Google's Q4 2025 earnings call highlighted a major surge in AI performance. The company’s Gemini models now handle over 10 billion tokens per minute, up from 7 billion the previous quarter. This translates to a staggering annualized run rate of over 430 trillion tokens, marking a 52-fold increase compared to the same time last year. Around 350 customers are processing more than 100 billion tokens each, while Microsoft reports over 250 customers expected to exceed 1 trillion tokens annually, indicating their larger clients are consuming significantly more. Cost efficiency is a key theme in Google's report. They announced an impressive 78% reduction in serving unit costs for their Gemini models, translating to a 4.5x increase in tokens processed per GPU hour. In contrast, Microsoft reported only a 90% increase in tokens per GPU last year. This suggests that Google is achieving significant efficiencies within its TPU infrastructure. As a result of this AI boom, Google’s backlog surged by 55% to $240 billion, while Google Cloud revenue reached $17.7 billion, reflecting a 48% growth rate that outpaces Azure's 39%. To support this rapid growth, Google plans to invest between $175 billion and $180 billion in capital expenditures in 2026. If Google is spending around $175 billion, combined investments from major players like Microsoft, Amazon, and Meta could push total data center CapEx to between $500 billion and $750 billion this year. Currently, AI infrastructure spending is about 1.6% of GDP, still far below the 6% peak seen during the railroad era, but indicating that this sector is on track for significant growth akin to major national infrastructure projects.