Google's AI Buildout Drove 37% Increase in Electricity Use in 2025

Google's electricity consumption hit a record high in 2025 — a 37% increase driven primarily by AI infrastructure expansion. According to Ars Technica's July 2, 2026 report, Google's data centers powering AI model training and inference are the biggest energy consumers.

Why Does AI Require So Much Energy?

Training AI models requires thousands of GPUs running simultaneously for weeks. A single training run of a GPT-5-level model consumes 50-100 MWh. Google operates dozens of AI models — Gemini, NotebookLM, Search AI, Cloud AI. In 2025, Google deployed 1.5M+ GPUs (TPU v5, NVIDIA Hopper).

Carbon Emissions Rising

Google pledged net-zero emissions by 2030, but AI growth has put this target at risk. Google's carbon emissions have increased 48% since 2019. AI energy efficiency is improving (NVIDIA H100 is 4x more efficient than A100), but demand is growing faster.