P.S. A Column On Things

By PAUL E. SCHINDLER JR. I am from Portland, Oregon, Beaumont ’66, Benson High ’70, MIT ’74. Some things are impossible to know, but it is impossible to know these things.

Big Bubbles

LLM SLOP LLM SLOP LLM SLOP (please note: there is no such thing as AI or SI. There are only large language models)

Railroad speculation, the 1929 stock-market boom, and the AI investment boom: an historical comparison

Research standard: I will distinguish documented figures from estimates, reported financing commitments, and figures that cannot presently be established. I will not manufacture a comparable number where the historical record does not supply one.

There is an important methodological issue with this comparison: railroad securities, stock-market margin loans, and borrowing to finance AI infrastructure are three different forms of financial exposure.

  • Railroads, circa 1870: Capital raised through bonds, mortgages, and stock to construct physical infrastructure.
  • 1929: Equity speculation, including loans to investors purchasing securities on margin, alongside corporate and investment-company leverage.
  • AI, 2026: Capital expenditure on data centers, power infrastructure, chips, and networking, financed through corporate cash flow, bonds, loans, leases, joint ventures, and private investment.

Consequently, comparing the headline amounts without separating debt from equity and investment commitments would produce misleading conclusions.

I will use nominal historical amounts and their inflation-adjusted 2026 equivalents where a defensible conversion is possible, and distinguish those from their share of contemporary GDP. An inflation adjustment does not itself convert a historical investment into an equivalent share of the modern economy.

1. The railroad boom and the Panic of 1873

The railroad expansion of the late 1860s and early 1870s was a major capital-investment cycle. It is important, however, to distinguish the investment boom from the financial panic: the Panic of 1873 began in September 1873, following the failure of Jay Cooke & Company, a principal financier of the Northern Pacific Railroad.

The historical evidence documents the following:

  • At the beginning of 1870, the United States had 48,860 miles of railroad.
  • In 1869, 6,568 miles of railroad were opened, with reported construction and equipment costs of approximately $358.7 million.
  • A University of Virginia historical account reports that approximately 10 percent of nonagricultural labor was devoted to railroad expansion and that about half of the nation’s metal production went toward railroad construction.
  • Railroad overbuilding and the subsequent reduction in construction contributed to the economic contraction that followed.

Sources: the 1870 Commercial and Financial Chronicle; the U.S. Census historical statistics reproduced by Virginia Tech; and the historical railroad-finance tables published by the Census Bureau.


Railroad debt and equity

The Census Bureau’s historical railroad-finance table provides a particularly

YearRailroad stockFunded debtTotal liabilities
1873$54.18 billion$51.09 billion$105.28 billion
1874$55.98 billion$65.70 billion$124.34 billion
1875$67.00 billion$67.00 billion$133.99 billion

Inflation adjusted — expressed in estimated 2026 dollars.

Two qualifications are essential:

  1. These are figures for the entire railroad system, not exclusively speculative or distressed railroad securities.
  2. Funded debt is a debt measure. Stock capital is equity. Combining them into total liabilities does not make the entire amount debt.

The 1873 funded-debt figure is therefore the defensible starting point for a comparison with later borrowing.

Census+1

2. The stock-market bubble of 1929

The Federal Reserve History account documents that the Dow Jones Industrial Average increased from 63 in August 1921 to 381 in September 1929, approximately a sixfold increase.

The subsequent collapse was severe:

  • October 28, 1929: the Dow fell nearly 13 percent.
  • October 29: it fell nearly 12 percent.
  • By mid-November: it had lost almost half its value.
  • By summer 1932: it was 89 percent below its September 1929 peak.

The Federal Reserve’s historical account also describes the widespread use of margin borrowing: investors purchased stocks with a fraction of the purchase price and borrowed the balance, using the purchased securities as collateral.

Federal Reserve History

Margin debt: the relevant comparison

One of the most useful modern scholarly reconstructions is Borowiecki, Dzielinski, and Tepper’s study, “The Great Margin Call: The Role of Leverage in the 1929 Wall Street Crash.”

It reports:

MeasureOctober 1929
New York City broker-dealer borrowings$8.525 billion
Bank-funded portion of NYC broker-dealer borrowings$1.885 billion
Nonbank-funded portion$6.640 billion
Estimated total margin loans against stocks$22.1 billion
Estimated margin loans to constrained investors$17.7 billion

The $22.1 billion figure is an estimate of total margin loans, not simply the amount borrowed by New York brokers. These measures must not be added together: they overlap.

The distinction between bank and nonbank lending is especially important. In 1929, nonbank sources supplied most of the reported New York broker-dealer borrowing.

The same study estimates total U.S. stock-market capitalization at approximately $126 billion in early October 1929. That makes its estimated $22.1 billion of margin loans approximately 17.5 percent of estimated total stock-market capitalization.

This percentage is a calculation from the study’s two reported estimates, not a separately reported historical statistic.

The contemporary GDP benchmark is $103.6 billion in 1929, according to the Bureau of Economic Analysis’s historical national accounts.

Thus, the $22.1 billion margin-loan estimate was approximately 21.3 percent of 1929 GDP.

That is an unusually large financial exposure relative to the economy, but it is important not to confuse margin debt with total stock-market capitalization or with the total amount of money invested in stocks.

3. The AI infrastructure boom of 2025–2026

The most useful evidence here comes from two institutions: the Bank for International Settlements (BIS) and the European Central Bank (ECB), together with Anthropic’s actual SEC registration statements.

This is preferable to relying on technology-company announcements or press estimates alone.

A. The debt financing is already measurable

The Federal Reserve’s November 2026 Financial Stability Report contains a section titled “Leverage in the Data Center Buildout.”

Its reported figures are:

MeasureAmount
Advanced-computing-related debt issuance, 2023$130 billion
Advanced-computing-related debt issuance, 2026 through OctoberMore than $680 billion
Outstanding public advanced-computing-related debt, October 2026More than $1.1 trillion
2026 share of nonfinancial investment-grade corporate debt issuanceApproximately 31%

The Federal Reserve reports that the composition of financing has evolved from primarily corporate bonds toward public and private debt, structured credit, and project finance.

It also identifies increasing lease commitments and backstop guarantees.

These are different measures: $680 billion is issuance during a period; $1.1 trillion is an estimate of outstanding public debt. They must not be added together.

Federal Reserve

The BIS independently documents the change in financing.

It reports that hyperscaler corporate bond issuance exceeded $100 billion in 2025, with additional infrastructure financing taking place through private credit and off-balance-sheet arrangements.

Bank for International Settlements

B. The hidden or off-balance-sheet commitments

The ECB’s November 2026 analysis is especially relevant because it examines the financing structure rather than just announced capital expenditure.

Its findings include:

  • Forecast cumulative capital expenditure by U.S. big technology companies of approximately $5.5 trillion through 2030.
  • Estimated annual AI investment needs exceeding 3 percent of U.S. GDP between 2025 and 2032.
  • More than $4 trillion in estimated off-balance-sheet commitments by big technology firms as of the second quarter of 2026.
  • The use of contractual commitments for computing capacity to support borrowing by data-center developers, chip owners, and other infrastructure providers.

The ECB explicitly distinguishes these contractual commitments from conventional balance-sheet debt. It describes some of them as debt-like in economic terms because they support borrowing against expected future cash flows, although they are not recorded as liabilities on hyperscalers’ balance sheets under U.S. accounting standards.

European Central Bank

This distinction matters enormously when comparing the present buildout with railroad bonds and 1929 margin loans.

A commitment to purchase future computing capacity is not automatically a loan. Nor is it automatically an unconditional debt obligation. The contractual terms, cancellation provisions, and financing arrangements matter.

C. The railroad comparison in GDP terms

The ECB provides an unusually useful historical comparison.

It reports that annual capital expenditure for development of the Northern Pacific Railway amounted to approximately 2.24 percent of U.S. GDP between 1870 and 1890.

It also reports that the internet investment boom of 1996–2003 represented approximately 1.10 percent of U.S. GDP.

Against those historical benchmarks, the ECB estimates that annual AI investment needs exceed 3 percent of U.S. GDP over 2025–2032.

European Central Bank

This is a particularly valuable comparison because it comes from one institution using a common GDP-based framework.

There is an important qualification: the Northern Pacific figure describes capital expenditure for one railroad’s development, while the AI figure covers a much wider investment ecosystem. The figures are not identical in scope.

Nevertheless, the reported percentages can be presented accurately without pretending that they measure exactly the same set of assets.

4. Specific comparison: railroad debt, 1929 margin debt, and AI debt

The following table separates the historical measures from the current ones.

Financial exposureOriginal reported amount2026-dollar equivalent or current amount
1873 railroad funded debt$1.837 billionApproximately $51.6 billion
1873 railroad capital stock$1.948 billionApproximately $54.7 billion
1929 estimated margin loans$22.1 billionApproximately $433.6 billion
1929 NYC broker-dealer borrowings$8.525 billionApproximately $167.2 billion
2026 advanced-computing debt issuanceNot applicableMore than $680 billion
2026 outstanding public advanced-computing debtNot applicableMore than $1.1 trillion

Historical-dollar conversions use the 2026 price index of 337.511 and the 1873 CPI of 12.000 or the 1929 CPI of 17.2. The 1873 conversion factor is approximately 28.13; the 1929 conversion factor is approximately 19.62. Values are rounded. The historical CPI series and 2026 index are from the cited inflation data.

A caution about this table: The 1873 railroad figures are system-wide funded debt and capital stock. The 1929 figures are estimates of securities-related lending. The 2026 figures are advanced-computing-related debt measures. The comparison is one of scale, not identical financial instruments.

5. RCA in 1929 and Anthropic in 2026: the technology-stock parallel

Radio Corporation of America (RCA)

Radio and broadcasting, 1920s

Anthropic

Artificial intelligence, 2026

This comparison requires one important correction: RCA was not a new 1929 IPO in the same sense as Anthropic’s proposed 2026 IPO.

A. RCA’s history and stock-market rise

RCA was established in 1919 to consolidate and license important radio patents.

Its stock was listed in 1924. It subsequently became one of the most prominent technology stocks of the 1920s.

The historical account in the Oxford University Press volume Financing New Industries describes:

  • RCA’s establishment in 1919.
  • Its public-market listing in 1924.
  • Syndicates purchasing RCA shares and attracting additional investors through rising prices.
  • More than a tenfold increase in RCA’s split-adjusted share price between 1925 and its 1929 peak.
  • A price-to-earnings multiple exceeding 70.
  • The fact that RCA did not pay a dividend until 1937.

The SEC provides another useful price comparison: RCA’s share price rose from $85 in 1928 to $549 in 1929.

SEC

The PBS documentary history of the crash describes RCA as the prominent technology stock of the decade and documents the involvement of a stock pool operated by Michael Meehan. It reports that RCA had fallen to $10 a share by 1931.

PBS

The Senate’s subsequent investigation of stock-exchange practices also documented RCA pool operations. It recorded a price increase in the newly issued radio stock from approximately $75–$80 to $109 during the March 1929 pool operation.

St. Louis Fed

The important historical point is that RCA combined:

  • A genuinely new and popular technology.
  • Rapid growth in the radio market.
  • A prominent public stock.
  • Speculative trading and coordinated stock-pool activity.
  • An extraordinary increase in its market valuation.

The company’s technology was real. The existence of a speculative stock market surrounding that technology is also documented.

B. Anthropic’s proposed IPO

Anthropic’s November 10, 2026, SEC filing is an actual primary source, not a press estimate.

The preliminary registration statement describes an offering of 825 million shares of Class A common stock. The prospectus identifies 801,351,354 shares as being sold by Anthropic and 23,648,646 shares by existing shareholders.

The filing is explicitly subject to completion. It should not be represented as a completed IPO.

Reuters reported on November 12 that the proposed offering could raise as much as $64.4 billion and that the company was seeking a valuation of up to $1.4 trillion. Those are reported offering terms and proposed valuation, not a completed market valuation.

C. Anthropic’s actual debt and contractual obligations

This is where the comparison with RCA becomes particularly interesting.

Anthropic’s own SEC filing reports:

Anthropic financial itemReported amount
Outstanding indebtedness, September 30, 2026$6.2 billion
Senior secured revolving credit facility$15 billion
Aggregate compute and infrastructure commitments$900.9 billion
Commitments expected before December 31, 2030$451.6 billion
Commitments due in the following 12 months$70.7 billion

The $6.2 billion is actual reported indebtedness, principally convertible promissory notes issued to Amazon through AWS.

The $15 billion revolving credit facility is a financing facility, not an additional $15 billion of outstanding debt.

The $900.9 billion is contractual purchase and lease commitments. It is not $900.9 billion of borrowed money.

The filing also states that Anthropic intends to finance these investments through revenue and capital it raises.

That distinction is crucial.

RCA’s speculative exposure was primarily expressed through its stock-market valuation and trading. Anthropic’s exposure includes a substantial amount of direct indebtedness and enormous future infrastructure commitments, in addition to its proposed equity offering.

There is no factual basis for calling the entire $900.9 billion a debt balance.

D. The comparison in context

CharacteristicRCA, 1929Anthropic, 2026
TechnologyRadio and broadcastingLarge language models and AI
Public-market statusListed in 1924Proposed IPO filed in 2026
Speculative interestDocumented stock-pool operationsProposed large public offering
Stock valuationMore than tenfold split-adjusted rise, 1925–1929Proposed valuation up to $1.4 trillion, reported by Reuters
Reported corporate debtSee RCA’s 1929 annual report$6.2 billion
Infrastructure exposureRadio manufacturing and broadcasting infrastructureCompute, chips, data centers, leases and purchase commitments
Dividend historyNo dividend until 1937Not a directly comparable historical measure

RCA’s 1929 annual report also provides an actual balance sheet. It reports $34.7 million in funded debt and other liabilities and $52.7 million in capital stock at year-end.

World Radio History

This is useful because it prevents a common historical error: assuming that RCA’s soaring stock price meant the company itself had borrowed an equivalent amount.

The market capitalization of a company’s stock, its outstanding debt, and its future contractual commitments are three different quantities.

6. Comparison by percentage of GDP

This is the most useful way to put the three episodes into economic perspective.

EpisodeFinancial or investment measureGDP comparison
Railroad expansion, 1870–1890Northern Pacific capital expenditure2.24% of GDP
Stock speculation, 1929Estimated $22.1 billion in margin loans21.3% of GDP
AI expansion, 2025–2032Estimated annual AI investment needsMore than 3% of GDP

These percentages measure different things:

  • The railroad figure is capital expenditure for one railway.
  • The 1929 figure is estimated outstanding margin credit.
  • The AI figure is annual investment requirements across the AI industry.

They should not be ranked as though they represent the same accounting category.

There is also an important distinction between the AI industry’s investment and its debt. The ECB’s estimate of more than 3 percent of GDP is an investment figure, not a debt figure.

What can and cannot be concluded

The historical sources establish that:

  1. Railroad expansion involved very substantial bonded indebtedness and equity capitalization.
  2. Stock-market speculation in 1929 involved substantial securities-backed borrowing, much of it funded outside the banking system.
  3. AI infrastructure investment is increasingly financed through debt, private credit, leases, and complex contractual arrangements.
  4. AI infrastructure financing now involves substantial off-balance-sheet commitments.
  5. The ECB’s estimated annual AI investment requirement exceeds the historical GDP share reported for Northern Pacific’s development.

The sources do not establish that the AI industry will suffer a crash equivalent to 1873 or 1929. Nor do they establish that the current market is necessarily a bubble.

Those would be forecasts or interpretations, not historical facts.

7. Inflation-adjustment methodology

For reproducibility, the conversions above use:

The historical indexes used are:

YearCPI index
187312.000
192917.2
October 2026337.511

Sources: the historical CPI series and the Bureau of Labor Statistics.

9. Final factual comparison

The most defensible conclusions from the sources are these:

Historical episodeDocumented financial characteristic
Railroad expansion and Panic of 1873Large-scale infrastructure investment financed through railroad stock and funded debt
Stock-market boom and Crash of 1929Extensive securities-backed borrowing, with estimated margin loans equivalent to approximately 21.3% of GDP
AI infrastructure boom, 2026More than $1.1 trillion in outstanding public advanced-computing debt, alongside substantial private borrowing and off-balance-sheet commitments

The RCA comparison adds another documented parallel: a revolutionary communications technology attracted enormous public enthusiasm, while its leading company’s stock became a vehicle for extraordinary speculation.

Anthropic’s proposed IPO is a modern example of a major technology company approaching the public market amid exceptionally large infrastructure financing requirements. But the company’s actual debt, proposed share offering, and future infrastructure commitments must be kept separate.

The strongest historical comparison is therefore not that the three episodes are identical. It is that all three involved major technological or infrastructure expansion accompanied by large-scale capital mobilization, while the forms of financial leverage differed substantially.

I have deliberately not supplied a 1873 railroad debt-to-GDP percentage or treated AI contractual commitments as borrowed debt, because doing so would require combining measures that the cited sources do not establish as equivalent.

Author's image

Paul E. Schindler Jr.

SEARCH PSACOT 2015-Present

Sign Up For Weekly Email Notification

* indicates required

Email Me

Email Me