Back to Power
Power · Anatomy

Anatomy of a 20 TW machine.

Where every watt actually goes — and why electrifying the whole stack needs roughly 12 TW of clean electricity, not 20.

The three slices

Cut civilization's ~19.9 TW of continuous power into the categories that match how the energy is actually delivered:

Slice 1 — the electric grid (3.4 TW)

Lights, computers, motors, air conditioning, AI compute, electric vehicles. The “modern” slice: ~17% of total. This is the grid. When people say “decarbonize the grid,” this is what they are talking about — and only this.

Slice 2 — transport (4.3 TW)

Cars, trucks, ships, planes. Almost entirely oil. Internal combustion engines waste 70–80% of the energy as heat — only 20–30% reaches the wheels. This is the slice with the cleanest electrification arithmetic.

Slice 3 — heat (12.2 TW)

The largest and least-discussed slice. Two sub-categories:

  • Industrial heat (3.8 TW): furnaces making steel, cement, glass, ammonia. Coal and gas, often above 1,500°C. Genuinely hard to electrify.
  • Building heat / other (8.4 TW): home heating, cooking, hot water. Mostly gas boilers and biomass. Easy to electrify with heat pumps if you have the wires.

The electrification dividend

Common framing: “we use 20 TW today, so we need a 20 TW clean grid to replace it.” This is wrong, and the error matters.

Electricity-driven machines are mechanically more efficient than fire-driven ones, because they skip most of the heat-loss step.

  • Cars: an EV is roughly 4× more efficient than a gas car. Replacing the 4.3 TW of oil transport needs ~1.1 TW of electricity, not 4.3 TW.
  • Building heat: a heat pump is roughly 3× more efficient than a gas boiler. Replacing the 8.4 TW of building heat needs ~2.8 TW of electricity, not 8.4 TW.
  • Industry: 1:1 at best, often worse. Physics is stubborn at 1,500°C.

Roll those up: today's civilization, fully electrified, runs on roughly 11–12 TW of clean electricity. Not 20.

That is the dividend. It is real, and it is large.

The new demand

We are not trying to maintain today. We are trying to advance. So while electrification is shrinking the requirement, growth is adding to it:

  • AI compute: ~0.1 TW today, doubling every ~2 years. The fastest-growing demand line in the grid.
  • Desalination: infinite fresh water requires infinite power. Currently small. Will not stay small.
  • Industry at scale: ten billion people need more steel and cement than seven billion. Industrial heat doesn't shrink even with efficiency gains.

What this means for the grid

Three sequential build-outs, each harder than the last:

  1. Electrify transport and heat. Capture the dividend. Convert ~12.7 TW of fire into ~5 TW of electricity demand.
  2. Grow the grid 3–4×. From today's 3.4 TW to roughly 12 TW just to decarbonize the existing machine.
  3. Grow the grid 30×. To 100 TW, the floor for what a Type I civilization looks like. See the Kardashev page for what that means.

The bottleneck is not generation. Generation is solvable: solar is now the cheapest electricity in human history, and nuclear is back. The bottleneck is wires — transmission, distribution, interconnection queues. Generation can be added in months. A new high-voltage line takes 7–15 years.

Every AI training cluster looking for 500 MW of continuous power is bumping into this wall, not into a chip wall.