The AI Bubble Debate – Consider These Patterns Across Two Centuries

The AI bubble debate misses a fundamental point: bubbles are endemic to technological revolutions, and that’s not necessarily a bad thing.

Consider the pattern across two centuries of infrastructure buildouts:

Canals (1790s-1830s): Canal mania saw massive capital inflows, most canal companies failed financially, yet transport costs fell 90% and enabled the industrial economy.

Telegraph (1840s-1860s): The “Victorian internet” saw speculative frenzy and overbuilding. Western Union survived and dominated briefly, but the real value accrued to businesses that could now coordinate across distances: finance, commodities trading, and news agencies.

Railroads (1840s-1870s): Railway stocks surged 5x then collapsed 70%. Most railroad investors were wiped out by 1900, but the rails got built and the value accrued to shippers (Standard Oil, meatpackers, retailers) rather than the railroads themselves. Canal workers and coach operators were displaced; new logistics and manufacturing jobs emerged.

Automobiles (1900s-1920s): Over 3,000 auto manufacturers entered the U.S. market. Fewer than 50 remained by 1930. Investors in 95%+ of early car companies lost everything, yet automobiles reshaped cities, retail, logistics, and labor markets. The derivative winners (oil companies, suburbs, highway contractors) captured enormous value. Entire industries around horses (farriers, stable hands, carriage makers) disappeared; assembly line workers, mechanics, and truckers took their place.

Electrification (1920s): Electric holding company stocks tripled, annualized returns exceeded 50% between 1925-1929, capital poured into infrastructure at breakneck speed. The bubble burst, but electricity transformed every sector of the economy. The value diffused downstream rather than concentrating in early winners. Manual laborers and domestic workers were displaced by machines and appliances; new roles in manufacturing, maintenance, and consumer industries emerged.

Aviation (1920s-1930s): Massive investor enthusiasm, dozens of airlines launched, most failed. The industry has been a notorious value destroyer for shareholders (Buffett’s famous quip about shooting Orville Wright). Yet aviation transformed global commerce, tourism, and logistics. The value went to Boeing, hotels, tourism economies, and global supply chains rather than the airlines themselves.

Semiconductors (1960s-1980s): Dozens of chipmakers, brutal competition, most early players disappeared or got acquired. The value diffused to the companies that used chips (PC makers, then software, then internet companies) rather than concentrating in chip fabrication.

Fiber optics (1990s): $2 trillion+ invested during the dot-com era. WorldCom, Global Crossing, and dozens of others went bankrupt. But the infrastructure enabled YouTube, Netflix, and cloud computing a decade later. Traditional media, retail, and travel agents were disintermediated; software engineers, content creators, and e-commerce operators filled the gap.

Dot-com (1995-2000): Pets.com, Webvan, and hundreds of e-commerce plays collapsed. Amazon lost 90% of its value. But the thesis was right, just early. E-commerce went from 1% of retail in 2000 to 15%+ today. The value accrued to survivors and second-movers (Google, Facebook, Netflix) rather than first-wave pioneers.

Every technological revolution follows this pattern: capital floods in, investors get burned, infrastructure remains, labor gets displaced and redeployed, and value diffuses downstream. The difference with AI is the displacement target: previous revolutions primarily affected manual and blue-collar work. AI aims squarely at knowledge work and white-collar jobs. That’s not a reason to think the pattern breaks, just a reason to note who’s nervous this time.

We’re seeing the familiar infrastructure phase now. “Compute” is the bottleneck attracting capital. The hyperscalers (Amazon, Alphabet, Microsoft, Meta, Oracle) have effectively abandoned their asset-light business models in favor of massive infrastructure buildouts. Since late 2023, physical assets at these five companies have increased 50-200%. Cash return on equity has declined 600-1300 bps as a result.

Here’s the valuation question no one’s asking: if you value the new PP&E at cost and assign a 10x P/B multiple to the rest, these market caps would be roughly half of current levels. Previously “virtual” companies now need to worry about capacity utilization and depreciation rates. That’s a meaningful shift.

AI has also given cover to long-overdue data infrastructure investments. Some of this spending isn’t really about LLMs at all, just initiatives that could have been greenlit years ago but lacked executive urgency.

The key insight: a technology doesn’t need to fizzle for its bubble to deflate. If AI proves as essential as electricity or the internet, its economic gains will likely be just as diffuse and derivative. First-mover advantages may not hold.