Essay #04
The abundance cascade
On a week abroad with 2GB of roaming data, you ration yourself. Whether something is finite or unlimited works on the mind far out of proportion to the actual scarcity. Give the same traveler an unlimited plan and the phone never rests; they are out among the sights from morning to night with a voice assistant running the whole time.
We decide where to put capital from inside a constraint. Each business that gets built makes other businesses possible, and one that is never started because energy is scarce takes with it all the others that would have grown out of it. From inside a constraint, we cannot really picture what we would do without it.
On energy, our civilization is that traveler with 2GB.
Three centuries of evidence. When the cost of energy collapses, existing demand gets cheaper to serve, and demand itself multiplies.
The longest record humanity has of driving a cost down is artificial light. On William Nordhaus's measurement, the real price of a thousand lumen-hours fell by 99.97 percent between 1800 and 1992, a factor of about three thousand.
Figure 5-1 Three centuries of light: the purest case of the cascade
| Year | Price |
|---|---|
| 1800 | $7.86 |
| 1850 | $1.8 |
| 1900 | $0.2 |
| 1950 | $0.02 |
| 1992 | $0.0023 |
| Setting | Share of GDP |
|---|---|
| United Kingdom, 1700 | 0.72 (best fit) |
| World, off grid, 1999 | 0.72 (best fit) |
| World, on grid, 2005 | 0.72 (best fit) |
| Range, 17 observations | 0.39 – 1.30 |
The three points on the right are the empirical settings the paper reports, three centuries apart, from tallow candles to grid electricity. The shaded band is the paper's range of 0.39 to 1.30 percent of GDP across seventeen observations; the line through it is its best fit of 0.72, not a value measured three times over.
Source: Left: Nordhaus (1996), in The Economics of New Goods. Right: Tsao et al. (2010), J. Phys. D 43(35), 354001.
So when light became three thousand times cheaper, did people spend less on it? Across three centuries, six continents and five different lighting technologies, the share of income spent on light stayed inside the same band. Nobody pocketed the saving; people bought more light: streets lit all night, factories running around the clock, a lamp in every room. Nothing in that mechanism is special to lighting, and light is only one use of energy.
What is waiting behind the price. What gets unlocked next can be read off the cost structure. Between a quarter and two fifths of the cost of desalinating seawater is energy, so cheap power takes fresh water off the list of binding constraints in dry regions. Energy is 40 to 70 percent of the operating cost of a vertical farm, so cheap power cuts agriculture loose from land and climate. Launch cost has already fallen twentyfold, from roughly 54,500 dollars a kilogram on the Shuttle to roughly 2,720 on Falcon 9. Cut it another hundredfold and earth orbit becomes somewhere you put factories.
The cascade runs back through the home. The goods that priced children out, in the previous essay, were the goods automation had not reached, and the reason it had not reached them was the same shortage of energy and of machines. There has only ever been one cure for Baumol's cost disease: raise productivity in the sick sector directly, with technology. When cheap power and machines that can do physical work reach housing, schools, childcare and hospitals, the one part of the price system that only ever rose starts to fall, and the price trap on children described in the last essay loosens. So cheap energy does two things. It multiplies demand for what already exists, and it brings back the people who would demand anything at all.
The order of supply, and its clock. The end state is fusion, but fusion is several decades and several hundred reactors away from a tenfold fall in cost. What carries the transition is ordinary large reactors and small modular reactors built in series in a factory, with fossil fuels as a bridge and renewables riding on top of firm power. Those start feeding the grid at the end of the 2020s and through the 2030s. Prices move in the 2040s.
Price is set by how many units get built, not by scientific breakthroughs. Unit cost falls at a steady rate every time cumulative production doubles: solar modules fell 20 to 25 percent per doubling, and more than 80 percent in the 2010s alone. What small modular reactors are really attempting is to turn a reactor from a civil engineering project built on site into a product that comes off a line, which would move nuclear from a construction learning curve onto a manufacturing one.
The demand side is slower still. Once the constraint comes off, supply grows on a timetable of years. Demand grows on a timetable of generations, and I call the gap between them the valley of demand.
An economy needs at least four head-counts: workers, customers, risk-takers and taxpayers. Machines can replace only the first. A factory with no one to sell to is a debt, not an asset.
And the four do not come back together. A person is a customer from the day they are born. Their earnings do not exceed their own consumption until about thirty. The founders of the fastest-growing companies are on average in their mid-forties, and the odds of success keep rising until nearly sixty.
Suppose the constraint came off completely in 2035, and that fertility in the rich countries jumped that same year to replacement level and stayed there. That cannot happen, but nothing faster can happen either, so the valley it produces is the shallowest one available, and it still lasts half a century.
For that half-century, the returns that can be relied on are the ones that do not depend on head-count: stakes in the automated capital itself, and the infrastructure of energy and computation. In a world short of customers, the power sits with whoever can buy.