Google and Meta Team Up on AI Chips

The deal will see Meta lease Google's custom-designed Tensor Processing Units (TPUs) to power the training and inference of its next-generation artificial intelligence models. This multi-year, multi-billion dollar agreement provides Meta with crucial access to high-performance computing amidst a market facing tight supply and soaring prices for AI hardware. There is also potential for the partnership to expand, with discussions underway for Meta to purchase TPUs directly for its own data centers as early as 2027. This move is part of Meta's broader strategy to diversify its AI hardware suppliers beyond Nvidia, which currently dominates the market with an estimated 85-92% share. In addition to the Google deal, Meta has also recently secured major agreements to acquire millions of GPUs from Nvidia and billions of dollars' worth of AI chips from AMD. This multi-sourcing approach aims to mitigate risks associated with relying on a single provider and gain leverage in pricing negotiations. Meta's increased reliance on external chipmakers follows significant challenges with its own in-house silicon development. The company has reportedly halted the development of its most advanced custom training chip, codenamed "Olympus," due to design complexity and manufacturing risks. While Meta continues to develop its "Meta Training and Inference Accelerator" (MTIA) program, these setbacks have accelerated its turn towards established chip suppliers. For Google, this partnership marks a major step in commercializing its TPU technology, which it has been developing for over a decade. Originally created for internal use to power services like Search and Maps, Google is now making its custom chips available to external companies, positioning itself as a direct competitor to Nvidia in the AI infrastructure market. The deal with Meta, a major AI player, serves as a significant validation of Google's TPU architecture.