Demand Lab / Live Model — Inputs are illustrative
Time horizon: 2024 → 2050
Lab / 002 Working model ● live

Watch the world
get thirstier.

A toy model of the planet's daily appetite for cognition — and what it costs in the older utilities to feed it. Drag the year. Watch the curves bend.

Year / Index
2036
20242030203620422050
Tokens consumed per person · per day
—
Doubling Period
14 mo
How quickly per-person token consumption doubles. Base case sits between today's training-compute curve and current inference scaling.
Readout / Year-On-Year
Tokens / person / day —
Global daily tokens —
Vs. today —↑
Substrate utilization —

What that draws from the older utilities.

Pulled from year 2036
Power
—TWh / yr
—
Annual electricity drawn just to feed inference and training pipelines.
Water
—km³ / yr
—
Cooling water cycled through campuses and evaporative loops.
Fiber
—Pbps
—
Aggregate east-west and edge bandwidth to keep inference local enough.
Land
—km² built
—
Hyperscale floor space for compute, transformers, and cooling plant.
Per Capita Year 2036

Your personal supply.

The substrate is metered the way every old utility was metered — a slow trickle no one noticed, until everyone needed it.

Tokens / day
—
≈ — novels of thinking
Watts / continuous
—W
≈ — a fridge always on
Liters water / day
—L
≈ — a shower per person, every day
Cost / month (USD)
$—
trending ↓ per token

If even the conservative curves are right,
the world will need a substrate.

Open a channel →