An instrument, not an oracle · updated —
What if the future has gauges you can read?
A handful of cost curves — computation, intelligence, energy, storage, biology — have been falling exponentially for decades, and they are starting to land on the same spot: the price of the essentials of a good life. This page charts the ones we could verify against primary sources, and only those. No invented numbers, no doom counters, no countdown to utopia — just the real gauges, read honestly, with every source one click away.
The Convergence Index
an interpretive model, not a measurement — formula below—
log-space progress toward seven stated cost milestones — not a percentage of the way to a good future
Read this before you quote that number. The index is a thesis wearing a dial. It answers one narrow, honest question: across seven verified cost curves, how far has each traveled — in log space, where exponentials live — from a stated starting line toward a stated milestone? We chose the milestones. We chose the starting lines. We weighted all seven equally because pretending to know better would be false precision. Change those choices and the number changes — that's what makes it a model, and we show the whole formula so you can disagree with it precisely.
And the part the dial cannot say: these gauges measure capability becoming cheap. Whether cheap capability becomes shared abundance is not on any gauge — it's a choice, made by people. That part is below, and it's the part that matters most.
Why these seven
The scarcity age had one price rule: everything essential costs a human's scarce time, so nothing essential gets cheap forever. The curves on this page are the places where that rule is visibly breaking — where an essential input to a good life is falling in price the way only information used to.
Each gauge earned its place by passing three tests, and plenty of candidates failed:
- Genuinely exponential — a sustained multiplicative fall (or rise) across decades or a clearly stated doubling rate from a primary source. Not a good year. Not a press release.
- Genuinely converging — the curves feed each other. Cheaper compute makes cheaper intelligence; cheaper intelligence designs cheaper energy and biology; cheaper energy powers cheaper compute. They aren't seven stories — they're one flywheel seen from seven angles.
- Genuinely verified — we could pull the actual series from a named primary source and link it. Two famous candidates (storage, robotics) failed this test honestly, and we say so below instead of charting them anyway.
Six of the seven are price curves. The seventh — solar actually installed — is the reality check: proof that at least one of these curves is not just getting cheap on a spec sheet but arriving in the physical world at planetary scale.
The gauges
Every chart is log-scale, because these curves live in log space — a straight line here means a constant percentage fall, year after year. Every chart names its source. Every gauge carries its own honest print, because every dataset has one.
How they converge
Any one of these curves alone is a nice story. Together they're a different kind of fact, because each one lowers the cost of the others.
Cheaper compute (Gauge I) is why a fixed unit of intelligence (III) got a thousand times cheaper in three years. Cheaper intelligence is what's stretching the task horizon (II) — from "finish my sentence" toward "carry my project." Intelligence that can carry projects gets pointed at the physical curves: better cell chemistry, better panel manufacturing, better grid design — pressing solar (IV) and batteries (V) further down their fifty- and thirty-year slopes. Cheap energy, in turn, is the feedstock of cheap compute — data centers are energy wearing a server rack — and the loop closes. Genomics (VI) is the proof the loop isn't confined to silicon: biology became an information good, and its price promptly fell off a cliff. And deployment (VII) is the loop touching the ground — gigawatts in fields, not figures on slides.
The floor of survival can drop faster than the ceiling of wages. That has never been true before. It's the one genuinely new fact on the board.
That's the thesis of this whole instrument, stated plainly: for the first time, the cost of being alive and capable — light, warmth, mobility, expert help, diagnosis, learning — is falling on compounding curves at the same time. Not evenly. Not everywhere. Not irreversibly. But measurably, and from primary sources, and that deserves a dashboard rather than a shrug.
What to watch — and what could derail it
An honest instrument panel includes the warning lights. These are the strongest real counterpoints we found — not straw men, the actual arguments, with sources. Any of them could bend these curves; some already are.
-
Capability is not adoption — the economist's brake
Daron Acemoglu's "The Simple Macroeconomics of AI" (NBER, 2024) estimates AI adds only ~0.5–0.7% to total factor productivity cumulatively over a decade — because most real work is messy, contextual, and slow to hand over. He may be wrong; he may be early; but a dashboard of capability curves owes you his number next to its own.
-
Diffusion takes decades — the dynamo lesson
Factories took ~40 years to reorganize around the electric dynamo (Paul David's famous 1990 paper). Today the U.S. Census Bureau (May 2026) finds only ~17–20% of U.S. businesses use AI at all, tilted heavily toward large firms. The curves race; institutions walk.
-
The grid is the queue — physical bottlenecks
Roughly 2,300 gigawatts of generation and storage sit waiting in U.S. interconnection queues, and the median wait from request to operation has doubled to over four years (Lawrence Berkeley National Laboratory, "Queued Up"). Panels can halve in price in a warehouse; they generate nothing there.
-
The stubborn half of the price tag — Baumol's cost disease
While TVs and software collapsed in price, hospital services, college tuition, childcare, and housing rose several times faster than overall inflation (AEI's "chart of the century"). The mechanism has a name — the Baumol effect — and it's the strongest argument that human-intensive essentials, plus anything made of land, won't ride these curves without a fight.
-
Cheaper can mean more, not spare — Jevons
When efficiency rises, consumption often rises faster — coal in 1865, and maybe compute now: "Jevons paradox strikes again," as Microsoft's CEO put it when cheap models appeared. Falling unit costs don't automatically become surplus for anyone; they can just become more usage.
-
Benchmarks are not Tuesdays — the extrapolation trap
METR itself warns that its task-horizon curve is measured on software tasks, and horizons run 40–100× shorter on messy visual computer work. A curve fitted on clean benchmarks and projected onto the whole economy is a hypothesis, not a schedule. We plot it because it's real; we label it because it's narrow.
-
Policy can bend curves backward — and did, this year
Tariffs and export restrictions pushed solar-and-storage system costs up ~9% in a single quarter of 2025 (Wood Mackenzie), and IRENA logged solar's first LCOE uptick in years. Learning curves are strong; they are not laws of physics, and trade wars are their oldest enemy.
The leading indicators we'd watch weekly, if you only watch a few: grid-queue wait times (does the physical world unclog?), the cheapest verified price of GPT-4-class inference (does the floor keep falling?), METR's next update (does the horizon hold outside software?), audited sub-$200 genome pricing, and battery pack prices each December. Those five will tell you more than a hundred headlines.
The honest both-and: capability is not distribution
Here is the sentence this page exists to keep honest: these curves make abundance possible; they do not make it happen.
Gravity doesn't distribute anything. Owners do. Laws do. Covenants do. The same falling curves could land as either of two very different worlds — the one where machines are broadly owned and essentials flow cheapest-first to the people who need them most, or the one economists have started calling digital feudalism: abundance in the warehouse, scarcity at the table, forever. Serious voices have warned about the second for years — Acemoglu and Johnson's Power and Progress documents a thousand years of technology gains that did not automatically spread, and Mariana Mazzucato's "Resisting Digital Feudalism" makes the case that the default drifts toward rent, not plenty, unless someone steers.
So no — this dashboard does not promise you a good future, and you should distrust any dashboard that does. Which road we take is not a forecast. It's a choice — made in laws and business models, and also in a thousand small design decisions by ordinary builders: what gets built for the farthest seat at the table first, what gets priced so the person who needs it most can reach it. There's an old covenant word for that build order — firstfruits: the first and best portion goes forward first, before you know how the harvest ends.
The gauges above say the capability is coming, on curves you can check for yourself. The ballot on what we do with it is still open. Open ballots are the opposite of doom.
The other half of this site
Getting everyone on the ride
Compounding curves have a cruel arithmetic: you can't compound what you don't have. The people who most need this ride are the least likely to have the savings, the credit, the slack, the network — or even the news that it's boarding. So we treated that as a problem to actually work, not a paragraph to gesture at: the ten constraints that keep people off the ride, ranked and evidenced, and the honest solutions — tagged by what one person can do today with ~$0, what takes a community, and what genuinely requires ownership or policy.
Method — the whole formula, nothing up the sleeve
The index answers: how far has each curve traveled from its baseline toward its milestone, measured in log space? In log space a fall from $100 to $10 and from $10 to $1 count the same — one order of magnitude — which is the only fair ruler for exponential curves. For a falling cost:
progress = ln(baseline ÷ current) ÷ ln(baseline ÷ milestone)
…and the mirror of that for a rising capability. Each trend's progress is clamped to
0–100%, all seven are averaged with equal weights, and that average is
the dial: —/100. That's the entire
computation — it runs in convergence.js on this page, from the data in
data.js, both plain files you can read. When a source series updates, the
dial moves.
| Gauge | Baseline | Latest verified | Milestone | Why that milestone (thesis choice) | Progress |
|---|
Where this model could be wrong
- The milestones are ours. Each one is argued, not derived. Move a milestone one order of magnitude and that gauge's progress shifts by roughly 10–15 points. That sensitivity is the honest cost of having a dial at all.
- Log-space progress flatters the past. Early orders of magnitude count as much as recent ones, so the dial reads high even while the remaining absolute gaps are large. The per-gauge "×-away" figures are the sober companion reading.
- Equal weights are a confession, not a finding. We don't know whether energy matters more than intelligence. Anyone claiming decimal-point precision about that is selling something.
- "Current" lags reality unevenly. Series end between 2022 (genome) and 2025 (batteries, METR). Each gauge states its vintage; the dial inherits the oldest.
- Cost curves are not outcome curves. Nothing here measures whether cheap capability reaches the people who need it. That is the open ballot above, and no formula closes it.
- Trends bend. Solar rose in 2022 and its LCOE ticked up in 2024; batteries rose in 2022; genome costs blipped in 2018. Every extrapolation on this page is labeled a pace, never a promise.
What we looked at and left off
Corrections
If a number here is wrong or has a better primary source, we want to know and we will fix it visibly. An instrument that can't admit error isn't an instrument — it's a billboard.