Essay #01
The capacity to build
In 1847 a country with no computers, no diesel and no central plan laid more railway in one year than most countries have built in a generation. Britain put almost 8 percent of that year's GDP into railways. The network more than quadrupled in ten years, and no government planned any of it. Private judgment did, one line and one company at a time, and it remade the country.
The same country has now spent seventeen years and 46.8 billion pounds on a high-speed line without opening a single mile. Hinkley Point C has more than doubled its original estimate and slipped to 2030. British industry pays the highest electricity prices of the twenty-eight countries in the standard comparison.
Britons did not become a different people in the two centuries between. What changed is what a builder could count on: how much energy, how much capital, and how much permission.
The three terms. Call that the capacity to build. It is the product of three things: institutions, the effective abundance of energy, and labor input, which today means automation.
Civilizational competitiveness C ≒ Institutions I × Effective energy abundance E × Labor input (automation) L
Because it is a product, one term near zero drags the whole to zero, however strong the other two. The third term, labor input, is really two questions. Does the country hold the technologies everyone else stops without, such as advanced chips and the machines that make them, or can it reach them? And can someone in it commit sums of a wholly different size on one decision, the way a sovereign fund can? By effective abundance I mean energy you can actually use, not energy you happen to own. The same reserves become almost free power in one country and stay in the ground in another. The price of a kilowatt-hour matters less than whether a builder can draw gigawatts for decades and nobody has to count.
The equation says nothing about the wealth a country already holds. It says how much the country can still build. The set of countries that are rich today and the set that can build today are different sets, and the first is living on inventory laid down in a period when all three terms were present.
The part that gets missed is what a constraint does before any decision is taken. Options never reach the point of being weighed and rejected, because they stop being proposed. Where a premise is known to be fixed, nobody writes a business plan on it, nothing reaches an investment committee, and nothing is conceived at all. The range of ventures we now feel to be realistic, and our common sense about the scale of an investment, were both formed on the assumption that the three terms sit where they sit today.
What the substrate is.
Figure 1-1 Energy commanded per person, in human-body equivalents
one mark = 100 watts of continuous power, roughly one human body at rest; each row is 24 marks
| Period | Watts per person | Human-body equivalents (100 W each) |
|---|---|---|
| Hunter-gatherer | 300 | 3 |
| World, 1900 | 890 | 9 |
| World, 2024 | 2,430 | 24 |
| United States, 2024 | 8,450 | 84 |
Twelve thousand years separate the top row from the bottom, and almost all of the difference arrives after 1800. Values before that date are order-of-magnitude estimates.
Source: Smil, Energy and Civilization (2017), and Smil's world series for 1900; Our World in Data (CC BY 4.0) for 2024. The 300 watts is the author's estimate.
A Palaeolithic hunter-gatherer commanded roughly 300 watts, fire included. The world average today is about 2,430 watts of primary energy per person, and the United States about 8,450. For an American that is eighty-four bodies at 100 watts each. A wealthy industrial civilization is one in which every person keeps something close to a hundred invisible servants.
Over twelve thousand years the amount a human eats did not grow; what grew is the amount of work done outside the human body.
Two directions. On that substrate, civilization advances in two directions, and each has an endpoint. Direction 1 drives toward zero the cost of arranging matter the way you want it: transport, manufacture, construction, agriculture and the opening of space are all versions of it. Direction 2 drives toward zero the cost of predicting well enough to act: communication, computation, science, education and AI are all versions of that. Both lower the cost of turning an intention into an outcome, one by moving the physical world and the other by knowing what to do.
The two compete for the same problems, and the side that can converge more cheaply wins. Which side that will be can often be read in advance. Break the problem into steps and count the mass that has to move at each. In shopping, search, comparison, ordering and payment move no mass at all; only the last leg, from warehouse to doorstep, does. So Amazon beat the shopping robot people used to imagine, because its solution moved less mass.
Why the balance broke. For thirty years capital and talent flowed almost entirely into direction 2. Software copies for almost nothing, runs with little friction, and needs little energy. Of those three properties only the last depends on the era, and in a constrained world capital goes where the constraint is loosest.
Then, in the 2020s, direction 2 hit the wall of energy, and efficiency did not hold the wall off. When the cost of converging falls, uses that were never viable become viable, and demand grows by more than the efficiency saved. World data center electricity demand rose 17 percent in 2025 and will roughly double by 2030 on the IEA's estimate; facilities built for AI alone rose 50 percent in a year. Over the same period the electricity used per AI task fell fast. The saving turned straight into more computation, used more widely, more often, and for heavier work.
Capital is already saying this more plainly than any forecast. In August 2025 the regulator returned the Palisades plant in Michigan from the decommissioning process to an operating licence, the first restart of its kind. Computing demand is making investment in energy supply politically unavoidable.
So I read the money now going into physical AI and energy as a return: capital left direction 1 thirty years ago, and it is coming back.