Google's AI Token Processing Grows 52x, Serving Costs Plummet

Google's Q4 2025 earnings call revealed a substantial acceleration in its AI capabilities, with Gemini now processing over 10 billion tokens per minute, a 52x increase year-over-year. This growth is coupled with a 78% reduction in Gemini serving unit costs, indicating significant improvements in efficiency. The company's 2026 CapEx investments are projected to be in the range of $175 to $180 billion, signaling a strong commitment to expanding its AI infrastructure. This level of investment, if mirrored by other hyperscalers, could drive data center CapEx to between $500B and $750B this year. The growth in token processing and cost reduction has direct revenue implications, with Google Cloud revenue growing 48% to $17.7 billion. While Microsoft reports a higher number of customers processing over 1 trillion tokens annually, Google's growth rate and cost efficiencies suggest a competitive advantage. The comparison to historical infrastructure spending, such as the railroad era and the national highway system, puts the current AI data center buildout into perspective. The long-term implications of this rapid growth and investment include the potential for more affordable and accessible AI services, but also the risk of increased market concentration and potential misuse of AI technology.