The sixteen source sections preserved in document order.
For the last few years, I've been searching for useful historical analogies to make sense of what we're living through. I eventually brought them together into a thesis I call the AI Supercycle.
History can help us understand the present, but only up to a point. Historical analysis is difficult precisely because the moment analysts begin converging around the same analogies and the same narrative
The 1873 comparison quietly assumes that America then held the position America holds now.
It did not. The gap is enormous.
America in 1873 was the emerging market. It was a capital-importing industrialiser, building its infrastructure substantially with other people's savings. The Federal Reserve's own historical work on the period carries the title Foreign Portfolio Investment When the United St
Press the best analogy available and the limits show up fast.
The monetary regime has no counterpart, and it is not what people usually say. America in 1873 was not on a simple gold standard. Greenbacks had been inconvertible since 1862. The Specie Resumption Act only passed in 1875, and resumption came on 1 January 1879. So this was a contested, partly-paper regime moving toward gold — and there
It is tempting to treat a past episode as a solved case. Setup, mechanism, break, aftermath, legible end to end. That is what makes an analogy feel like analysis.
It is nothing of the sort.
The causal chain we recite was assembled afterwards, by people who already knew the ending. It was not visible while things were happening. Often it was not even the argument being had.
Britain in 1846 was not
Two things. Neither is prediction. Neither requires knowing which year we are in.
A repertoire of mechanisms. Each episode isolates one specific way a buildout goes wrong, described concretely enough to check whether it is present now. The unit of transfer is the mechanism and the thing it tells you to look at. Not the ending.
Smaller mistakes. Capital-heavy general-purpose technologies get overbu
Britain, 1850s to 1870s. Capital export rose through the 1850s and 1860s and averaged about a third of national savings from 1870 to 1914. Overseas assets went from around 7% of national wealth in 1850 to roughly a third by 1913. American railways took close to 9% of it.
What it teaches. The creditor survives the borrower's crisis. Britain funded booms that broke, repeatedly, took the write-downs,
America, 1873. Cooke, the Northern Pacific, the September suspension.
What it teaches. The intermediary breaks before the assets do. Cooke did not fail because trains stopped running. It failed because it funded long-dated assets with short-dated money and could not place paper it had already borrowed against. The railroads mostly failed afterwards — as a consequence of the credit event, not a cau
British railways, 1840s. In 1846 Parliament passed 272 Acts incorporating railway companies, covering about 9,500 miles of proposed route, with authorised capital around £132 million.
Authorised is not spent, and the difference matters. Actual railway investment was £30 million in 1846 and peaked at £44 million in 1847 — almost 8% of British GDP, and twice that year's military budget. It did not f
Telecom, 1996 to 2001. More than 2,000 competitive carriers raised around $82 billion by the end of 1999. Industry capital spending went from about $56 billion in 1997 to roughly $120 billion in 2000.
Then the equipment makers began financing their own customers. Lucent disclosed maximum commitments of about $8.1 billion. Nortel committed around $3.1 billion with roughly $1.4 billion drawn. Cisco
Electrification, 1892 to 1935. Samuel Insull built something unambiguously real. Chicago Edison became the largest electricity producer in the world, and he drove the economics that put power into ordinary homes.
Above those operating utilities he stacked a pyramid — dozens of holding companies over hundreds of operating ones, capped by two large investment trusts, with an original investment of w
Shale, 2010 to 2020. US shale burned roughly $300 billion of negative free cash flow between 2010 and 2019, impaired more than $450 billion of invested capital, and produced over 190 bankruptcies. Production more than doubled, peaking near 13 million barrels a day. And it caused no systemic event.
The tempting explanation is that the paper never reached the financial system's core. That is false,
Aircraft finance, 1994 to now. Enhanced equipment trust certificates have financed aircraft since the early 1990s, with senior tranches often rated well above the airline's own credit. Airlines go bankrupt constantly. The structure works anyway.
Get the law right, because the shorthand version is wrong. Section 1110 of the US Bankruptcy Code does not simply exempt aircraft from the automatic stay.
None of these supplies an ending. What they supply is a set of questions with something observable attached:
Is the creditor's exposure spread across borrowers, geographies and horizons?
Is the intermediary funding long assets with short money?
Is the adjustment appearing in the unbuilt rather than the built?
Have revenue quality and credit quality stopped being independent?
Is anyone borrowing ag
One caution first, because this is where memos like this overreach.
What the August 2026 announcements establish is directional. Nvidia signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR around independent compute-financing platforms, with an ambition to mobilise more than $500 billion of third-party capital. The institutions are to underwrite
Three propositions. Theses, not findings.
The efficiency clock. The claim that no previous buildout had an input whose unit cost collapsed by orders of magnitude while the buildout ran is provocative and not yet tested against a systematic survey. If it holds, it cuts both ways — expanding the market while obsoleting the specific collateral in the specific vehicle.
Absorption speed. Dark fibre wai
The consensus has largely settled on 1873, and for good reason. It is probably the closest historical analogue we have: a genuine technological breakthrough, a massive physical buildout, a credit shock rather than a collapse in end demand, and a financial intermediary that failed before the underlying infrastructure did.
But the analogy is also incomplete.
In 1873, the United States was the capita
