Google and Meta Team Up on AI Chips
Google has signed a multibillion-dollar deal to supply Meta with its custom AI accelerator chips, forming a powerful alliance to challenge Nvidia's market dominance. The partnership aims to fast-track Meta's AI service deployment and highlights a major trend of tech giants forging hardware alliances to control their own AI supply chains.
This partnership is a direct response to Nvidia's staggering grip on the AI chip market, where it holds an estimated 85% to 92% market share. The high demand for Nvidia's GPUs, like the H100, has led to tight supply and sky-high prices, pushing major AI players to seek alternatives. The deal gives Meta access to Google's custom-built Tensor Processing Units (TPUs), specifically designed to accelerate AI workloads. Unlike general-purpose GPUs, TPUs are optimized for the matrix math that is central to AI models, which can make them faster and more power-efficient for specific tasks. Google has been developing TPUs for over a decade, giving them a significant head start in custom silicon. This move follows significant setbacks in Meta's own AI chip development. The company recently scrapped its most advanced in-house training chip, code-named "Olympus," due to design and manufacturing challenges. This has forced Meta to rely more on external partners, leading to recent multi-billion dollar deals not only with Google but also with Nvidia and AMD. Under the multi-year, multi-billion dollar agreement, Meta will initially lease the TPUs from Google to power the training and inference of its next-generation AI models. A potential second phase of the deal, starting as early as 2027, could see Meta purchasing the custom chips directly to install in its own data centers. Google's latest TPU, the 7th generation "Ironwood," boasts peak performance that rivals Nvidia's Blackwell architecture. The 6th generation "Trillium" chip already offers 4.7 times the performance of its predecessor and is used to train Google's own Gemini AI models. This alliance is part of a larger industry trend of tech giants like Amazon, Microsoft, and OpenAI developing their own custom AI chips. The goal is to reduce dependency on Nvidia, cut costs, and gain more control over their hardware and software integration for optimized performance.
