Every token runs on a real chip drawing real power — and demand is so far ahead of supply that it's driving the biggest infrastructure build-out in a generation and the fastest efficiency gains tech has ever seen. The scale of the gap is the whole story: token demand is growing roughly a hundred times faster than the electricity to serve it. Compare the two growth rates over the same two years:
Even the "constraints" are really signs of how fast this is scaling — delays and power limits are the growing pains of a historic build-out, and they're pushing every player toward radical efficiency.
When every watt has to count, the result is the fastest efficiency progress in tech history plus a once-in-a-generation infrastructure boom. Smart pricing does the rest — flat-rate buckets keep AI affordable and predictable for people, metered rates let machines scale — so everyone keeps getting access while the grid races to catch up. The gap is already closing from both ends: more power coming online, and far more done per token. See the optimistic case →
Sources — Google I/O 2026 / Google AI Blog (token volumes: 9.7T → 480T → 3.2 quadrillion tokens/month). IEA (data-center electricity ~415 TWh in 2024, projected ~945 TWh by 2030) & Brookings, Apr 2026 (approaching ~1,000–1,050 TWh, the world's "5th-largest consumer" — a faster ramp than the IEA base case). Dell'Oro Group, Mar 2026 (2026 data-center capex > $1T); Big-4 hyperscaler 2026 capex ~$600B, ~75% AI (company guidance, Amazon/Microsoft/Alphabet/Meta). Bloomberg / Sightline Climate, Apr–May 2026 (~50% of US 2026 projects delayed or canceled). Gartner (40% of AI data centers power-constrained by 2027). Inference 80–90% of compute load: industry power-requirement analyses, 2026.
Method — "×330" and "×2.4" are growth multiples over the same ~2-year window from the cited figures; they compare rate of growth, not absolute volume. Bar lengths are proportional to those multiples.