Google and Meta Team Up on AI Chips
In a major challenge to Nvidia's market dominance, Google has signed a multibillion-dollar deal to supply Meta with its custom AI chips. Meta will shift some of its AI workloads to Google's next-gen silicon, a huge validation for Google's hardware ambitions and a signal that Big Tech is actively seeking alternatives to Nvidia's GPUs.
This partnership is part of Meta's broader strategy to diversify its AI chip suppliers beyond Nvidia, which holds a dominant market share of around 92% for discrete GPUs. Meta has also inked multi-billion dollar deals with AMD for its Instinct GPUs, aiming to mitigate supply chain risks and gain leverage in pricing negotiations. Meta's 2026 capital expenditure for AI infrastructure is projected to be between $115 billion and $135 billion, a significant increase from the $72 billion spent in 2025. This massive investment is geared towards building the necessary data centers, servers, and networking capabilities to support the exponential growth in AI workloads across its platforms like Facebook, Instagram, and WhatsApp. While expanding its partnerships, Meta is also heavily invested in developing its own custom silicon. Its first-generation AI inference accelerator, MTIA v1, is designed to optimize performance for its specific deep learning recommendation models. The company's infrastructure chief, Santosh Janardhan, has emphasized that a combination of chips from Nvidia, AMD, and their own custom silicon will be necessary to handle their diverse workloads. Google's Tensor Processing Units (TPUs) are custom-designed chips specialized for the matrix and vector-based mathematics essential for AI models. First deployed internally in 2015, TPUs have evolved through multiple generations, with the latest versions offering significant power and efficiency for training and running large-scale AI models. This deal provides Google with a major customer for its TPUs, strengthening its cloud business and challenging Nvidia's market leadership.
