Google consumed 27.9 terawatt-hours of electricity in 2025, a 37% increase from the previous year, as the company’s accelerated buildout of artificial intelligence infrastructure placed unprecedented strain on its energy resources. The figure, disclosed in Google’s annual environmental report, represents enough power to supply approximately 2.6 million American homes for a year and marks the steepest single-year increase in the company’s electricity consumption on record.
The surge stems directly from the computational demands of training large language models and serving AI-powered products, including Gemini and AI-enhanced search features rolled out throughout 2025. Data centres required to support these services operate at significantly higher power densities than traditional computing infrastructure, with individual AI training clusters drawing megawatts of continuous power for weeks or months at a time.
Google’s energy trajectory now poses a direct challenge to its stated climate commitments. The company pledged in 2020 to operate on carbon-free energy by 2030 and achieve net-zero emissions across all operations. However, the 2025 electricity increase pushed Google’s total greenhouse gas emissions up 48% compared to its 2019 baseline, according to the environmental report. The company’s renewable energy purchases, whilst substantial, have not kept pace with the velocity of AI-driven demand growth.
The business implications extend beyond Google’s balance sheet. Hyperscale data centre operators are now competing for limited grid capacity in key markets, with some regions experiencing multi-year queues for new power connections. This constraint is reshaping corporate site selection strategies and driving renewed interest in on-site generation, including small modular nuclear reactors and advanced geothermal systems. Google has signed agreements for 2.7 gigawatts of renewable energy capacity, but bringing these projects online typically requires three to five years from contract signing to commercial operation.
Enterprise customers evaluating AI deployment strategies must now factor energy costs and availability into total cost of ownership calculations. Cloud computing prices have remained relatively stable despite surging electricity costs, suggesting hyperscalers are absorbing margin pressure to maintain competitive positioning. This dynamic cannot persist indefinitely; analysts at Bernstein Research estimate that sustained double-digit energy cost growth could force cloud pricing adjustments of 8-12% by late 2026.
Google’s competitors face identical pressures. Microsoft reported a 30% increase in electricity consumption for its fiscal 2025, whilst Amazon Web Services has not yet disclosed comparable figures. The collective demand from AI infrastructure buildouts is material enough to influence regional power markets; in Northern Virginia, home to the world’s largest concentration of data centres, AI facilities are projected to account for 25% of total electricity demand by 2028, according to grid operator PJM Interconnection.
The energy intensity of AI workloads varies significantly by model architecture and deployment pattern. Inference—serving responses to user queries—consumes less power per operation than training but occurs at vastly higher frequency. Google’s search engine processes approximately 8.5 billion queries daily; adding AI-enhanced features to even a fraction of these queries multiplies the computational load. The company has not disclosed the specific energy cost per AI-enhanced search query, but independent estimates from researchers at the University of California suggest it could be 6-10 times higher than traditional search.
Regulatory scrutiny is intensifying. The European Union’s Energy Efficiency Directive now requires large data centre operators to report detailed energy consumption metrics and demonstrate progress towards efficiency targets. Several U.S. states with significant data centre concentrations, including Virginia and Texas, are considering similar disclosure requirements. These mandates will force greater transparency around the energy costs of AI services and may influence corporate deployment decisions.
The immediate outlook centres on whether efficiency improvements can bend the energy curve. Google has highlighted advances in custom AI accelerator chips and cooling systems that reduce power consumption per computation. However, these gains must outpace the growth in AI workload volume—a challenging proposition given current product roadmaps. The company’s ability to meet its 2030 carbon-free energy target whilst scaling AI services will serve as a critical test case for the broader technology sector’s sustainability claims.
