Alphabet is accelerating development of proprietary ASICs alongside Amazon and Meta to diversify its AI infrastructure and reduce dependence on third-party GPUs. These custom chips are designed for specific tasks in generative AI workloads, signaling a strategic shift in procurement priorities.

Cloud providers are shifting purchasing priorities toward energy-efficient processors measured by joules per token, reflecting a move from raw compute capacity to power-aware throughput. As workloads transition from model training to inference, demand will favor chips optimized for speed, latency and lower power consumption.

This trend could erode demand for general-purpose GPUs and pressure Nvidia’s market share, while bolstering Google’s long-term cloud margins and competitiveness. Investment in in-house ASICs may also accelerate specialized applications in robotics and physical AI systems, further differentiating Google’s offerings.