AI doesn't determine the outcome. Distribution does. Here are three scenarios — measured in Big Macs and kilowatt-hours per hour — that share the same AI curve and differ only in how we move the surplus.
Every economy is a heat engine. Value flows from a hot reservoir (productivity, capital) through a work-extraction mechanism (the market) to a cold reservoir (wages). The temperature difference between the reservoirs determines how much work can be extracted.
Value enters from the hot reservoir — productivity, technology, capital. The market extracts work — profit, goods, services. Waste heat dumps into the cold reservoir — wages.
The engine's efficiency is determined by the ratio between the reservoirs. The larger the gap between TH and TC, the more work you extract. Capitalism maximises this gap by design.
AI superheats TH. The question — the only question — is what happens to TC.
Big Macs per hour of income, 1970–2045. One metric. Three scenarios. The technology is identical. The distribution mechanism is not.
By 2045, the same economy, the same AI, the same burger. Version 1 gives you one burger per hour — subsistence. Version 2 gives you nine — comfortable, roughly what America had in the 1970s. Version 3 gives you twenty — genuine post-scarcity for the median person. The spread between V1 and V3 is 19×. Same technology. Same thermodynamics. Different architecture.
kWh of electricity per hour of income, 1970–2045. Same metric. Same three scenarios. But electricity is different from burgers — AI directly consumes electricity. The hot reservoir burns the cold reservoir's fuel. Data center demand is projected to double by 2030 (IEA). In Version 1, the engine eats the thing people need to live.
The Big Mac spread is 19×. The electricity spread is 40×. Because AI doesn't eat burgers — but it burns electricity. In Version 1, data centers compete directly with households for the same grid. AI demand doubles by 2030 (IEA), triples by 2035, and residential prices spike because there's no investment in new generation for people — all the new capacity feeds the hot reservoir. The engine consumes its own exhaust.
In Version 3, the sovereign fund invests in clean generation — solar, wind, nuclear, storage. AI optimises the grid. Electricity gets cheap because it's treated as infrastructure, not a market commodity. The same AI that drives prices up in V1 drives them down in V3 — if you invest structurally. By 2045, one hour of V3 income buys 1,500 kWh. The green line doesn't just win. It leaves the chart.
A Big Mac is a consumer good. AI can automate the supply chain and make it cheaper. But electricity is an input to AI itself. Every GPU cluster, every inference call, every training run burns kilowatt-hours. In Version 1, the hot reservoir isn't just growing — it's consuming the resource the cold reservoir needs to survive. AI superheats TH and depletes TC's fuel supply simultaneously. That's why V1 drops from 76 kWh/hr in 1970 to 37 by 2045 — below half. The minimum-wage worker in 2045 can buy less electricity per hour worked than their grandparent could in 1970. The engine ate the grid.
Let the system run. The cold reservoir freezes. Maximum extraction efficiency. The engine stalls when the exhaust has nowhere to go.
This is the default. No new policy, no new institutions, no intervention. AI gets deployed, cognitive labor gets repriced, minimum wage holds nominally but inflation devours its purchasing power. The cold reservoir — the bottom of the income distribution — approaches absolute zero.
A Carnot engine at maximum efficiency is an engine at maximum extraction. Every joule of value created by AI flows to capital. Wages become a rounding error. It's efficient. It's also a death spiral: when the cold reservoir can't buy anything, demand collapses, the economy contracts, and the hot reservoir cools too. The engine stalls.
Version 1 isn't sustainable. It's the system running to equilibrium. And equilibrium in thermodynamics is called death.
Build heat pumps. Push energy back up the gradient. Unions, progressive taxation, institutional trust. It works — as long as you keep pumping. The moment you stop, the system relaxes to Version 1.
From 1945 to ~1973, median wages and productivity were coupled. A dollar of new productivity meant a dollar of new wages. This wasn't natural — it was engineered. Unions, the GI Bill, progressive taxation, Bretton Woods, strong democratic institutions. A massive, sustained, expensive heat pump pushing value back from capital to labor.
It worked. The middle class expanded. Purchasing power grew. The cold reservoir stayed warm. Nine Big Macs per hour by 2045 isn't utopia, but it's comfortable. It's roughly what America felt like in the 1970s — a single income could support a family.
The problem: it requires constant external force. You're fighting the gradient. The moment political energy wanes — 9/11, financialisation, regulatory capture, institutional erosion — the system relaxes to Version 1. That's exactly what happened. The heat pump was dismantled. Not by accident. Deliberately. For the benefit of the hot reservoir.
Version 2 is engineering against physics. It works until the engineers lose.
Don't fight the gradient. Change the shape of the space. Some of the productivity increase from AI is structurally reinvested — in more productivity, in training, in human flourishing. People reskill. AI handles labour market matching. A basal floor — healthcare, education, reskilling — means nobody falls out of the economy. They find their next niche. The system grows, and keeps growing, because the cold reservoir never freezes.
Version 3 isn't about fighting the gradient. It's about changing the topology of the space. AI frees up productivity — call it 50% of everyone's time. The question is where that freed time goes. In V1, all of it flows to capital. In V2, unions and regulation push some back. In V3, the split is structural:
25% back to the individual — time, autonomy, the ability to reskill, pursue new work, or simply live. 50% back to the employer — to enter new markets, build new businesses, create the jobs the reskilled people will fill. 25% to social utilities — healthcare, education, reskilling infrastructure, labour market matching. The flywheel that keeps the whole system turning through the transition.
The basal floor is the key. Not a welfare floor — a launch floor. Healthcare so people can take risks. Education so people can reskill. AI-powered labour market matching so people can find their next economic niche in weeks, not years. The floor isn't where you stay. It's what you push off from.
Think of it as the AI version of the Norwegian oil sovereign fund. Norway didn't spend its oil wealth. It invested it structurally — so that when the oil runs out, the economy doesn't collapse. The same logic applies to AI productivity gains. If you spend them on quarterly returns, they're gone. If you invest them in the architecture of a growing economy — training, matching, the basal floor — they compound.
The structural reinvestment creates natural drag on concentration. Not by regulation. By architecture. The same way TCP/IP creates congestion control — not by law, but by protocol.
At low velocities, Newtonian mechanics works perfectly. Capital accumulates freely. Invest, grow, compound — no resistance. Capitalism is excellent at this. Leave it alone.
But as concentration approaches c — the maximum possible concentration — the structural reinvestment creates drag. The 25% flowing to social utilities, the reskilling infrastructure, the labour market matching — these aren't taxes. They're the geometry of the space. Each additional dollar of concentration costs more to accumulate because the economy around it keeps producing new competitors, new niches, new businesses. The basal floor means displaced workers don't fall out — they bounce into new roles. The sovereign fund means the surplus compounds into growth, not extraction.
You can't lobby against the speed of light. You can't capture a regulator to change γ. The architecture doesn't have an office in Washington. It doesn't have a board. Once the structural reinvestment is embedded in how the economy works — the way TCP/IP is embedded in how the internet works — it persists without political energy. That's the difference from V2. V2 requires continuous pumping. V3 changes the shape of the space.
This is why Version 3 is geometry, not engineering. Version 2 applies force against the gradient. Version 3 changes the gradient itself. The equilibrium state of the system — the place it naturally relaxes to — IS equitable. Not because anyone is forcing it. Because the shape of the space requires it.
Structural reinvestment that compounds. The 25% to social utilities funds healthcare, education, and AI-powered reskilling — so displaced workers find their next niche, not the unemployment line. The 50% to employers funds new businesses and new roles — so the jobs exist when the reskilled workers arrive. The 25% back to individuals funds autonomy — so people have the breathing room to retrain, relocate, or reimagine what they do. Not redistribution — reinvestment in the growth engine itself.
And here's the kicker: AI is what makes this possible. You can't do real-time labour market matching with pen and paper. You can't reskill millions of people without adaptive, personalised education. You can't run the sovereign fund without system-wide computation. The same AI that threatens Version 1 is the tool that enables Version 3.
Same Big Mac. Same AI. Three distribution mechanisms. The nominal numbers are irrelevant. The ratios are everything.
| Year | 🍔 Big Mac | 🤖 AI $/hr | V1 income | V1 🍔/hr | V2 income | V2 🍔/hr | V3 income | V3 🍔/hr |
|---|---|---|---|---|---|---|---|---|
| 1970 | $0.49 | — | $1.60 | 3.27 | $3.40 | 6.94 | $2.50 | 5.10 |
| 1990 | $2.45 | — | $3.80 | 1.55 | $10.00 | 4.08 | $8.00 | 3.27 |
| 2010 | $3.73 | — | $7.25 | 1.94 | $16.50 | 4.42 | $13.00 | 3.49 |
| 2025 | $5.79 | $8 | $7.25 | 1.25 | $23 | 3.97 | $19 | 3.28 |
| 2030proj | $9 | $3 | $15 | 1.67 | $32 | 4.57 | $38 | 4.80 |
| 2035proj | $10 | $1.50 | $18 | 1.22 | $45 | 5.63 | $60 | 7.50 |
| 2040proj | $9 | $0.80 | $35 | 1.09 | $55 | 7.22 | $90 | 12.0 |
| 2045proj | $8 | $0.50 | $48 | 1.04 | $55 | 9.0 | $120 | 20.0 |
In Version 1, the Big Mac goes to $48 because inflation absorbs the compression. In Versions 2 and 3, the Big Mac peaks and declines — peaking around $10 as automation enters food production, then falling as AI-driven efficiency makes physical goods cheaper. By 2045, the V3 Big Mac is $8 — cheaper in real terms than today. The burger follows the AI cost curve down, because when the surplus is structurally reinvested instead of extracted, automation makes things cheaper. That's the whole point.
The technology is fixed across all three versions. The AI is the same AI. The Big Mac is the same Big Mac. The thermodynamics are the same thermodynamics.
The only variable is distribution. How the surplus moves. Whether the engine is allowed to run to equilibrium, whether external force is applied, or whether the space itself has a shape that makes extreme extraction asymptotically impossible.
Version 1 is what happens. Version 2 is what we had. Version 3 is what we build.
AI doesn't determine the future. AI superheats the hot reservoir. What happens to the cold reservoir — whether it freezes, gets pumped, or is protected by the geometry of the space itself — is entirely a design choice.
The numbers are wrong. The ratios are path-independent. The direction is not in question. The question is which architecture we choose.