How the AI ‘bubble’ compares to history 🏛️

The Future on Margin (Part I) by Howe Wang. How three centuries of booms were built on credit, and how they break

Jan 03, 2026

One of the hallmarks of 2025 in AI analysis, was discussion and debate around the AI boom relative to the AI bubble. These days I gravitate towards original macro analysis around the AI Infrastructure buildout, economics and what history says.

I’ve been a (secret) fan of Howe Wang’s writing for a while but was blown away by his November piece Early in AI, Late in the Trade. I’m obsessed with this topic of how does the AI bubble compare with history? Not just Railway, but all of them?

Read the Introduction to the Series

I asked Howe to go into even more depth, and he went into so much depth it’s a 2-part series. This is the first one. Howe writes about many of the things I most enjoy reading about - like public technology spending, machine learning infrastructure, and enterprise AI adoption. My point is these are the central themes in 2026 for AI, and even driving the bulk of U.S. GDP. This is the big picture that actually matters.

Federal Tech Spending at the intersection of In-depth Investment Analysis 🌍

The datacenter AI rollout for infrastructure is a global event.
Demand for compute is rising exponentially.
There are historical cycles of instructure build-outs that are comparable, that we can learn from. This isn’t just about technology.

Imagine my surprise when after I asked Howe Wang for a piece, he writes back:

“This is by far the longest piece I’ve written (about 30 pages), and I really appreciate you encouraging me to do it.” 😱

This is the kind of enthusiasm and passion for shared topics I live for. We decided to break it into two parts.

The story of history and AI Infrastructure 🌊

If you know someone who might enjoy reading about this topic, don’t hesitate to share it with them.

Howe’s work and research have been cited in the Wall Street Journal, The Information, and other leading outlets. His advisory firm helps enterprises make AI/ML systems work effectively and reliably in production.

Please support the author if you enjoy topics like:

The economics and geopolitics of AI, data, and frontier technologies
Adoption patterns of enterprise AI and ML applications
Federal and enterprise AI/data spending trends

Make a Pledge

Together the series represents 6,000 words, 40,000 characters. Please understand what’s coming. The AI Infrastructure wave is also like nothing we’ve seen before. 🌊 History can still be our guide 🗺️. This is the first part in our series, to listen to it see the audio embed at the bottom.

The Future on Margin (Part I)

How three centuries of booms were built on credit, and how they break

Part I: Mississippi, South Sea, and the Railway Dream (1719–1914)

Ed: the following is meant to be read slowly and with deliberation.

History’s most famous bubbles are usually taught as morality plays: greed, gullibility, mass delusion. That’s true enough, but it doesn’t help you see the next one coming. A better way to read them is as engineering diagrams. Booms don’t form randomly. They form when the same few switches flip at once, and they’ve been flipping for three centuries.

The first switch is a future you can picture and almost touch. Not abstract optimism, but something you can point to: a continent connected by rail, homes lit by electricity, commerce rewired by the internet. A general-purpose technology that starts to feel inevitable once you’ve seen the early version work.

The other switch is a financing regime that makes it cheap to own that future today. Easy credit. Leverage. Installment terms. Securitization. Holding-company pyramids. Anything that turns a slow, capital-intensive buildout into an asset you can buy with a small down payment and carry with borrowed money. When those two combine, markets don’t wait for a 40-year diffusion curve. They compress it into three to five years of price action. The mistake is thinking technology’s timeline is the market’s timeline, and treating long-run importance as a reason prices should rise next quarter.

This essay is the longer version of my earlier post about the timeline mismatch. I want to show how these booms were financed: which instruments were sold, where leverage was embedded, and what assurances made the assets feel safe. Then the unwind arrives, and it is rarely gradual. In a levered system, belief becomes collateral. Prices fall, collateral shrinks, credit tightens, and the loop reinforces itself. The trigger is usually pretty boring: a rate hike, a missed quarter, a magazine cover doing the math, one default that makes lenders reprice the whole sector. Once confidence cracks, the structures that lifted prices start forcing exits.

Here’s the lineup: John Law’s paper-credit loop in Paris. South Sea debt engineering in London. Frontier land speculation that wrecked even the guy who mapped the frontier. Railway calls that turned paper routes into cash demands. That’s Part I. Part II is the upgraded architecture: utility pyramids that looked stable until collateral fell, RCA as the gatekeeper trade of the 1920s, and the dot-com stack of margin, junk bonds, vendor financing, and IPOs that funded a real buildout until the funding disappeared. Different centuries, same architecture.

Across these manias, the pitch is always “buy the future early.” The future is vivid. What makes it a mania is credit, the ability to turn conviction into position size. Start with France, because it is the purest form: paper used to buy more paper, until the whole system has to settle in something real.

I. Paper for Paper: Mississippi Company - Paris, 1719 - 1720

Exhibit 1: Mississippi Company Promotional Engraving Coat of arms depicting two Native Americans holding a golden horn of plenty from which the Mississippi River flows.

Source: Leonard Schenk, “Louisiana by de Rivier Missisippi” (ca. 1720), Old World Auctions

The Mississippi scheme was a story where the French state effectively outsourced public finance to John Law and told the market: this is now the machine. Louisiana was the future you could hang on the wall: rivers, trade routes, mines, a whole continent of upside, but the real substance was fiscal. Buying Compagnie des Indes shares meant buying into a company that was absorbing government debt and consolidating government-like cash flows, with banknotes and state paper woven directly into how the structure was financed. Shares were mass-market by design: you didn’t need a pile of hard cash if you could pay with government debt or bank paper, and the state’s involvement made it feel official. Prices did what they always do when “official” meets easy financing: ~500 livres in mid-1719 ran to ~10,000 by late 1719.

Then the stress test hit, not because Louisiana “wasn’t real,” but because holders tried to turn paper profits into gold/silver, and the system revealed what it actually was: confidence plus refinancing. Metallic payments were restricted, banknotes were elevated to legal tender, and then came the decisive move: a share-price peg at 9,000 livres, where the Banque Royale stepped in and exchanged its notes for company stock: printing money to buy the stock and keep it from falling. Once you’re doing that, the price isn’t being discovered anymore; it’s being defended. And defending it required repeated, very large note issues to absorb selling pressure, which showed up where it always shows up: higher prices in the real economy. Researchers estimate average monthly inflation around 4% from Aug 1719 to Sep 1720, with a 23% spike in January 1720.

Exhibit 2: Mississippi Company Share Price Chart (1719-1720) Stock price trajectory showing the rapid rise from ~500 livres (mid-1719) to ~10,000 livres (late 1719), followed by collapse back to 500 livres by 1721.

Source: Peter M. Garber, “Famous First Bubbles,” Journal of Economic Perspectives 4, no. 2 (1990): 35-54. https://pubs.aeaweb.org/doi/pdf/10.1257/jep.4.2.35

When confidence finally cracked and the cleanup turned deflationary: shares were down to ~2,000 by September 1720, ~1,000 by December 1720, and back near 500 by 1721. Louisiana didn’t move. The “future” didn’t vanish. What broke was the financing regime. Mississippi was a paper-credit loop that only works as long as the balance sheet can keep refinancing exits.

II. Debt Engineering as Equity: South Sea Company - London, 1720

Exhibit 3: South Sea Company Stock Certificate Inscribed stock certificate for £368 joint stock dated 173[8], signed by David Abarbanel as agent for stockholders Catherine Ray and Reverend Richard Ray.

Source: Spink & Son Ltd., https://www.spink.com/lot/17017000226

Britain runs the same play, just with less colonial fantasy and more explicit financial engineering. The South Sea story people remember is “trade with South America.” The story investors actually bought was: refinance the British state permanently. After years of war, Britain was sitting on a messy stack of government obligations - redeemable bonds, long annuities, short annuities. The South Sea Company offered to bundle a large share of that debt into a single corporate balance sheet. If you could consolidate public debt inside one company with Parliament’s blessing, then in theory you could lower the state’s funding costs, skim the spread, and turn government liabilities into a rising equity asset.

It felt credible because the state made it credible. Parliament approved the plan in March 1720. Ministers and MPs owned shares. The political class was financially aligned, not just through equity but through loans secured against stock. The structure is what matters. Investors didn’t need to show up with piles of cash. Debt holders could swap government obligations for shares on attractive terms, and new buyers could subscribe to stock by paying only a small amount upfront and owing the rest over time. That pulled demand forward and quietly embedded leverage into the system. As long as the share price held up, everything worked.

Prices followed confidence in the machine, not near-term earnings. Shares traded ~£120 in January, jumped as political approval removed uncertainty, and ripped higher as conversions succeeded and it became clear the company was absorbing a huge chunk of public debt. By summer, prices were running toward £700 and peaking near £950, even as new shares were sold at extreme levels - including subscriptions around £1,000 with only a fraction paid down.

Exhibit 4: South Sea Company Share Price with Policy Events (1720) Annotated price chart showing South Sea stock rising from £120 (January) to ~£950 (peak), with markers for major subscription offerings and Parliamentary approvals.

Source: Peter M. Garber, “Famous First Bubbles,” Journal of Economic Perspectives (1990)

Then liquidity tightened and investors started needing cash. The whole structure reversed. Britain didn’t stop being Britain, and public credit didn’t suddenly become worthless. What failed was the assumption that refinancing would always be available. When that window closed, installment buyers became forced sellers and loans against stock turned into margin calls. In weeks, South Sea shares collapsed: from ~£775 at the end of August to ~£290 by early October.

III. Mapped, Subdivided, Leveraged: Frontier Land Speculation - United States, 1790s - 1830s

Exhibit 5: American Land Policy & Peak Land Sales Chart (1790s-1850s) Dual chart showing peak land sales in million acres (left) and price per acre in U.S. Land Policy (right). Peak sales reached 20+ million acres in 1836 before the collapse.

Source: Digital History, University of Houston. “American Land Policy, 1785-1900.” https://www.digitalhistory.uh.edu/disp_textbook.cfm?smtID=11&psid=3836

Early American frontier land speculation is the same psychology stretched across decades instead of months, and in some ways it’s the purest version because the “future” is so easy to believe. The asset wasn’t a chartered company with opaque accounts. It was land: surveyed, titled, mapped, subdivided, and wrapped in a national story that mostly ran in one direction. Population growth, westward settlement, canals, then railroads, rising agricultural output. The belief was: “This will be worth more, and I can carry it until then.”

Speculators bought large tracts with minimal cash down, used short-term notes and bank loans to hold them, then subdivided and sold parcels, often flipping paper claims several times before a place generated real cash flow. As long as prices rose, the interest looked manageable and the leverage felt prudent. The whole system rested on one assumption: that you could always refinance patience, roll the debt, find the next buyer, extend the bridge one more season.

The unwind, as always, only required credit tightening. The triggers varied: in 1819, the Second Bank of the United States called in loans and demanded specie from state banks that had been lending freely against land; in 1837, Andrew Jackson’s Specie Circular required payment for government land in gold and silver instead of paper, just as the Bank of England was raising rates and pulling credit from American borrowers. When banks pulled back, buyers disappeared, and notes couldn’t be rolled, the “inevitable future” stopped being reassuring and became a liability.

One detail I find impossible to shake is Meriwether Lewis. He led the expedition that mapped the West, understood exactly what it was and how long development would take. And yet he died insolvent in 1809, caught in land speculation and government debt, despite being granted land himself. The future he believed in did arrive. The land did become more valuable. The country did expand exactly as envisioned. It just didn’t happen on the timeline his balance sheet required. Being right about the future doesn’t protect you if you finance it too early.

IV. When Maps Became Assets: Railway Mania

Exhibit 6: “A Railroad: From Paris to the Moon” Nineteenth-century French engraving satirizing railway stock speculation and the 1840s Railway Mania. The image shows a steam locomotive and six cars climbing a track supported by two slender spires anchored in Paris’s Fifth Arrondissement, ascending directly to the moon. The ticket window reads “A Railroad: From Paris to the Moon” and the caption begins “Industry knows no more obstacles.”

Source: Engraving reproduced in Adam Gopnik, Paris to the Moon (New York: Random House, 2000)

I read an essay collection called Paris to the Moon years ago and filed that image away as peak French whimsy: modernity, optimism, all of it. Only later did it click: the “train to the moon” cartoon isn’t really about romanticism. It was a satire about British railway stock mania at the time, about how easily we price the destination before the tracks exist, and how normal it feels from the inside.

Britain, 1844 - 1850

In Britain, the mid-1840s Railway Mania is what happens when a real technology collides with a too-easy financing regime. Rail already worked by the early 1840s (locomotives weren’t a parlor trick), but the country wasn’t yet a finished, everyday “train nation.” That gap mattered, because it created the perfect story: national network effects, time compression, cheaper freight, compounding commerce. And it created the perfect pitch: buy the future now.

Exhibit 7: Mileage of New Railway Lines Authorized by Parliament for Great Britain (1826-1850) Chart showing annual mileage of railway lines authorized by British Parliament, with dramatic spike reaching ~4,500 miles in 1846 at the peak of Railway Mania

Source: Federal Reserve Economic Data (FRED), Federal Reserve Bank of St. Louis. “Mileage of New Railway Lines Authorized by Parliament for Great Britain.” National Bureau of Economic Research https://fred.stlouisfed.org/series/M04058AUSM346NNBR

The cleanest chart of mania isn’t a stock chart. It’s Parliament. In 1844, lawmakers authorized 805 miles of new railway. The next year: 2,816 miles. Then 1846 hit - 272 Acts of Parliament, 4,540 miles sanctioned, proposed routes totaling 9,500 miles. Railway capital formation consumed nearly 7% of GDP, possibly closer to 8%. That’s not infrastructure policy. The funding window was wide open. Then it slammed shut. By 1848, Parliament authorized just 330 miles. The index of railway shares, which had peaked at 167 in mid-1845, cratered to 80 by October 1848. Of the 9,800 miles authorized between 1844 and 1850, some 3,760 miles, 38%, were abandoned outright. Not because rail stopped being useful. Because a third of the “network” had only ever existed on paper.

Exhibit 8: Railway Securities Listed on LSE and All Railways Share Index (1844-1850) Dual-axis chart showing the number of railway securities listed on the London Stock Exchange (solid line, left axis) rising from ~100 in 1843 to ~300 by 1845, while the All Railways share index (dashed line, right axis) peaked around 1,900 in late 1845 before collapsing to ~600 by 1850.

Source: Gareth Campbell, “Deriving the Railway Mania,” Financial History Review 20, no. 1 (2013): 1-27, https://doi.org/10.1017/S0968565012000285

What made it executable wasn’t just excitement; it was structure. Railway equity was designed to be bought on small upfront cash with future installments, which meant you could control a lot of exposure with limited money down, and you could trade in the promise of paying later. Then, as promotion hit full fever, the market did the most dangerous thing it can do in a capital-intensive industry: it confused maps for funding. The share-price behavior in the data lines up with that script. Campbell’s railway share index peaks in August 1845 and falls ~18% over the next three months, right as promotion reaches “unprecedented levels.” And the broader drawdown is not cute: depending on how you treat delistings and failures, the “All Railway” index falls ~66-70% from peak to trough in 1850.

Exhibit 9: Railway Share Index and Weekly Railway Calls (1843-1850) Dual-axis chart showing weekly railway calls in billions of pounds (bars and solid moving average line, left axis) spiking from minimal levels in 1843 to peaks of £2-3 billion in 1845-1846, while the Railway Share Index (dashed line, right axis) rose from ~1,000 in 1843 to peak near 2,000 in late 1845 before collapsing to ~600 by 1850. Calls were installment payments demanded from shareholders who had subscribed to railway shares by paying only a fraction upfront, with the remainder due as construction progressed.

Source: Gareth Campbell, “Deriving the Railway Mania,” Financial History Review 20, no. 1 (2013): 1-27, https://doi.org/10.1017/S0968565012000285

The part people miss is how the bust actually feels mechanically. The market doesn’t need to “change its mind” about railroads. It just needs to stop financing every railroad at once. That’s where calls come in: the industry’s weekly cash demands rise as projects move from prospectus to shovel, and suddenly “owning the future” turns into “please wire money by Friday.” The tightening catalyst is the usual blend of higher rates, weaker liquidity, and a broader credit wobble. 1847 is a classic stress point in British credit conditions, and once the marginal buyer disappears, the entire installment-and-new-issue ecosystem stops working.

United States, 1870s - 1900s

In the U.S., the railroad story was not one clean spike like Britain’s 1840s. It was a multi-decade buildout punctuated by credit-driven surges, and you can see those waves in the most literal metric: miles of track laid. After the Civil War, construction accelerated into the early 1870s, then dropped sharply as the first financing regime broke: ~7,139 miles in 1872, falling to ~4,623 in 1873, then down near ~1,524 by 1877. That first unwind was a credit shock. The Panic of 1873 was a railroad-finance crisis with a bank run attached: heavy railroad-bond promotion met a sudden refusal to refinance, a major underwriter (Jay Cooke & Co., tied to Northern Pacific) failed, and investors abruptly repriced what “safe” meant.

Then came the second, bigger wave: the 1880s, the American version of “the map becomes the asset.” Track laying went vertical again, ~11,569 miles in 1882, peaking around ~12,983 in 1887. The narrative felt self-evident: a continental grid that turned distance into time, pulled towns into existence, and converted prairie into export revenue. But railroads were also a capital-markets product. Promoters drew lines on paper, land companies and local boosters fought to get “on the route,” and Wall Street (and London) packaged the dream into securities that could be sold at scale. Bonds were marketed as “safe,” often framed as first-mortgage claims on physical assets, while equity was sold as the levered upside on national progress. And because these projects were long-duration and cash-hungry, the system drifted toward financial engineering by default: stacked capital structures, staged funding, and the quiet assumption that the next tranche would always be there.

Exhibit 10: Miles of Railroad Built for United States (1870s-1900s) Chart showing annual miles of railroad track constructed in the United States.

Source: Federal Reserve Economic Data (FRED), Federal Reserve Bank of St. Louis. “Miles of Railroad Built for United States.” National Bureau of Economic Research via FRED15. https://fred.stlouisfed.org/series/M04056AUSM346NNBR

What eventually broke was the belief that the bridge between financing and revenue could always be rolled. Railroads burned cash up front and earned it back slowly; coupons and short-term notes operated on a calendar. Once investors stopped treating railroad paper like a quasi-sovereign claim on America’s destiny, “unlimited capex” turned into a basic question: who was writing the check? By the early 1890s, the market had plenty of reasons to hesitate, including saturation, rate wars, and the simple fact that the system had been over-capitalized. The bust did not require every project to be fraud. It only required enough marginal projects to be weak, and enough leverage to be unforgiving, once the new-issue window narrowed. The Panic of 1893 turned that technological optimism into a run, and the aftermath was measurable: in 1894, ~40,819 miles were in receivership, rising to ~87,856 miles in 1895.

The third wave in the early 1900s looked more mature, but it followed the same law. Construction remained meaningfully positive: ~4,484 miles in 1906, ~3,824 in 1907, before dropping to ~1,801 in 1908. Over the long arc, the technology itself kept working. Railroad gross earnings trended steadily higher across the period. On an annual average basis, U.S. railroad earnings rose from ~$7.3 million a month in Jan 1870 to ~$114.8 million by Jan 1914, a nearly 16× increase, or ~6.5% annual growth over almost half a century.

Exhibit 11: U.S. Railroad Industry: Gross Earnings vs Stock Prices (1866-1914) Dual-axis chart comparing U.S. railroad gross earnings (blue line, left axis, monthly averages) with American railroad stock prices (red line, right axis, indexed). Shows steady 16× growth in earnings from ~$7.3M/month (1870) to ~$114.8M/month (1914), while stock prices experienced at least six distinct boom-bust cycles including crashes in 1873, 1893, and 1907. Demonstrates divergence between operational success and market speculation.

Source: Chart created by Howe Wang. Data from National Bureau of Economic Research (NBER) and Federal Reserve Economic Data (FRED)

And yet railroad stock prices moved nothing like that. They went through repeated booms, crashes, and long drawdowns: at least six distinct valuation cycles over the same span. The stock market was not primarily pricing whether railroads were useful or whether freight kept moving. It was pricing optimism and financing conditions: the availability of credit, the ability to refinance, the ease of placing new bonds, and the confidence that the next risk-off shock could be survived.

Railways are the moment the “buy the future early” trade industrializes. The asset becomes a map, the position becomes an installment obligation, and optimism quietly becomes a funding plan. When the marginal lender blinks, the unwind doesn’t need a scandal. It just needs the next cash call to arrive after refinancing disappears.

In Part II, the architecture upgrades. Utilities build pyramids of control. RCA becomes a margin-financed tollbooth. And the dot-com era perfects the trick: credit doesn’t just fund the boom, it gets booked as demand.

We’ll be sharing part II of the series shortly.

I’m Howe Wang. I write the Procure.FYI Substack, which focuses on investment analysis at the intersection of public technology policy, AI/data economics and geopolitics, and enterprise AI/ML adoption. Thank you for reading, if you enjoy this post share it with a friend and subscribe to my Newsletter too: