Tea Leaves #5: AI TRACKS TO NOWHERE

Originally published July 15, 2026

This week’s Economist pithily summarized recent AI fundraising: “Nowadays it is easy to get desensitised to big numbers.

“Every week stonking deals are announced.”

This isn’t surprising: Extraordinary productivity boosts can trigger “stonking” deals.

But history suggests those deals don’t initially fund the most productive investments.

For proof, consider the British railway mania of the 1840s. In many ways, railways were 19th century data centers – catalysts for generational productivity growth.

As Edward Chancellor writes, at the mania’s apex, rail investments were “projected [to exceed] the country’s national income,” funding tracks “twenty times the [country’s] length.”

Tracks that led “to faraway places that could never repay the cost of investment.” Tracks to nowhere.

Like rail construction that preceded it, AI projects do not want for capital. But capital wants for direction.

Listen to executive interviews and speeches on the subject, and you’ll hear plenty about “falling behind.” Plenty about “catching up.” You’ll hear less about expected returns.

Just the urgent need to “adapt.”

A strategy grounded in fear may minimize executive regret, but it doesn’t maximize the good. It trades the discomfort of thoughtful decisionmaking for the comfort of the herd.

In The Price of Time, Chancellor illustrates this point with the story of Watchet. While the small British parish wasn’t exactly “nowhere” in the 19th century, it certainly wasn’t a metropolis that required three independent railways – precisely what was proposed in the manic wave.

Hidden in a footnote, Chancellor provides an update on Watchet: While “[none] of [the three proposals] got off the ground,” a singular passenger line was finally opened later.

And that line “continues in operation today.”

With the help of AI, businesses will build profitable Watchet railways. But continued pursuit of this technology for its own sake all but guarantees we’ll lay more unprofitable tracks first.

Tracks to nowhere.