Can You Survive Being Right Too Early?

In late July, the market decided the AI build-out had gone too far. The Nasdaq fell into its second correction of the year, the Philadelphia Semiconductor Index dropped into a technical bear market, and chip stocks shed more than $1 trillion in market value. One week later, the S&P 500 closed at a record and the Dow cleared 54,000 for the first time in its history.

Nothing was resolved in between. Nobody proved the spending will earn its return, and nobody proved it won't. This episode is about why the crash of 1929 does not, on its own, explain the Great Depression — what actually did, where the modern equivalent is hiding, and why the number that decides your outcome was never the forecast.

In this episode:

The forecaster who called 1929 — and then told everyone to buyOn September 5, 1929, Roger Babson warned that "sooner or later a crash is coming, and it may be terrific." Seven weeks later he was proven right. What gets left out of the story: by most accounts he'd been saying a version of it for a couple of years, and after the market rallied in the spring of 1930 he changed his call and urged investors to buy. The Dow was back at 294 that April. Two years later, to the day, it closed at 63.

Why the crash wasn't the DepressionThe Federal Reserve's own history says the damage from the 1929 crash faded within a few months — by the autumn of 1930, recovery appeared imminent. Then more than 9,000 banks suspended operations between 1930 and 1933, around 30% of every bank that existed at the end of 1929. The Dow bottomed 89% below its peak and didn't recover until November 1954. A crash is a fire in one room. What connects the rooms decides whether the building burns.

The funding quietly changed underneath the boomJ.P. Morgan Asset Management puts hyperscaler AI capex at 93% of cash flow from operations this year, up from 33% in 2023. Microsoft's $15 billion "capex cut" was a lease-classification change, not a cut. The BIS counts more than $200 billion of private-credit loans to AI companies, Morgan Stanley sees a $1.5 trillion financing gap through 2028, and AI data center and power issuance made up 45% of net US high-yield issuance this year.

The smoke detector is beeping; the building is not on fireApollo's Torsten Slok tracks the cover ratio on hyperscaler bond deals: nearly 5x in February, below 2x by July. Against that, the Chicago Fed puts the average bank's exposure to AI-adjacent industries at 0.8% of assets. The risk hasn't vanished — it has moved into private credit, separate vehicles and lease structures, where it is harder to see.

The bull case, stated fairlyAzure grew 43% last quarter. Google Cloud grew 82%. Nvidia's data center revenue grew 92%, to $75.2 billion. The revenue is real. So was Amazon's — it grew more than 140% between 1999 and 2002, while the stock lost about 95% of its value.

The leverage in your own accountFINRA reported margin debt at a record $1.53 trillion in June, up 51.5% in twelve months. Amazon fell roughly 95% from its 1999 high and took a decade to get back. Own it outright and you kept the only thing that mattered: the option to wait. Own it with borrowed money and your broker closed the position before your thesis had time to be proved right.

What it means for youYou don't get to choose whether it burns. You get to choose, in advance, how far it can travel into your life. That comes down to three questions, none of which require a view on AI: if this position fell by half and stayed there for five years, does anything in your life actually break? Is any of it borrowed — including the borrowing you don't file under borrowing? And when do you genuinely need the money?

New episodes weekly. Subscribe on Apple Podcasts, Spotify, YouTube, or Amazon Music.

Conviction Bet is independent investment commentary. Nothing in this episode is investment advice.