Meta is making a move in the artificial intelligence race and is considering a significant change in the technology powering its data centers. According to recent reports, Mark Zuckerberg's company is negotiating with Google for a massive purchase of TPU chips , the AI ​​silicon developed by Alphabet's parent company, which could alter the current balance of the AI ​​hardware market.
These talks, still in a preliminary phase, suggest that Meta would allocate billions of dollars to incorporate these processors into its infrastructure starting in 2027 and "rent" TPU-based computing capacity through Google Cloud from 2026. The move would put more pressure on Nvidia, until now the almost hegemonic provider of GPUs for large AI workloads worldwide.
Meta is negotiating Google TPU chips for 2027
According to sources cited by The Information, Meta is in talks to acquire Tensor Processing Units (TPUs) from Google and deploy them in its data centers starting in 2027. These chips would be used to train and run advanced artificial intelligence models, a key component in products like Facebook, Instagram, and WhatsApp, and in the development of future smart assistants and services.
The potential agreement would not be limited to the direct purchase of hardware. Meta is also considering acquiring access to Google chips through Google Cloud starting next year, which would allow it to test and scale these technologies more quickly before committing to even larger investments in its own equipment for its facilities.
For Meta, which plans tens of billions of dollars in AI-related capital expenditures, the move would be a way to diversify its reliance on Nvidia and, at the same time, secure computing power in a context of severe shortages of high-performance chips.
From Google's perspective, closing a deal of this magnitude with one of the world's largest buyers of AI hardware would give a boost to its custom chip strategy and strengthen the appeal of its cloud against giants like Amazon Web Services or Microsoft Azure, especially in Europe, where many companies are accelerating their generative AI projects.
What are Google TPUs and why do they matter?
TPUs, short for Tensor Processing Unit , are processors designed by Google specifically to accelerate machine learning and neural network tasks. Unlike traditional GPUs, initially developed for graphics, TPUs fall into the category of custom silicon for AI , designed from the outset for workloads such as training large language models or large-scale inference.
Google launched its first generation of TPUs in 2018 , initially focused on internal use within its own data centers and the Google Cloud platform. Since then, the company has iterated the design with increasingly powerful and efficient versions, culminating in its most advanced chip, known as Ironwood, recently unveiled to attract new AI customers.
This type of custom design allows Google to optimize price, performance, and energy consumption compared to generic solutions. At a time when the cost of training next-generation AI models is skyrocketing, any gain in energy efficiency or cost per unit of computing becomes a critical factor for large European technology companies, banks, insurers, and industrial groups that are deploying AI in their services.
Google's flagship Gemini 3 has been primarily trained using TPUs , which the company presents as proof that the technology works in real-world, large-scale environments. For potential customers like Meta, who already handle massive data workloads, this operational experience is a key factor to consider when choosing a hardware platform.
Potential impact on Nvidia and the AI ​​hardware market
Until now, most large language models and enterprise AI projects have relied on Nvidia GPUs as the de facto standard . The company's dominance rests on both the performance of its chips and its software ecosystem, especially the CUDA platform, used by more than four million developers worldwide.
However, Meta's potential move to migrate part of its infrastructure to Google's TPU could usher in a new era of competition in the market. Analysts at the brokerage firm XTB point out that data centers account for approximately 90% of Nvidia's revenue, so even a partial loss of key customers in this segment could have a significant impact on its financial results and future growth rates.
In recent years, demand for custom chips, such as TPUs, has skyrocketed as many companies seek alternatives to Nvidia's expensive processors . Projects like Anthropic's—which has expanded its agreement with Google to use up to one million AI chips valued at tens of billions of dollars—demonstrate that the market is beginning to open up to other suppliers.
If Google succeeds in getting Meta and other major tech companies to adopt its TPUs, Nvidia would face increasing pressure on price, energy efficiency, and product offerings . This could force the company to adjust its business strategy, accelerate the development of new generations of GPUs, and strengthen its software tools to maintain its dominant position against alternatives such as Google's custom chips or the internal projects of other giants. Increased competition is also driving agreements between manufacturers; for example, some industry players have formed alliances to secure the supply and design of custom chips.
At the same time, the emergence of more specialized hardware options could benefit European companies that are investing heavily in AI, since greater competition usually translates into better access to computing power , something key for startups and large industrial groups driving automation, data analysis and smart assistant projects.
Google Cloud's strategy and the fight for the AI ​​cloud
The potential deal with Meta fits into Google's broader strategy to make its TPUs a cornerstone of its cloud business . Until recently, these chips were primarily used in the company's own data centers; now, the tech giant is taking steps to enable external customers to deploy them in their own facilities or consume them on demand as a cloud service.
Within Google Cloud, leasing AI chips—including Nvidia GPUs—has become an increasingly important revenue stream . The goal now is for customers to migrate some of those workloads to TPU, where Google has more control over design, supply, and margins.
Some Google Cloud executives estimate that this strategy could allow the company to capture up to 10% of Nvidia's annual revenue in the data center chip segment, a figure that translates into billions of dollars. To achieve this, the company is intensifying its sales and technical campaign, especially among large corporations and AI service providers.
In Europe and Spain, where data regulation and concerns about digital sovereignty are increasingly important, Google Cloud is competing to host the AI ​​projects of banks, telecoms, insurance companies, and public administrations . Having a broader catalog of its own hardware can be an additional advantage when negotiating large contracts, especially when cost and energy efficiency are key factors.
The shift towards custom chips is not without its challenges: to compete with Nvidia, Google must offer mature development tools, optimized libraries, and a partner ecosystem that facilitates code and model migration. At this point, the historical weight of CUDA remains a significant barrier that the company will have to overcome if it wants to gain sustained market share.
Market reaction and analysts' interpretation
The leak that Meta is considering buying Google's TPU chips had an immediate effect on the stock market. In pre-market trading on Wall Street, Alphabet's shares rose by around 4% , bringing the company closer to a record market capitalization of nearly $4 trillion, if the gains were to continue throughout the session.
Meanwhile, shares fell by close to 3-4% in pre-market trading, reflecting investor concern that one of their largest AI hardware customers might explore alternatives in the medium term. Broadcom, which collaborates with Google on the manufacture of its artificial intelligence chips, also saw gains of around 2%.
Analysts at firms like XTB believe that a firm agreement between Meta and Google would represent a significant shift in the AI ​​industry, weakening Nvidia's near-monopoly in data center infrastructure. They point out that such moves could force a profound reorganization of the market, both technologically and commercially.
Furthermore, they emphasize that the adoption of TPU by top-tier customers could serve as public validation of Google's commitment to custom silicon. If major social media platforms, generative AI companies, and European corporations begin to rely on these types of chips, it is likely that other players will follow suit to avoid falling behind.
Experts also draw attention to the energy dimension of the problem: AI data centers consume increasing amounts of electricity and water, a particularly sensitive issue in the European Union. In this context, any improvement in the energy efficiency of chips could become a decisive factor when choosing a technology provider and locating new infrastructure.
The potential alliance between Meta and Google regarding TPUs points to a new phase in the race for artificial intelligence hardware, in which Nvidia's dominance is beginning to be questioned and custom chips are gaining ground as a serious alternative to major technology platforms, also in the European market.
