AI Too Cheap to Meter - (Part 2)

In Part 1, we looked at how AI makes intelligence cheap, where scarcity moves and how work and firms get reorganised. Those changes ripple up into the wider economy. This part covers three questions: what abundance does to prices, money and measurement; who owns the gains; and what could stop abundance from arriving at all.

Prices, money and measurement

Macroeconomics is built on stable relationships between output, jobs and prices, and an abundance shock tests all three.

1. Good Deflation vs Bad Deflation:
   • Productivity-led price falls raise real incomes, as happened with computers, phones and bandwidth
   • The danger is when falling prices meet falling wages and weaker demand
   • Central banks will need to tell the two apart in real time, with sector-level data rather than one headline number
2. Monetary Policy Under Strain:
   • A 2% inflation target can push policymakers to fight healthy, technology-driven price declines
   • The usual trade-off between inflation and unemployment can break when prices fall and some sectors shed jobs at once
   • Expect serious debate on nominal GDP targeting and new policy frameworks
3. GDP Stops Telling the Full Story:
   • GDP counts market transactions, so free or near-free AI services barely register
   • A world with a capable tutor, doctor and lawyer in every pocket could look like slow growth on paper
   • Most AI value shows up as consumer surplus, which needs new measures such as time saved and access to services
4. The J-Curve of Productivity:
   • Solow noted in 1987 that computers showed up everywhere except the productivity statistics
   • General-purpose technologies look like costs first, because organisations must rebuild workflows before gains appear
   • The productivity surge arrives after the restructuring, not during the hype
5. Money for a Machine Economy:
   • Agents paying other agents need instant, programmable, low-fee settlement
   • Stablecoins, CBDCs and real-time rails like UPI become the plumbing of machine-to-machine commerce
   • Regulators will need rules for agent identity, liability and spending limits

Who owns the gains

The biggest question isn't how much we produce. In an abundance economy, the central question becomes: who owns the machines that produce it?

1. Labor Share Falls, Capital Share Rises:
   • When AI substitutes for labor, more national income flows to owners of models, chips, data and energy
   • Society can be richer in aggregate while households with no capital fall behind
   • Abundance and inequality can grow side by side
2. Compute Is the New Land:
   • Ricardo worried about landlords collecting rent from a fixed resource
   • Frontier compute behaves similarly: concentrated, capital-intensive and rent-generating
   • Open-weight models, public compute and antitrust become tools of distribution, not only competition
3. New Models for Sharing the Gains:
   • Sovereign AI Funds: Public stakes in AI infrastructure paying citizen dividends, modelled on Norway's fund and Alaska's Permanent Fund Dividend
   • Universal Basic Income: A floor funded by productivity gains
   • Compute Dividends: A baseline allocation of AI capacity for every citizen
   • Tokenized Ownership: Fractional, on-chain stakes in AI-native firms open to ordinary investors
4. The Tax Base Erodes:
   • Most states rely on taxing wages and consumption linked to wages
   • As labor's share of output shrinks, governments must shift toward capital, land, compute or token-usage taxes
   • Each option trades off between investment, competitiveness and capital flight

What could stop it

Abundance is possible, but it isn't guaranteed.

1. Energy and Infrastructure Limits:
   • If power generation and grid buildout lag, compute costs stay high and abundance stalls
   • Permitting, transmission and clean energy investment become economic policy, not only climate policy
2. Concentration and Rent-Seeking:
   • A few firms controlling frontier models and chips could capture most of the surplus through pricing power
   • Abundance for shareholders is not the same as abundance for society
3. Transition Shock:
   • Even if long-run outcomes are good, the transition can be brutal for displaced workers and regions
   • The Industrial Revolution saw output soar while workers' real wages lagged for decades, a period often called Engels' Pause
   • Retraining, safety nets and portable benefits decide how long that pause lasts this time
4. Regulatory Fragmentation:
   • Conflicting national rules, export controls and data localisation can split AI markets into blocs
   • Fragmentation raises costs and slows diffusion, especially for emerging economies

Conclusion

An abundance economy asks central banks to tell healthy deflation from harmful deflation. It asks statisticians to measure value that never shows up as a price, and it asks the financial system to settle payments between machines.

Without deliberate design, the gains concentrate with whoever owns compute, models and energy. Sovereign funds, dividends, open models and new tax bases are the tools for spreading them.

And none of it is guaranteed. Energy shortfalls, concentrated ownership, a painful transition and fragmented regulation can each stall abundance. None of them is a technology problem; all of them are policy and infrastructure problems.