Google and Meta Ink AI Chip Deal to Challenge Nvidia

Google's Tensor Processing Units, or TPUs, are custom-built chips designed specifically for AI workloads. The first generation debuted internally in 2015 to power products like Google Search, and by 2018, they were available to external customers through Google Cloud. The latest generation, codenamed "Ironwood," is the company's seventh iteration and is designed to handle the massive computational demands of both training and running large-scale AI models. This deal is part of a broader trend of major tech companies developing their own custom silicon to reduce reliance on third-party chipmakers. Meta has its own in-house AI chip program called the Meta Training and Inference Accelerator (MTIA). The first version of MTIA was aimed at improving the efficiency of Meta's recommendation algorithms for platforms like Facebook and Instagram. The partnership comes as demand for AI chips, particularly those from Nvidia, has skyrocketed. Nvidia currently dominates the AI accelerator market, holding an estimated 70% to 95% market share. This has led to intense competition for a limited supply of their powerful, and expensive, H100 GPUs, with a single chip costing between $25,000 and $40,000. For Meta, this agreement provides a crucial alternative source for the massive amount of computing power needed for its AI ambitions, mitigating supply chain risks and potentially lowering costs. For Google, it represents a significant commercial validation of its TPU technology and opens up a new revenue stream, directly challenging Nvidia's market position. The collaboration also extends to software, with Google working to optimize Meta's open-source PyTorch AI framework for its TPUs, making it easier for developers to switch from Nvidia's platform.