reported that Meta, the owner of Facebook and Instagram, is negotiating to base part of its future AI infrastructure on Google’s specialized chips, known as tensor processing units (TPU). The news was first reported by The Information and later confirmed by several business media outlets.

Nvidia’s stock briefly fell by about four percent, to approximately $175, a significant drop from this year’s peak in October when the stock price was around $212. Trading was exceptionally liquid, with over 250 million shares changing hands in extended trading, indicating investor nervousness about Nvidia’s future dominance in AI hardware.

At the same time, Alphabet, Google’s parent company, continues to grow towards a valuation of around $4 trillion, fueled by optimism that its AI chips could become a key alternative to Nvidia in large data centers.

From Gaming GPUs to Generative AI

For years, Nvidia has been almost synonymous with AI infrastructure. Its graphics processing units (GPUs), originally developed for computer graphics and gaming, have become the standard tool for training and running the most advanced artificial intelligence models. The H100 and H200 series GPUs today form the backbone of global AI infrastructure, from OpenAI’s and Anthropic’s models to the internal systems of the largest banks and tech giants.

Analysts estimate that Nvidia holds between 80 and 90 percent of the market for so-called AI accelerators, with some estimates going as high as 95 percent, which is practically a monopoly in a segment that will be worth trillions of dollars in the coming years. Meta itself announced last year that it plans to purchase over 350,000 H100 chips, which is a massive order but also a sign of dependency on a single supplier.

Google’s approach seems fundamentally different. Instead of universal GPUs, the company has been developing its own specialized chips, TPUs, for over a decade, which have been designed almost exclusively for machine learning tasks. These are ASICs (application-specific integrated circuits), chips designed for narrowly defined jobs. They are faster and more energy-efficient for certain AI tasks but perform poorly in ‘general’ computing and cannot replace a CPU or GPU in universal applications. In other words, Nvidia and Google are not playing the same game but are competing for the same money from the largest customers.

What Meta Changes

According to The Information, Meta is considering integrating Google’s TPUs into its data centers starting in 2027, while it could begin renting their capacities through Google Cloud as early as next year. For companies like Meta, Amazon, Microsoft, or Alphabet, the key word is ‘scale.’ They do not need thousands, but hundreds of thousands or millions of chips, with guaranteed delivery and a decrease in price per unit. Nvidia’s GPUs are extremely powerful but also expensive, and global demand has long outstripped supply. If Nvidia cannot deliver enough GPUs, large customers want and seek another source, and Google’s TPUs provide just that, a second supply channel, reducing the risk of supply chain disruptions and, very importantly, a negotiating lever in price discussions. Even a relatively modest redirection of orders in favor of Google, for example, ten percent of the consumption of a giant like Meta, is enough to change market sentiment and drop Nvidia’s valuation by hundreds of billions of dollars, which is happening these days.

For Google, this shift represents a confirmation of its long-term strategy. TPUs have been an internal technology for years, available only within Google. Today, they are a commercial product rented through Google Cloud, and partners like Anthropic have already signed contracts giving them access to up to a million TPUs, worth tens of billions of dollars, with over one gigawatt of computing capacity going online in 2026.

End of Monopoly or Just a Cold Shower?

The news of Meta’s possible pivot has not gone unnoticed at Nvidia. The company, which is unusual for it, publicly reacted on the social network X, stating that it is ‘excited about Google’s success’ and that it continues to supply chips to Google. In the same breath, it emphasized that ‘Nvidia is a generation ahead of the industry’ and that it is the only platform that powers every AI model, wherever computing takes place.

Nvidia emphasizes the flexibility of its GPUs compared to specialized ASICs like TPUs, stating that GPUs can be used for a significantly broader range of tasks, from rendering graphics to scientific simulations to AI, while TPUs are optimized practically exclusively for machine learning.

However, the reality is that competition is accelerating. In addition to Google, AMD is also pushing hard with its MI series of accelerators, as well as custom chips from hyperscalers like Amazon (Trainium, Inferentia) and Microsoft. Meta is also working on its own AI chips to reduce reliance on Nvidia.

What Next

In the short term, this pivot is most visible in stock prices. Nvidia has lost more than ten percent of its value since its peak in October, following an almost euphoric rise and a valuation above four trillion dollars. Alphabet, on the other hand, continues its strong ascent fueled not only by TPUs but also by the success of its own Gemini 3 model.

In the long term, the signal is clearer than short-term price jumps. The era of a single dominant supplier of AI hardware is likely over before it even began. The largest customers want a diverse supply chain, customized chips for specific AI tasks, and more negotiating power. Nvidia will likely remain the standard for the broadest range of applications for years to come, but the race is intensifying.

For everyone else, from smaller cloud providers to companies just entering generative AI, this could mean greater supply, lower prices, and more choices in the long run.