Unscarcity
Sign in for free: Preamble (PDF, ebook & audiobook) + Forum access + Direct purchases Sign In

Unscarcity Research

The Abundance J-Curve: Why Cheap Later Means Expensive Now

Electricity ran 9.9% over two years against a 6.2% basket, and three Fed officials dissented toward a hike. The buildout's bill arrives before its payoff.

18 min read 3948 words Updated August 2026 /a/abundance-j-curve-inflation

Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.

The Abundance J-Curve: Why Cheap Later Means Expensive Now

Or: the machine we are building to make everything free is, at this moment, making your power bill expensive — and the institution charged with price stability is professionally obligated to try to stop it.

The Receipt

Start with the document, because the document is unusually clear.

The Bureau of Labor Statistics published the July 2026 CPI on August 12. Headline inflation ran 3.4% over twelve months. Core inflation — everything except food and energy, the number economists actually watch — ran 2.5%, which is nearly boring. The gap between those two is energy, and energy was up 14.7%.

Most of that is gasoline, up 24.6% on the back of the Iran conflict. That is a textbook supply shock: it arrived from outside the economy, it will pass, and every central banker alive has been trained since the 1970s to look straight through exactly that kind of number.

Now look at the line underneath it. Electricity: up 4.2% over the year. Twelve months earlier, in the July 2025 report, electricity was up 5.5%. Compound the two and household electricity has risen 9.9% in two years. Over the same two years the entire consumer basket rose 6.2%.

Electricity ran about sixty percent hotter than everything else Americans buy, two years running, and it did it quietly, underneath a gasoline headline three times louder.

Brookings puts the longer arc bluntly: electricity costs are up 42% since 2019 against a 29% CPI, while metered load associated with data centers rose 31% since 2022. The Energy Information Administration’s August 2026 outlook has the average residential rate going 16.5 cents per kilowatt-hour in 2024, to 17.3 in 2025, to 18.3 this year, to 18.6 in 2027 — with national generation climbing from 4,309 billion kWh to a projected 4,632 across the same span.

Gasoline is a war. Electricity is a construction schedule. Only one of those passes on its own.

Three Dissents and a Named Culprit

On July 29, 2026, the Federal Open Market Committee held the federal funds rate at 3.50–3.75%. Three regional presidents — Beth Hammack of Cleveland, Lorie Logan of Dallas, and Neel Kashkari of Minneapolis — dissented, each preferring a quarter-point increase. It was the first time since September 2016 that three policymakers dissented in the same direction.

Two days later Kashkari published his reasoning, and it deserves slow reading, because a sitting Fed president wrote the thesis of this article on Minneapolis Fed letterhead:

“The initial cause of that inflation was, in part, a series of supply shocks, such as supply chain disruptions from the pandemic, the war in Ukraine, the trade war and, most recently, the Iran conflict. The massive investment in data centers has also added a new demand element to the high inflation Americans are experiencing.”

— Neel Kashkari, Statement on My FOMC Dissent, July 31, 2026

Note the word he chose. Not supply. Demand.

That distinction is not rhetorical, it is jurisdictional, and Kashkari says so in his very next paragraph: monetary policy is the right tool for demand-driven inflation and a poor one for supply shocks. Classify the AI buildout as a supply disruption and the Fed looks through it. Classify it as a demand element and the Fed is supposed to crush it. He then reaches for the most loaded analogy in central banking — the 1970s, when policymakers diagnosed a series of successive supply shocks, waited, and eventually concluded that tight money was necessary anyway.

Read that again with the transition in mind. The most senior officials in American monetary policy are debating whether the correct response to building the infrastructure of abundance is Volcker.

There is an irony worth pausing on. One of the three dissenters runs the Dallas Fed — the same bank that, five months earlier, published the most careful public model of exactly how much of this inflation the data centers are actually causing. We will get to their number, because it is smaller than the rhetoric and more alarming than the rhetoric, in that order.

The J-Curve Is Not a Metaphor

Economists already have a name for this shape, and it did not come from a futurist.

In The Productivity J-Curve: How Intangibles Complement General Purpose Technologies (American Economic Journal: Macroeconomics, 2021), Erik Brynjolfsson, Daniel Rock and Chad Syverson showed something specific about general purpose technologies — electricity, the computer, and now AI. A GPT is never just the machine. It demands enormous complementary investment that the national accounts measure badly or not at all: process redesign, retraining, new business models, new physical plant. Those costs are incurred and counted early. The output they enable arrives and is counted late.

The result is a curve. In the down-stroke, measured productivity looks terrible, because you are paying full freight for capability you have not learned to use yet. In the up-stroke it looks miraculous, because you are harvesting benefits whose costs were expensed years ago. Correcting for computer-related intangibles alone, the authors found total factor productivity at the end of 2017 was 15.9% higher than the official figures showed.

This is not a new observation, either. Paul David’s The Dynamo and the Computer (American Economic Review, 1990) documented the canonical case: electric power was technically available in the 1880s, and American manufacturing productivity did not accelerate until the 1920s. Forty years. The delay was not the dynamo. It was that factories had to be rebuilt around it — the old buildings were designed around a central steam shaft, and until someone tore that up and redesigned the floor plan around unit drive, electricity was an expensive way to do what the shaft already did.

Which brings us to the extension this article wants to make, the thing the pitch decks miss.

The productivity J-curve has a price-level twin.

Same mechanism, different instrument. When the complementary investment a GPT requires happens to be a metered utility, it does not appear in the accounts as capital expenditure. It appears in the Consumer Price Index as the cost of living. A gigawatt of new generation built to serve a training cluster lands on a household’s bill as a rate increase — indistinguishable, to the index, from one caused by a fuel spike or a lazy monopoly.

The CPI was designed for an economy where investment and consumption are separable activities. Steel bought for a factory is not in the basket. Electricity bought to run a GPU warehouse raises the price of the identical kilowatt-hour your refrigerator draws, and that is in the basket. The nation’s most consequential capital-formation program is being partially routed through a consumer price. No index on Earth is built to see through that.

We are measuring the construction of abundance with an instrument calibrated to detect scarcity, and it is returning the only answer it knows how to give: too expensive.

How Big, Really?

Vibes are not evidence, so here is the Federal Reserve’s own arithmetic.

In March 2026 the Dallas Fed modeled precisely this question and published the range. If every proposed data center were connected and run flat out, annual PCE inflation in 2030 would be 1.02 percentage points higher through retail electricity prices alone. The authors call that scenario unrealistic, and they are right to. Under plausible buildout and utilization assumptions, the effect is 0.04 to 0.13 percentage points by 2030, with the mid-capacity case running 0.05 points this year and rising every year after.

Two things are true about those numbers at once, and holding both is the whole discipline here.

First: a tenth of a point is not a crisis. Anyone telling you data centers caused the inflation of 2026 is selling something. Gasoline and shelter dwarf it.

Second: it grows every single year, and it is the one component whose cause is a signed contract rather than a passing shock. The Dallas Fed’s own framing is that the impact “grows substantially over time” in every scenario they ran, and that beyond 2030 the effects rise sharply absent major new gas or nuclear generation.

Then there is the finding nobody has priced politically. The sensitivity analysis shows that if new solar arrives at only a quarter of proposed projects, the inflationary impact nearly doubles. Which means a meaningful share of “AI inflation” is not a fact about AI at all. It is a fact about permitting. The inflation is, in part, a policy choice about how fast a country lets itself build generation — a choice made in county zoning meetings and interconnection queues, then billed to households and blamed on the Fed.

The Cruelty of the Trough

Now the asymmetry that makes this politically radioactive.

The costs of the buildout are metered, monthly, and universal. Everyone with an electric meter pays, including the four-fifths of American firms and the large majority of households with no involvement in AI whatsoever.

The benefits are denominated in a good almost nobody buys directly. Between November 2022 and October 2024, the cost of querying a model at GPT-3.5-level performance fell from $20 per million tokens to $0.07 — a more than 280-fold collapse in roughly eighteen months, per Stanford’s AI Index. That is one of the fastest deflations ever recorded in any good, anywhere.

Not one household’s grocery bill fell by a cent because of it. Nobody buys tokens.

For cheap intelligence to become cheap anything else, it has to travel through firms into prices. So look at the pipe. The Census Bureau’s Business Trends and Outlook Survey put the national AI use rate at 19.8% of firms as of May 3, 2026, hovering between 17% and 20% for six straight months — and that is on the generous definition, since the Bureau widened the question in November 2025 from AI used “in producing goods or services” to AI used “in any business function.” One in five firms, counting the ones using it to draft emails. Information sits at 39.7% and finance at 33.9%; the sectors where the median household actually spends money sit far below the average.

There is a good objection here, and it deserves a straight answer. The Fed’s own research note on monitoring AI adoption (April 2026) reports the firm-weighted rate at about 18% while the Atlanta Fed’s employment-weighted estimate reaches 78% — because 95% of American firms are tiny and the giants employ everyone. So most workers are already at an AI-adopting employer, even though most firms are not.

Both numbers are real, and neither rescues the household. Adoption is not pass-through. A firm that adopts AI and banks the savings as margin has changed its income statement, not your receipt. That is the entire argument of value migration: when an input goes free, the money does not evaporate, it relocates to whatever stayed scarce. Right now what stayed scarce is electricity — which is precisely why the scarce thing is the thing getting more expensive on your bill.

So the trough has a shape you could not design better if you were trying to kill a transition on purpose: diffuse metered costs today, concentrated invisible benefits tomorrow, and an election in between.

Why This Is the Bootstrap Paradox Wearing a Different Suit

The Bootstrap Paradox named the funding problem: you must build post-scarcity infrastructure using scarcity-era wealth, which depreciates as the project succeeds. The J-curve names something adjacent and, in the short run, more dangerous.

It is not only that scarcity-era wealth must fund the transition. It is that scarcity-era instruments must score it — and they score it as a malfunction.

The CPI reads capital formation as cost of living. The federal funds rate, the one lever the Fed actually holds, has no setting for “cool household demand but keep financing the terawatts.” Raise rates and you raise the cost of capital for the most capital-hungry construction program America has attempted in generations. Jason Furman’s decomposition of the first half of 2025 found that information-processing equipment and software amounted to about 4% of GDP while producing 92% of GDP growth; strip it out and the economy grew 0.1% annualized.

Tightening into that is not a precision strike on data centers. It is a blunt instrument swung at the only load-bearing wall in the room.

This is the same structural problem the Electron Gap describes in physical terms — exponential compute demand meeting a grid that grows a few percent a year — translated into monetary policy. And it sits one layer above the fight over who pays for AI’s electricity, a distinction worth being precise about, because the two problems are genuinely different and confusing them is how good policy gets wasted.

Cost allocation is a distributional problem: should a retiree in Ohio subsidize a hyperscaler’s substation, or should the hyperscaler pay its own way? Large-load rate classes are a real answer to a real question.

The J-curve is a temporal problem. Solve allocation perfectly and the trough is still there. You will have changed who pays during the down-stroke. You will not have changed the fact that somebody pays for a decade before anybody benefits. Perfect fairness inside the trough is still the trough.

And the resolution, in this framework’s terms, is what the Foundation exists to provide. If energy is a guaranteed floor rather than a metered market good, the buildout’s cost stops landing on the household ledger as a monthly threat and starts landing where capital formation belongs — which is also, not coincidentally, the argument for pricing the new economy in kilowatt-hours rather than in an index that cannot tell investment from consumption.

Texas Already Voted

If you want to know whether the trough decides the politics, stop theorizing and read the news from three weeks ago.

On August 3, 2026, the governor of Texas — the most build-friendly jurisdiction in the developed world, the state whose entire brand is “we will let you build it” — ordered a pause on new data center approvals pending a full audit by the Public Utility Commission and ERCOT. The trigger was arithmetic: ERCOT’s interconnection queue had gone from 233 gigawatts in January to 474 gigawatts by summer, roughly 90% of it data centers, in under six months.

The effect was immediate and measurable. In its August outlook the EIA cut its Texas load-growth forecast for 2027 from 14% to 6% — less than half — citing the pause directly.

That is the whole thesis in one news cycle. Not a bubble popping. Not a technology failing. A buildout getting slower because voters with electric bills got loud, in the reddest, most permissive energy state in America, eighteen months into the down-stroke. The trough is not a forecast. It is already setting policy, and it is setting it in the direction of less buildout, which — per the Dallas Fed’s own sensitivity analysis — is a decision that makes the remaining electricity more expensive, not less, if the generation slows alongside the load.

What Would Actually Shorten the Trough

Three levers, in descending order of how little anyone is currently pulling them.

Build generation faster than you build load. The Dallas Fed showed the inflation impact nearly doubling if solar underdelivers. That is an enormous policy lever disguised as a technical footnote. Every month shaved off an interconnection queue is basis points off the CPI. Permitting reform is anti-inflation policy, and almost nobody argues for it in those words.

Make the payoff legible sooner. Brynjolfsson, Rock and Syverson’s central point is that the down-stroke is partly a measurement artifact. We have no statistical apparatus that reports “this quarter’s price increase purchased four gigawatts of permanent generating capacity.” Until something publishes that number next to the CPI, the buildout will keep appearing in the only frame available: a bill with nothing attached.

Stop making the central bank the only institution with a lever. The Fed did not ask to arbitrate the pace of the AI transition. It has a dual mandate, one blunt instrument, and now a demand shock that is also a capital-formation program. Kashkari’s dissent is honest, coherent, and — if you take the abundance thesis seriously — aimed at the wrong target. That is not a criticism of Kashkari. It is an indictment of an institutional design in which the only body that can respond quickly to a civilizational buildout is the one whose only move is to make it more expensive.

The Honest Caveats

Two, and they cut hard.

The J-curve framing is abusable. “Costs now, benefits later, trust us” is the argument every boondoggle in history made on its way to the write-off. It is unfalsifiable in the moment. The only defense is to state in advance what the up-stroke should look like and when — falling unit costs in named sectors, measurable productivity in the firms that adopted, electricity prices decelerating as generation catches load — and to treat their absence as evidence rather than as a reason to extend the runway.

If it is a bubble, there is no up-stroke. A J-curve requires the second half. Furman’s statistic cuts both ways: an economy where one investment category produces 92% of the growth is either bootstrapping its successor or dangerously concentrated, and we will not know which until the capacity is either used or stranded. Rail mania built a permanent continental network and wiped out a generation of investors, in that order, on the same track. Both halves of that sentence are load-bearing.

What this article claims is narrower and, I think, survivable: whatever the buildout ultimately proves to be, its costs are being measured in scarcity-era instruments that structurally cannot represent it as investment. That is true if it is a bubble and true if it is the foundation of the next century.

The Bottom Line

The labor cliff argument has always been that the machines arrive faster than the institutions. This is the same story told through a price index, and it is arriving ahead of the layoffs — the bill before the pink slip, which is a sequencing almost nobody planned for.

The terawatts and the trillions required to reach post-scarcity have to be spent inside an economy with exactly one vocabulary for spending: cost. Not investment. Not construction. Not bootstrapping. Cost. So the transition registers on the instruments as inflation, the institution built to fight inflation reaches for the only tool it has, and the households paying the bill are asked to be patient for a payoff denominated in a commodity they have never purchased.

That is the trough. And the trough, not the destination, is where the politics of the abundance transition will actually be decided — because a civilization that loses its nerve in the down-stroke never sees the up-stroke, and never learns what it gave up. The Star Wars path does not require anyone to choose it. It only requires that the buildout finish under owners who never had to ask permission, while everyone else was busy fighting the electric bill. Compute landlords can wait out a trough. Households cannot.

Nobody in this story is a villain. The Fed is doing its job. Kashkari is reading his data correctly. Abbott is answering constituents who are not wrong about their bills. The CPI is measuring exactly what it was built to measure. That is what makes it hard: there is no conspiracy to expose, only an instrument panel calibrated for the wrong century, and a decade of expensive electricity between here and the free kind.

Unscarcity is a book about building the institutions before you need them, rather than discovering in the trough that you never had any. Get the book — the down-stroke is not a forecast, it is on your last three electric bills.

Further Reading

References

Every abundance transition in history ran through a trough. The difference this time is that the trough shows up on a monthly bill with your name on it, itemized, while the payoff shows up as a number in a research paper about tokens. That is not a fair fight for public patience. Which is exactly why the institutions have to be built before the down-stroke, not during it.

Share this article:

Share Share Share