What if we are really in the foothills of the singularity?

Both series sit on one axis, because both are indices at 1873 = 100. The scissors is therefore real, not an artefact of dual scaling — the widening wedge is the productivity dividend, and it is the entire case that supply-driven deflation is survivable.

But two footnotes matter. Benign on average was brutal in episodes. The era opens with the Panic of 1873 and is punctured by the Panic of 1893 — visible here as the output dip — one of the worst depressions in pre-Fed American history. Supply-shock deflation does not exempt you from financial crises; it schedules them.

Second, the aggregate concealed a distributional war. Farmers' nominal debts stayed fixed while crop prices sank; each bushel of wheat bought less debt relief. That produced the Free Silver movement and Bryan's Cross of Gold speech in 1896. It lost. Deflation's economics are about supply curves. Deflation's politics are about who owes what, denominated in which unit.

The first principle is Ricardo's: when a factor of production becomes superabundant, its price falls toward marginal cost, and the economic surplus migrates to whatever remains scarce and complementary. AI and robotics make intelligence and manual labour reproducible. They do not make land reproducible. Or copper. Or grid interconnection slots. Or permits. Or trust, attention, and status.

So the precise prediction is divergence, not collapse. Everything reproducible falls toward its energy-plus-materials cost. Everything non-reproducible absorbs the rents.

Which means "deflation" is a sectoral statement, not an aggregate one. Whether the headline index falls at all depends on basket weights and the monetary response — the price level becomes, in a real sense, a policy choice rather than a fact about the technology.

This is Baumol's cost disease: the sectors where productivity lags become the expensive ones. A television costs a fraction of what it did; a hospital stay costs three and a half times more. The overall index stayed positive only because services carried it.

The genuinely novel thing about this shock is that AI attacks the Baumol sectors directly. Medicine, law, education, and software are exactly where a scalable cognitive workforce lands first. For thirty years the stagnant sectors were the safe ones. Now they are the target.

Note the tell at the bottom of the chart: computer software already sits deep in negative territory. That is what the arrival looks like from the inside — and it happened without anyone calling it deflation.

The hardware is not the constraint. Entry-level units already sell below $6,000; capable research platforms run $13,500–16,000; Tesla targets $20,000–30,000 for Optimus at scale — with the honest caveat that Tesla has missed every Optimus deadline since 2021 and external sales keep slipping.

So why does a robot-hour still bill at human parity? Because you are not buying the robot. You are buying reliability. The gap between the floor and the price is error rates, supervision, integration, downtime, and teleoperation — everything that makes the machine trustworthy in a real plant.

That matters because the two numbers fall on different curves. The bill of materials falls with manufacturing scale, and capital already sees it: Morgan Stanley estimates building the same robot without Chinese components would roughly triple the actuator bill alone; Figure, with no public product pricing at all, is valued near $39 billion. But the reliability tax falls with learning curves — and learning curves, historically, bend only one way.

Industry surveys now describe equipment availability — not capital, not permitting — as the binding constraint on industrial expansion. That is a remarkable sentence. Capital is the thing there is most of right now, and it cannot buy its way past a transformer queue.

Behind transformers queue the other governors: multi-terawatt grid interconnection backlogs, advanced packaging and high-bandwidth memory, EUV lithography capacity, grain-oriented electrical steel from a handful of mills — and, ironically, licensed electricians and grid engineers, some of the scarcest labour on Earth for the coming decade.

Growth theory has a formal version of this: when some sectors explode in productivity, aggregate growth ends up governed not by what we do brilliantly but by what remains essential and hard to improve. The bottleneck sectors become the expensive ones, then the dominant ones in the price index, then the political ones.

This is also where the energy transition stops being a side quest. The AI buildout is an energy buildout — the same capex wave either locks in carbon for forty years or retires it. And the governors are the silver lining: physics and permitting will not let this happen overnight. The slow ramp may be the only thing that gives institutions time to adapt.

Phase one: an investment boom with firm or rising real rates, inflation concentrated in the bottleneck sectors, and central banks leaning against overheating even as goods prices soften. Enormous investment demand, plus expectations of faster future growth, pushes equilibrium real rates up.

Phase two: the capacity lands, unit costs crater across sector after sector, and the deflationary undertow arrives — at which point conventional easing pushes on a string, because supply-driven price declines don't respond to cheap credit. Demand does. Which drags fiscal policy onto the stage.

Most singularity commentary skips straight to phase two and gets the sign on interest rates backwards. The market is currently voting on phase one, and the vote is legible in a single series.

Falling prices raise the real weight of fixed nominal debts. Collateral values sink, defaults rise, banks tighten, spending contracts, prices fall further. Every turn of the loop makes the next turn easier.

This — more than fiat ideology or a 2% target — is why sustained deflation is intolerable to modern policymakers. Every mortgage, corporate bond, and government liability is a fixed nominal claim that grows heavier as prices fall. The 19th century chose creditor stability over debtor relief and survived the choice. A world levered above 300% of GDP doesn't get that choice.

Notice the third statistic above. There is no clean aggregate leverage series for the 1880s, and inventing one to make the comparison look rigorous would be the exact failure this piece is arguing against. The comparison is qualitative — and it is still decisive.

Abba Lerner formalised the policy side in 1943 under the name functional finance: in a fiat system, taxes are not fundamentally a revenue device — they are the outflow valve that drains purchasing power to balance money against goods. The binding constraint on the sovereign is inflation, not solvency.

Here's the twist. In a superabundance scenario the valve runs in reverse. The hard problem is not draining purchasing power to fight inflation. It's injecting it — because the shock lands precisely on the wage channel, the mechanism through which output has been distributed to households for two centuries.

Which means the load-bearing institutions of the transition are claims: negative income taxes, citizen dividends, sovereign wealth funds holding the robot economy's equity (Alaska has run the prototype for fifty years), broad-based capital ownership. The Free Silver fight was over which metal backed the unit of account. This era's version is over who owns the machines.

During the transition, demand elasticity rules. When automation makes a service dramatically cheaper and demand for it is elastic, total employment in the sector can rise — the paradox Jevons identified for coal. ATMs made branches cheap, so banks opened more of them, so they hired more tellers. Healthcare is the canonical case ahead: make superior care radically cheaper and consumption explodes, pulling in human-adjacent roles faster than automation displaces them. Electricians are about to have the decade software engineers just had.

In the long run, Ricardo cuts the other way. Comparative advantage is a mathematical guarantee that humans always have something to trade. It is not a guarantee about the wage. Nothing in the theorem prevents the market-clearing price of human labour in a given task from falling below any level a person would accept.

The ATM chart is usually deployed as reassurance. Read to the end, it is the opposite: Jevons bought tellers twenty good years and then took the job anyway. Plan for the first truth; build institutions for the second.

Long-dated real yields. If markets price transformative growth, they stay elevated or climb. A collapse back to 2010s levels means the market has stopped believing.

The goods–services CPI split. Services deflation — in medicine, education, professional work — is the tell that the Baumol sectors are cracking. It has essentially never happened in postwar data. This is the single highest-information gauge on the panel.

Transformer lead times. The tachometer on the governor. When 128 weeks starts shrinking, the physical throttle is opening.

Delivered robot cost per task. Watch billed rates in real deployments fall from ~$25/hour toward the sub-$1 hardware floor. Each halving is a step-function repricing of manual labour.

Energy and compute capex as a share of GDP. You cannot fake an industrial revolution — it shows up in concrete and copper.

For half a century the labour share was so stable that economists treated it as a constant — one of the "stylised facts" of growth. Then it wasn't. Roughly seven points of national income moved from wages to capital in a generation, and that was before a scalable cognitive workforce existed.

Plate 08 showed the arithmetic of the asymmetry. This is the same asymmetry, measured, over seventy years. It is the trend the shock accelerates.

The Great Deflation ended with a losing speech about a cross of gold, and the grievances it left unresolved shaped politics for a generation. This one will stage the same fight over a different question: not which metal anchors the currency, but who holds claims on the machines.