in short
In July 2026, the Stanford Digital Economy Lab released "We Must Act Now," a statement warning that AI could drive an economic transformation larger than the Industrial Revolution but unfolding over a vastly shorter time. Sixteen Nobel laureates signed it, and the signatory list has since passed 2,000 (Stanford Digital Economy Lab; UVA Economics of AI). Coverage of the same letter pointed out that the comparison has a dark side, since the Industrial Revolution also brought sharpening wealth disparities along with its new wealth (Gizmodo). Nobody serious is dismissing the scale of what is being built. What is contested is where it lands, and the honest position is that nobody currently knows which future we're racing toward, only that we're racing.
Every piece in this series has, in its own way, been circling the same question without quite naming it. Why does a war on the other side of the planet threaten a chip fab. Why does Elon Musk need a factory the size of a small city. Why is a Dutch company's shipping manifest now a matter of national security. The proximate answer to all of it is the same: everyone involved believes that being second in this race isn't a normal competitive disadvantage. It's terminal. That belief is what turns an ordinary industrial buildout into something closer to an arms race, and arms races, historically, don't end because someone wins. They end because something outside the race itself forces a stop.
The Belief That Makes Racing Rational
The logic driving the money is worth stating plainly, because it's not irrational even if it turns out to be wrong. If advanced AI genuinely compounds, better models help design better chips, better chips train better models, and the lead extends rather than closes, then falling behind isn't a setback you recover from next quarter. It's permanent. Under that belief, spending $16.8 billion on a factory that might not work, or locking in a twenty-year nuclear contract to guarantee power nobody's certain they'll need at that scale, isn't reckless. It's the only rational move available. The uncertainty isn't really about whether the technology works. It's about whether the compounding is real, and that's a question nobody racing has actually answered yet. They are committing capital to it, at a scale we've spent this entire series documenting in gigawatts and gallons and wafer counts, before the question has been settled.
Nowhere: The Money Disappears, the Concrete Doesn't
There's a real possibility this ends the way the late-1990s fiber-optic buildout ended (IEEE ComSoc): the companies that built it went bankrupt, the capital was wiped out, and the cable sat in the ground for a decade before anyone figured out what to do with it, at which point it quietly became the backbone of the internet everyone now takes for granted. If the current AI buildout goes nowhere in this sense, the physical infrastructure this whole series has tracked, the reactors, the fabs, the water reclamation plants, doesn't vanish. It just changes owners, usually at a steep discount, and gets used by whoever's still standing once the reckoning happens. This is the least catastrophic version of failure, and it's also the one history says is genuinely likely: at a September 2026 economics forum in Shanghai, a university president put it this way: "the new form of the AI economy has been established, but AI economics has not truly begun." Installation, in other words, is running ahead of proof that the deployment pays for itself (Bitget, translating Wall Street CN's report).
Somewhere: The Gains Are Real and They Land on Whoever Already Owns the Scarce Thing
This is the outcome our own research points toward most directly, and it's worth being honest about why. Every structure this series has examined, sovereign funds bundling land, power, and compute into a single asset class; hyperscalers vertically integrating into their own nuclear supply; Terafab's entire premise of escaping a queue by owning the factory outright, is built to capture value for whoever already controls the scarce input, not to distribute it. That isn't a conspiracy. It's just what capital-intensive races have always produced. Anton Korinek, one of the economists building the actual research agenda on this question, has said it more starkly than most: "there is no guarantee that we can earn a decent living based on the value of our labor anymore," once AI's productivity effects fully land, which is why he's calling for a new system of income distribution, not assuming markets sort it out unassisted (Wikipedia, quoting a Vox interview).
History has a name for the lag this implies, and it's a sobering one: economic historians call it Engels' Pause. Between 1780 and 1840, British output per worker rose by roughly 46 percent while real wages rose by only about 12 percent; only after 1840 did wages catch up, with output per worker up 90 percent and real wages up 123 percent between 1840 and 1900 (Robert C. Allen, Oxford). Later work has revised some of the exact figures, though not the basic gap (Warwick CAGE). The technology back then wasn't fake, and the eventual gains weren't fake either. They were badly delayed. Nothing guarantees that pattern repeats, but it's the honest base rate, not the reassuring one, and every physical bottleneck we've documented in this series, who gets the ASML allocation, who owns the water rights, who signs the nuclear PPA, is really a question about who ends up owning the pause.
Everywhere: The Old Optimistic Story, Which Has Been Right Before
The other serious argument, made without irony by people who have been right about large calls before, is that this ends the way electrification and the steam engine eventually did: broad, diffuse, better for nearly everyone, eventually. "AI = massive ramp in productivity = massive ramp in demand = massive jobs boom," as one prominent technologist put it in 2026, echoing, whether he intended to or not, almost exactly what the economist Andrew Ure wrote in 1835 defending the steam engine against the same fears now being raised about AI (Baron Public Affairs). Ure was, eventually, more right than wrong. Britain did get richer. New jobs did appear that nobody could have named in 1835. But the same brief notes that in that same year Parliament heard from a handloom weaver whose weekly earnings had fallen by 80 percent (Baron Public Affairs). The honest caveat, the one this framing tends to leave out, is that "eventually" did a great deal of work in Ure's sentence: wages lagged output for decades, and nothing in that record suggests the catch-up was automatic.
Physics Doesn't Care Which Ending You're Backing
Here is the thing our own research is best positioned to say with real confidence, after five pieces spent inside the actual supply chain: the belief driving the race is moving faster than the physics underneath it, and the physics is what actually governs the clock. ASML expects to deliver about ten High-NA machines in 2027, not fifty, no matter how much conviction exists that the compounding is real. A fab takes three to five years to build regardless of who's paying for it. A desert doesn't generate more water because the returns look good on a pitch deck. Every constraint this series has documented, the machines, the materials, the labor, the water, sets a hard ceiling on how fast any of the three endings above can actually arrive, independent of which one turns out to be true. The race doesn't end when someone decides to stop. It ends when the physical world stops cooperating with the pace of belief, the way capital supercycles tend to end: not with a decision, but with a wall.
Where It Actually Lands
So: does the race end? Almost certainly, the way infrastructure supercycles typically end, not in a single dramatic moment, but in a slow collision between capital, physics, and time, resolving into some mix of all three outcomes rather than a clean choice between them. Some of this buildout will go nowhere and quietly become someone else's cheap infrastructure a decade from now. Some of the value really will concentrate with whoever already owns the scarce input, the way it has in every race like this before. And some of it, if the harder version of history repeats, will eventually diffuse outward, but only after the same kind of institutional fight that made "eventually" true the last time, not automatically, and not on anyone's optimistic timeline.
What the physics and the capital markets will not do is decide who benefits once the dust settles. That's not a technology question. It's a policy and power question, being asked right now, while the concrete is still being poured. The sixteen Nobel laureates who signed that July statement didn't predict an outcome. They called for institutions and guardrails to be built before the outcome arrives, because the premise behind the statement is that the outcome depends on institutions as much as on the technology. What lands, and for whom, will be decided by what gets built, politically, not just physically, while everyone is still busy racing.
Sources
source
covers
"We Must Act Now" statement; 16 Nobel laureates; 200+ initial signatories
Signatory list passing 2,000
The Industrial Revolution comparison's dark side
Shanghai forum on AI economics; installation vs. deployment
Korinek's labor-income and redistribution argument (Vox interview)
Output per worker vs. real wages, 1780\u20131900
Revisions to Engels' Pause estimates
Andreessen's 2026 post; Ure's 1835 argument; handloom-weaver testimony
Fiber-optic boom and bust as a precedent
