Shape of the economy

The next decade will be defined by where economic scarcity moves.

In Shape of the labor market, I argued that artificial intelligence is making cognition cheaper while robotics will increasingly make physical execution cheaper. The same technologies reshape the economy one level higher. The systems doing the automating require compute, semiconductors, electricity, machines, materials, infrastructure, and capital.

AI makes digital production cheaper. Embodied autonomy extends that process into the physical world. At the same time, electrification, reindustrialization, defense spending, and aging populations create demand that is independent of AI. The result is not a weightless economy. Digital abundance increasingly creates physical scarcity.

The useful question is no longer simply which industries will grow. It is:

Which industries are most likely to capture disproportionate economic value over the next decade?

What makes an industry attractive?

Growth alone is insufficient. The strongest industries combine rising demand, constrained supply, favorable technology exposure, and attractive value capture.

I use these factors to compare the major economic industries below. The ratings are directional judgments rather than precise forecasts. Their purpose is to expose where multiple independent forces point in the same direction.

Map of the economy

Scroll left and right to reveal all data in the table.

Where the winners cluster

The physical stack

The clearest concentration sits upstream of AI applications themselves: semiconductors, semiconductor equipment, data centers, electrical equipment, power generation, grid infrastructure, critical materials, and specialized construction.

AI demand propagates down a physical stack. More intelligence requires more compute. More compute requires more chips and data centers. More data centers require more electricity, cooling, networking, and electrical equipment. More generation requires transmission, transformers, turbines, construction, and materials. Supply generally becomes slower and harder to expand as the causal chain moves from software into atoms.

That is why industries that looked comparatively dull during the asset-light software era become more attractive in an AI-intensive one. Global data-center electricity consumption is expected to more than double by 2030, while new manufacturing and electrification add demand to the same physical system. (IEA) The opportunity is not any single source of energy or commodity. It is the infrastructure required to turn computation into reliable physical capacity.

The autonomy stack

The second cluster forms around industrial automation, machinery, advanced manufacturing, autonomous systems, and defense. AI initially automates work performed on screens. Embodied autonomy extends the same process into factories, warehouses, transportation networks, military systems, and other structured physical environments.

This coincides with a push toward more domestic industrial capacity. Developed economies are unlikely to recreate twentieth-century manufacturing through twentieth-century labor intensity. The more plausible model is more automated factories with more capital per worker, supported by robotics, machine vision, sensors, motors, controls, electrical equipment, industrial software, and specialized maintenance.

Defense is the strongest version of this pattern because technological demand overlaps with geopolitical demand. Global military expenditure reached approximately $2.9 trillion in 2025, while modern systems increasingly incorporate drones, autonomy, semiconductors, satellites, cybersecurity, advanced manufacturing, and robotics. (SIPRI) Software can scale quickly. Industrial capacity cannot, which makes the ability to manufacture physical systems increasingly valuable.

Demographic scarcity

Healthcare ranks highly for a different reason. Its demand does not depend on AI adoption. Aging, chronic disease, expanding medical capabilities, and high willingness to spend on health continue independently of the technology cycle.

AI can automate administration, documentation, diagnostics, research, and portions of clinical decision support, but much of healthcare remains physical, regulated, and difficult to scale. Healthcare and social assistance are projected to add roughly two million U.S. jobs from 2024 through 2034, more than any other major sector. (BLS) The technology can make practitioners more productive without removing the underlying scarcity of surgery, nursing, rehabilitation, dentistry, and direct care.

Shape of the economy

The strongest pattern in the map is that many of the best-positioned industries sit upstream of AI applications rather than inside them. Software remains economically important, but producing software becomes easier as intelligence becomes cheaper. The physical inputs required to deploy that intelligence remain much harder to reproduce.

The economy therefore moves in several directions at once:

The marginal worker can become less important at precisely the same time that the marginal chip, megawatt, transformer, robot, factory, data center, mineral deposit, or transmission line becomes more important. That makes ownership more consequential. The owner of a productive asset can combine software, machines, capital, energy, intellectual property, and labor while each individual input becomes more leveraged.

If I were choosing where to spend the next decade accumulating expertise, building a company, or owning productive assets, I would prefer industries where technology creates demand for the scarce input I control rather than industries where technology continuously makes my output cheaper to reproduce.

Shape of the labor market asked where human labor remains scarce. The economic equivalent is to ask what remains scarce when intelligence no longer is.

Increasingly, the answer is compute, energy, materials, machines, industrial capacity, infrastructure, specialized expertise, and ownership.