Google Inks AI Chip Deal with Meta

The move to secure Google's hardware follows significant setbacks in Meta's own custom silicon projects. Meta recently halted development of its most advanced in-house training chips, code-named "Olympus" and "Iris," amid concerns that the designs were too complex for mass production and not as stable as Nvidia's established products. This Google agreement is part of a much larger AI infrastructure spending spree for Meta, with capital expenditures for 2026 projected to hit between $115 billion and $135 billion. The company also recently signed a multi-year deal for millions of Nvidia GPUs and a separate five-year partnership with AMD valued at up to $100 billion. For Google, this represents a major strategic pivot from primarily using its Tensor Processing Units (TPUs) for internal projects to offering them commercially. It's a direct bid to capture a piece of the lucrative AI hardware market, with Google Cloud executives reportedly setting a goal to win up to ten percent of Nvidia's annual revenue. Nvidia currently dominates the market for AI training chips, controlling an estimated 80% to 92% of the market share. This dominance is built not just on its powerful GPUs but also its proprietary CUDA software, which has become an industry standard for AI development. The deal gives Meta access to chips with a fundamentally different design. Unlike Nvidia's general-purpose GPUs, Google's TPUs are Application-Specific Integrated Circuits (ASICs) designed purely for the mathematical operations at the core of AI models, which can make them more power-efficient for large-scale training. The current multi-year agreement involves Meta renting the TPUs, but the two companies are also discussing a potential second phase. Starting as early as 2027, Meta may begin purchasing the custom Google chips outright to install in its own data centers.