Meta is considering using Google TPU chips for its data centers in 2027, NVIDIA ( NVDA ) stock price fell in pre-market.

In pre-market trading on Tuesday, Nvidia's stock price fell by 3.2%, following reports that Meta is in talks to use Google's AI chips. Boosted by this news, Alphabet's stock price rose by 2.1%, as investors are digesting the potential changes in the AI hardware landscape.

The Information reported on Monday that Meta is considering deploying Google's Tensor Processing Units (TPU) in its data centers before 2027. The social media giant may start renting TPUs from Google Cloud as early as next year.

For Google, securing Meta as a client will validate the effectiveness of its custom chip technology. The TPU was initially launched in 2018 for internal use within Google Cloud's business. These chips have undergone multiple generations of development, with each generation specifically designed for AI workloads. The customization features of the TPU give Google an advantage. Experts point out that chips built for specific tasks can deliver higher efficiency compared to general-purpose computing.

Meta is one of the largest AI infrastructure investors in the world. The company expects its capital expenditure to reach $70 billion to $72 billion this year alone. This strong purchasing power makes Meta's chip choices influential across the entire industry.

Technology companies have been actively seeking alternatives to Nvidia's graphics processors. Although Nvidia still maintains its market leadership, the trend towards diversification is becoming increasingly evident.

Google recently reached an agreement with Anthropic to procure up to 1 million TPUs. Seaport analyst Jay Goldberg described the deal as a “strong validation” of the technology. He noted that many companies had previously been evaluating TPUs, and now more companies may be considering adoption.

The differences in chip architecture are crucial. GPUs were originally designed for rendering graphics in video games. It turns out they are very well suited for training artificial intelligence because they handle massive amounts of data and parallel computations effectively. TPUs, on the other hand, take a different approach; they are application-specific integrated circuits designed from the ground up for discrete applications, specifically designed by Google for artificial intelligence and machine learning tasks. (CoinCentral)