The AI CEO's Dilemma

Whether we like it or not, we're all AI economists now: will progress hold, will the buildout tank the economy, and what will AI do to work?

deslop.media's AI Economy Model v1 is a fictional world with rules taken from our real one. Every year, AI CEOs decide how much Gold goes to Data Centers and how much compute to training, hoping to earn it back over a long horizon. Smarter models open new markets only they can serve, faster frontiers let laggards catch up cheaply, and nobody knows how much work the frontier will find.

What can be learned from this game?

So we ran the sandbox as an instrument: 68,200 seeded 12-year seasons across four sweeps of demand regimes. Here is what we pinned down before running anything. Three labs, two of them bots playing fixed strategies. No funding market: nobody raises outside money, so each lab lives on its starting 1,000 Gold. And demand? We don't claim to know: each regime is drawn from bounded ranges, every draw weighted equally. Together the three troves put 3,000 Gold on the table; the industry "pays back" when the labs finish the season holding at least that much among them. Every percentage that follows is a share of our model worlds, not a forecast about the real one.

Watch coverage, not elasticity

Across 10,800 of them the observer saw only what the public sees: revenue, served work, the going rate. The best early tell of payback is monetization coverage, cumulative revenue as a share of invested capital: by year 3 it scores 0.89 (a coin flip scores 0.5).

The year-3 rule of thumb in this economy: if cumulative revenue has reached about 4% of invested capital, payback runs at 92%; if it has not, 31%.

The exact line is 4.2%, 126 Gold on 3,000. Using measured intelligence elasticity — a measure of how additional model capability or intelligence creates additional volume of demand — to predict payback at year 3 gets it right only 43% of the time, worse than a coin flip, and it worsens to 37% by year 6. In v0.1 of the AI Economy that's because the statistic is relative, not absolute. In worlds headed for a poor outcome, demand is light early on, and even small amounts of additional work become big percentage jumps — the doomed worlds look responsive, but from too small a base.

Missing the curve mortgages the out-years

Payback is backloaded: worlds on course at year 3 still earn a median 95% of lifetime revenue after it, and 77% after year 6. The back half or not at all.

Falling behind narrows the door, closing it almost shut by year 10: more than 80% behind, payback is 4.6%.

Failing worlds are the front-loaded ones: by year 6 an eventual failure has served 27% of its lifetime work, a payer 18%. If the work you can find today is most of the work there is, the capital is already stranded.

A step function of intelligence elasticity

Price elasticity of demand is the dial from Econ 101: when something gets cheaper, how much more do people buy? Near 1, demand stretches in proportion; near zero, it barely moves. Here, it is the shape of the price ladder: money spread across cheap markets or concentrated at the frontier.

Intelligence elasticity of demand is the same idea with smarts in place of price: when a model gets one notch smarter, how much new work unlocks? At 1, each step opens as much new demand as the last; below 1, the market wants ever-bigger IQ leaps and demand thins toward the frontier.

Formal definitions, and every number this article leaves out, live in the companion manuscript.

Stretching the ladder across 24,000 seasons, payback is a step function of intelligence elasticity: at or above 1, the industry paid back in every run under every price structure; around 0.9 and below, it collapses toward zero unless prices are flat and cheap work carries the industry.

One notch sparser drops payback to 45%; two notches, near zero. The flat ladder holds at or near 100% even at the sparsest arrival tested.

Where the Gold ends up

The Data Center builders get paid up front, so they never carry payback risk; the labs do, and customers keep most of the value. Industry payback is not lab payback — the usual story: all 4,000 baseline seasons paid back while a serve-only lab recovered its trove in 11% of worlds.

Industry payback is decided by whether demand keeps arriving within reach of the Intelligence it can afford to build — and the earliest honest evidence is not any measured elasticity, but whether coverage is on its curve by year 3.

None of this values a stock or ranks a real lab. The full model is in the manuscript; the decisions are yours.