The Auto Sector: A Hundred Years Before the AI Buildout

Two hundred and fifty-three companies were building automobiles in America in 1908.

By 1929, only forty-four were left, and three of those, Ford, GM, and Chrysler, controlled the industry.

That collapse is worth examining now, not because the AI buildout will repeat it exactly, but because it's the closest precedent available for understanding which companies are actually built to survive what's coming.

Capital markets commentator Barry Ritholtz argued on a recent episode of the Prof G Markets podcast that the AI boom is closer to the Industrial Revolution than to the dot-com boom and bust most people reach for.

Most people who've picked up that same Industrial Revolution framing have sharpened it into a comparison with the railroad boom specifically. Niall Ferguson has compared the AI data-center buildout to the 1870s-1890s railway boom and worries about a repeat of the Panic of 1893.

Liaquat Ahamed, who won a Pulitzer for his history of the 1929 crash, just published a book on the 1873 railroad bust and says the AI parallel keeps him up at night. The White House's own Council of Economic Advisers made the same railroad comparison in a January paper.

This memo takes Ritholtz's Industrial Revolution framing and applies it differently, into the 1920s global automobile sector instead of the railroads. The 1920s auto industry represents a useful environment and one worth examining in its own right.

The forty-four companies that survived to 1929 were the lucky or well-capitalized few. Everyone else is what the trade press of the era called the independents, and almost all of them were gone within two decades.

It wasn't falling demand that killed them. Car sales grew through nearly the entire 1920s.

What killed the independents was the sheer scale of capital mass production required, a bar most of them couldn't clear once the 1920-21 recession and then the Depression arrived and credit tightened.

Ford won the first phase of the shakeout outright, on production cost. The Model T was a single design, built on dedicated tooling.

That's exactly why Ford had to shut down for six months in 1927 to retool the entire operation for the Model A.

GM's manufacturing chief, William Knudsen, built something deliberately different at Chevrolet. He designed a flexible production system that could absorb model changes without a shutdown.

Paired with Alfred Sloan's annual model changes and GM's in-house financing arm, that flexibility is what let GM overtake Ford rather than just survive next to it.

Winning the capital-intensity phase didn't make Ford's lead permanent. Building for flexibility, not just for scale, decided who won the next one.

Both, depending on which layer of the AI stack you're looking at. The hyperscalers, the handful of companies building cloud infrastructure at global scale, chiefly Alphabet, Microsoft, Amazon, and Meta, are funding this buildout through real capital markets against their own balance sheets.

Alphabet alone just closed the largest equity offering in corporate history, $85 billion, including a $10 billion stake from Berkshire Hathaway, on top of more than $55 billion in fresh bond debt raised since November. That layer looks like real capital formation.

The layer sitting on top of it looks more like Cisco in 2000 and 2001, when the networking giant financed its own customers' purchases and watched its stock collapse by roughly 90% when those customers couldn't pay.

Nvidia, OpenAI, Oracle, AMD, and CoreWeave have built a documented web of circular deals, investment tied to purchase commitments, that analysts now size above $800 billion.

The “this isn’t 1999” defense holds up well for the hyperscaler balance sheets. It holds up much less well for the chip-and-lab layer immediately above them.

By 1930, Ford and GM were outproducing Japan's entire domestic auto industry by a factor of roughly 43 to 1. Tokyo didn't respond the way London did, with a tariff that let an inefficient domestic industry survive at a permanently smaller scale.

In 1936, Japan passed a law forcing Ford and GM to stop production in the country entirely, and designated Toyota and Nissan as the licensed replacements.

That's the closer fit for China today. US export controls block Nvidia's most advanced chips from Chinese buyers, and Beijing has answered with concentrated state backing of Huawei, SMIC, and its domestic cloud players rather than trying to outspend Washington.

China's total AI infrastructure investment in 2025, an estimated $125 billion against a US figure roughly six times larger, isn't evidence of falling behind on capital. It's evidence of building a designated replacement behind a wall, the way Tokyo built Toyota.

The US-Japan-China story is the sharpest one, but it isn't the whole picture, and the other outcomes map onto today's AI buildout just as cleanly.

Britain's auto industry survived behind a 33% tariff that lasted until 1956, never matching Detroit's scale, never disappearing either. The EU's AI Act and data-sovereignty rules amount to the same trade.

They preserve a degree of independence behind a regulatory wall, without producing a domestic champion behind it.

Germany's eighty-six manufacturers fell to twelve survivors in the Depression. Daimler and Benz had already merged in 1926 to survive, not to dominate.

That's the more likely fate for today's mid-tier AI labs that can't raise capital at hyperscaler scale, acquisition or merger, not collapse.

Canada's 35% tariff forced Ford and GM to build Canadian plants, and Imperial Preference let cars built in Oshawa reach British Commonwealth markets at a lower tariff than cars built in Detroit.

Canada became the world's second-largest auto producer by 1929 with zero domestic ownership of the industry, purely because hosting Detroit's production was a passport into a market Detroit couldn't reach as cheaply on its own.

The Gulf states' compute deals with US hyperscalers, the UAE's arrangement with Microsoft among them, read the same way: access and energy in exchange for hosting infrastructure that isn't theirs.

Argentina, Brazil, and Mexico never developed an independent auto industry at all. Ford and GM ran assembly operations there from the 1910s and 1920s onward, and no domestic manufacturer ever emerged to consolidate.

Most of the Global South's relationship to AI infrastructure today looks like that: consumption of hyperscaler capacity with no domestic model or chip industry underneath it.

The auto shakeout took two decades and needed a recession plus a Depression to hit an already overbuilt field. And the GM-versus-Ford split suggests the physical assets themselves mattered less than how flexibly they were built.

Ford's single-model tooling was close to worthless once that model died. GM's flexible lines weren't.

A modern GPU or data center looks structurally closer to GM's flexible plants than to Ford's. It can run different model generations, get leased, or get resold.

If the AI buildout corrects the way auto sector did, the actual capital destruction may have less to do with whether the spending was real, and more to do with whether what got built was flexible enough to survive the correction.

That's the open question this memo leaves on the table, not a settled one.

Barry Ritholtz, Prof G Markets with Scott Galloway and Ed Elson, June 19, 2026.

Niall Ferguson and Liaquat Ahamed's railroad-boom comparisons per Fortune and the Heritage Foundation. White House Council of Economic Advisers, "The Revolution of Artificial Intelligence," April 2026.

Germany and Russia's state-directed railroad histories, and Argentina's foreign-capital-financed railway competition, per standard economic histories of 19th-century rail expansion and Colin Lewis's research on British railway investment in Argentina.

Alphabet equity and debt financing per SEC filings, CNBC, and Alphabet's June 2026 investor presentation. Circular AI financing analysis per Bloomberg and Noah Smith.

GM/Ford manufacturing history per the Harvard Business School case "Ford vs. GM: The Evolution of Mass Production" and ASSEMBLY magazine's GM centennial coverage. Japan's 1936 Automobile Manufacturing Industries Act per Toyota's corporate history archive and Project MUSE.

If you have a perspective or disagreement, reply directly. I read every response.