It was a stormy night for international markets.
The last few months of market narrative have centred on tech bubbles colliding with debt.
It’s like the dot-com-era debt machine is back, only this time the collateral is Nvidia chips. I recently covered that issue here.
The sell-off on Tuesday was the market turning on the debt-funded build layer of the AI boom.
The builders of the AI revolution carry the capital risk in this cycle, and Tuesday was their reminder.
Here’s what happened, why it ran so hard, and the one number that tells us whether it’s over.
What happened Tuesday
The Philadelphia Semiconductor Index fell 7.9%, and every one of its 30 members closed lower, led by Micron and Marvell.
Source: TradingView
[Click to open in a new window]
Both companies had been running hot, more than doubling this year, so they had the furthest to fall.
The Nasdaq and the S&P 500 both closed at more than one-week lows, and the iShares Semiconductor ETF dropped about 6%.
The selling began overnight in Asian and Korean chipmakers and thundered into the US session.
These markets have been on a tear; I’m not surprised the market is starting to wonder when the music stops.
What we saw Monday and Tuesday is investors shaking off a few weeks of hype and reassessing what they actually own.
Monday, the first domino fell
Tuesday did not come out of nowhere. Let’s look at the numbers leading up to it.
The selling started on Monday, the 22nd, when Alphabet fell roughly 10% on a $180 billion AI capex plan with no clear timeline for the cash to come back.
Palantir fell close to 9%, Amazon and Meta each lost about 4%, and SpaceX, not two weeks past its record IPO, dropped 16% after filing its first-ever bond offering.
There was a twist.
Memory chipmakers rose that same Monday, with SK Hynix, Micron and other memory names climbing ahead of Micron’s earnings this Wednesday — more on that later.
By Tuesday, the whole ship went down.
The AI trade is crowded; I have been vocal about it. And any wobble is an invitation to take profits.
So why go looking for a wobble in the future of computing? Debt.
The initial understanding was that this build-out would be funded from cash flow rather than borrowing; that is no longer the case.
Hyperscalers raised around $108 billion of debt in 2025, with projections of up to $1.5 trillion of new tech-sector debt in the coming years, as capex climbed to between 45–57% of revenue.
The real fear is that data centres being built today could be made obsolete by efficiency gains before they ever pay for themselves.
Nobody wants to be left holding the bag on those data centres if they go offline.
One weak link
Part of Tuesday was mechanical.
The leveraged semiconductor ETF, SOXL, fell nearly 20% before the market even opened.
Because it resets exposure daily, a falling market forces them to sell into the weakness, which drives more selling.
The chip businesses themselves did not change overnight.
We want proof
We will get a real signal soon.
Micron reports after the close on Wednesday the 24th, and its forward guidance on orders will tell us whether the AI spending cycle is still intact or starting to cool.
For perspective, the June 5 sell-off erased more than a trillion dollars in a single session, then recovered over the next two trading days.
The market judged it an overreaction rather than a change in fundamentals.
These dips have been bought back quickly so far.
The companies I have been writing about all month are the boring businesses that actually profit from AI; these sit a mile away from the chaos.
Next week, I’ll be digging further into these companies and why they succeed while the names that took a hit on Tuesday do not.
Until next time.
Regards,
Paul Dichiera,
Fat Tail Daily
Comments