The Escape Bet

This is a forecast about a wager. Fifty years of slowing growth turned the developed world toward protecting shares over growing totals, at home first, then at the border. AI is the shared bet on a way out. Whether it pays depends on whether AI diffuses like the great general-purpose technologies; but value created and value captured are different questions, and the wedge between them, deficit-financed and priced on belief, is where the system can break.

Mercantilism is not, at root, a trade doctrine. It is the use of state power to protect an incumbent’s share rather than to grow the total. The tariffs of the 2010s were not a rupture but the moment its internal logic finally reached the coastline.

China pulled the trigger, but the gun was already loaded. The same challenge met integration in the fast-growth postwar decades, and met quotas and managed trade when Japan, an ally, arrived inside the drought. The deeper enabling condition is the slowdown of productivity growth after the 1970s and the rent-seeking that followed.

Once a country completes industrialization, growth decelerates; and when growth is scarce, people and businesses raise their returns not by creating new value but by seeking protection and redistribution: rent from land, licenses, platform monopolies, and eventually tariffs. Nobody becomes greedier; the payoff table changes. The consequence is visible in aggregate: corporations invest less and distribute more, and a growing share of measured profit is rent on protected positions, not return on new production.

The demand side eroded in parallel. Inequality has shifted income toward households that spend less of it, while aging has slowed labor-force and potential growth, with mixed effects on saving; together they have weakened private demand for three decades. The shortfall must be filled for the market to clear at a non-declining output level, and the filler, ex post, has been deficit-financed government spending, entrenched by the 2008 and 2020 crisis responses and accommodated by central-bank balance sheets. Before the 2020s, China’s trade surplus partly financed it, recycling export dollars into Treasuries. But borrowed demand is not a growth model.

China shows the same symptoms at middle income, because its investment-driven growth model built the rent machinery directly into the state: land finance, state monopolies, and a household sector suppressed by design to subsidize investment. The problems surfaced earlier and more acutely: at middle income, with legitimacy resting on growth, and with a shrinking population that, with property deleveraging and weak household demand, exerts persistent disinflationary pressure.

AI enters as the remedy to everything above. No major country will delegate it merely to grow the global common pie. In the short run the capex itself creates demand: the escape bet doubles as a demand program. Over the medium run investment must earn its expected returns, though sunk spending and contracted pipelines can carry it past the point where belief weakens; the gap today is significant.

Aside from a few capability-enabling innovations of the AlphaFold type, much of current commercial AI monetization rests on high-skill labor substitution or augmentation, which weakens wage-financed demand unless offset by lower prices, new complementary work, investment, or redistribution. Within an automated sector, deflation preserves nominal revenue only where quantity responds (software may be one); in price-inelastic activities nominal spending and the sector’s share of GDP shrink even as quantities rise, and the aggregate outcome depends on whether freed spending lands elsewhere and new tasks emerge. The elasticity of demand for cognition may be the single most important unknown of the decade.

The current surge in AI and semiconductor pricing is short-term crowding demand against fixed capacity, quantities bottlenecked at the fabs, the early windfall concentrating upstream through scarcity prices, full utilization, and richer system mix. Extrapolating it forward ignores the industry’s historic cyclicality and the uncertainty over how long the excess demand lasts.

The future return on AI capex reduces to revenue and cost: revenue to the rate and breadth of adoption and to competition; cost to how efficiently the capital and the remaining labor are utilized.

On adoption, the bear case is political as much as technical: unlike the industrial labor displaced in the 1980s and 1990s, today’s high-skill workforce has voice where it lacks walls and walls where it lacks voice, and both slow adoption below what the technology alone would allow. The revenue base is narrower than the capex implies: direct frontier-model revenue concentrates in two firms; deeper consumer penetration runs through falling prices, the consumer face of the elasticity question, where volume must outrun price; and the payoff on enterprise AI spend remains limited and uneven, so the trajectory of AI revenue is far harder to extrapolate than the trajectory of AI capability.

The bull case is broader than substitution and rides the same condition: if efficiency keeps collapsing the price of capability, new tasks, new product categories, and new firms and institutions can open, value arriving through creation, with AlphaFold-like R&D automation the extreme version: possible, unbounded if it happens, not predictable from here.

On competition, first-mover capability advantages at the model layer have so far diffused with historically short lags, and the asymmetry matters: the layer where America leads most visibly is the most contestable, the hardest physical chokepoints sit with allied firms abroad, no country controls the full stack, and the rents may not accrue to the firms making today’s investments.

On cost, past spending is sunk, and construction, procurement, and power agreements give the near-term pipeline strong though uneven contractual inertia, so disappointment cascades down the supply chain rather than stopping the spend; and the marginal dollar of AI capex migrates from operating cash flow to technology bonds and private credit, capital chasing return in a high-savings, low-growth world. The savings glut is financing its own escape bet.

One consequence runs deeper. Labor never backed money the way gold did; its role is humbler and larger: the pervasive input, and the channel through which most households earn their claims on output. As high-value labor is replaced, scarcity migrates to what remains scarce: land, energy, rent-derivable resources, branding and status, and the power that governs distribution. Wealth becomes less about skills and more about rent, as in the petro-states, a partial preview of a post-labor economy: not post-scarcity, but rent-ordered and politically distributed.

The same migration is under way among states. Deficit-financed demand is the shared condition of most major issuers, and no rival currency pairs the dollar’s depth with political neutrality, so doubt about fiat is expressed not by switching currencies but by holding the asset that is no one’s liability. Hence record central bank gold buying since 2022, sped by the reserve freeze that taught states fiat claims carry political risk; the dollar keeps its transactional dominance while its store of value role erodes at the margin.

This is what makes the present configuration fragile. The system runs on the expectation that the AI bet pays; any sufficiently negative signal can collapse it, reversing a major investment impulse and materially raising recession risk. That is the way belief-backed systems fail: slowly, then suddenly. Until the growth problem is genuinely solved, the regime will keep tariffs and export controls to protect domestic profitability, hardening around strategic industries while easing at the consumer margin.

Three futures follow, and honesty requires holding all three open. The dividend may arrive and re-anchor the system: AI raises the return to creation above the return to capture, productivity shows up in the statistics, and the debts look justified. It is the only genuinely good exit. The muddle may simply continue: the bet neither clearly pays nor clearly fails, capacity is duplicated behind walls, deficits and belief carry the system on; it is the single most likely path, and Japan has proven it can persist for decades. Or the belief inverts, through disappointing returns, a financing accident, or a political rupture, and a structure that stood on expectation is repriced all at once.