Inside Google’s massive AI capex (live)

AI infrastructure growth is currently driven by massive capital expenditure, with companies like Google allocating over $175 billion annually to support expanding compute needs. As the industry transitions from model training to inference, the requirement for gigawatt-scale data centers is evolving toward more flexible, geographically distributed deployments. Reliability standards, traditionally set at four nines, are being re-evaluated as compute costs dominate total service expenses, prompting a shift toward lower-reliability power delivery in exchange for increased capacity. Amin Vahdat, Chief Technologist for AI Infrastructure at Google, emphasizes that vertical integration—co-designing chips, software, and power systems—is essential for efficiency. While labor, power, and chip supply chains remain critical rate limiters, the future of data center design relies on optimizing power-to-space ratios and managing microgrid-like control systems to handle diverse, latency-sensitive workloads.