The 33× AI Efficiency Shock: How Google just slashed AI energy costs in 12 months
Google revealed the energy for an average AI text query dropped 33× in one year—cutting from 0.3 Wh to 0.009 Wh per request. Translation: the same GPU cluster that once handled 10 M queries can now serve 330 M without buying a single new card. If you run SaaS, internal copilots, or any inference-heavy feature, expect your cloud bill to shrink or your margin to spike. The breakthrough came from on-die telemetry, liquid cooling loops tuned per rack, and a new inference stack that fuses model, compiler, and hardware schedules. Expect AWS, Azure, and GCP to race the same playbook into GA tiers soon. Source: https://arstechnica.com/ai/2025/08/google-says-it-dropped-the-energy-cost-of-ai-queries-by-33x-in-one-year/
arstechnica.com · 3 min read