Last week felt like someone quietly turned down the volume on the greatest show in the markets.
You had the biggest company on earth dropping numbers that would have sent any normal stock limit-up, and instead we got a collective shrug followed by a multi-day slide. Nvidia beat on every line, raised guidance again, and the CEO basically declared we’re standing at the dawn of a new industrial revolution. Normally that’s rocket fuel. This time? The rocket wobbled.
I’ve been around long enough to know that when the crowd stops cheering the obvious winner, something underneath has shifted. So let’s dig in and figure out whether the recent AI trade pullback is the kind of healthy reset that smart investors use to load up—or the first visible crack in a narrative that got way ahead of itself.
The Moment Everything Changed (Even If It Looked Normal)
On the surface, nothing looked catastrophic. The S&P 500 dipped below its 50-day moving average, bounced hard off the 100-day on options-expiration Friday, and closed the week basically flat to down half a percent. Normal stuff in bull markets.
But zoom into the engine room of this entire move—the handful of AI infrastructure names—and the picture gets uglier. The usual suspects dropped five, seven, ten percent in a matter of days. Bitcoin got smoked for good measure. Breadth collapsed. Money-flow indicators flipped from aggressive buying to distribution.
And the real tell? Even Nvidia, the undisputed king, couldn’t hold more than a few percentage points post-earnings. In every prior quarter that print would have added another ten percent in a straight line. This time the market whispered “yeah, we knew that already” and moved on.
That whisper matters.
What the Credit Markets Are Screaming That Stocks Are Ignoring
While equity traders argue about forward P/E ratios, the bond and derivatives crowd has been quietly hitting the panic button.
Credit-default-swap spreads on some of the biggest AI borrowers have absolutely exploded higher. We’re talking moves from comfortable sub-50 basis points to triple digits in a matter of months. One infrastructure player saw its five-year protection cost jump to levels implying double-digit annual default probability under conservative recovery assumptions.
Now before anyone screams “default apocalypse,” relax—these are still low absolute probabilities. But the direction and speed tell a story. The market is suddenly demanding a much higher premium to hold debt from companies that are levering up at a historic pace to build the AI future.
“For the first time in twenty years, a majority of global fund managers say corporations are over-investing.”
Bank of America Global Fund Manager Survey
That quote stopped me cold when I read it. Over-investment isn’t just an academic concern—it’s how empires die. Japan in the late 80s. U.S. telecom in 1999-2000. When everyone can draw a straight line from today’s capex to tomorrow’s cash-flow bonanza, you’re usually at the exact moment you should start worrying.
The Math Nobody Wants to Do
Let’s run some quick numbers that rarely make it to CNBC.
Collectively, the big tech platforms have already committed north of a trillion dollars in AI-related capital expenditure over the next several years. Some analysts think the final tally could approach three trillion when you roll up every hyperscaler, cloud provider, and data-center REIT.
That’s not pocket change. That’s roughly the GDP of India being plowed into server farms and custom silicon.
And the payback period? Most internal models assume meaningful cash returns start showing up around 2027-2028 and then scale dramatically from there. In the meantime, interest expense is no longer zero. Debt issuance is measured in tens of billions per company. Refinancing risk is real for the first time in a decade.
Put another way: the AI trade has moved from “growth at any price” to “growth at this specific price with this exact timeline.” That’s an entirely different game.
Why Nvidia’s Print Was Actually Bearish (Yes, Really)
I’m going to say something that will annoy the perma-bulls: Nvidia’s quarter was a sell-the-news event disguised as a beat.
Don’t get me wrong—the numbers were ridiculous in a good way. Data-center revenue up 41% quarter-over-quarter. Supply still sold out for a year. Gross margins staying north of 70%. All fantastic.
But the market isn’t paying for what you did last quarter. It’s paying for what you’re going to do in 2027. And baked into current prices is something close to flawless execution across the entire supply chain for years.
When the best possible print barely moves the stock, it tells you expectations were even higher. That’s not strength. That’s the sound of a bar being raised so high that even superhero results feel merely mortal.
The Santa Rally Setup Is Still Alive—But On Life Support
Seasonally, December remains one of the strongest months of the year. Buybacks are running at all-time highs. Professional managers are still underweight tech after chasing value and small-caps all summer. Liquidity conditions are decent.
All the ingredients for a year-end melt-up are sitting on the counter.
The problem? The oven might be broken.
- Breadth is terrible—fewer than 40% of Nasdaq stocks above their 50-day moving average
- Volatility measures are creeping higher
- Bond yields are no longer falling reliably on weak data
- Credit spreads are widening for the exact names driving the rally
In past years, any one of those would have been noise. Today they feel like cracks in the foundation.
So… Buy the Dip or Respect the Warning?
Here’s where I land after turning this over for days.
The long-term AI story is not broken. The productivity gains are real. The infrastructure buildout is happening. Companies that execute will create enormous wealth over the next decade. If you’re investing fresh capital with a five-plus year horizon and you truly believe we’re in the early innings of a technological revolution, then yes—current levels will probably look cheap in 2030.
But—and this is a big but—the margin of safety has evaporated.
.margin of safety has evaporated.We are no longer being paid to take risk. We are being paid to be right about timing, execution, monetization speed, competitive dynamics, energy constraints, regulatory backlash, and about a dozen other variables that used to be footnotes.
In my experience, when the crowd moves from “this is obviously going up forever” to “this will be huge if everything works perfectly,” the easy money has already been made.
What I’m Actually Doing With My Own Money
Full disclosure: I own some of the core AI infrastructure names. I added aggressively below the February 2024 lows and rode the entire move up. I have trimmed exactly zero shares on this pullback so far.
But I’m also not adding here. Cash levels are at multi-year highs for me personally. If we get a washout toward the 200-day moving average on the Nasdaq—call it another 8-10% lower—I’ll start nibbling again. Until then, I’m content letting the market figure out whether this is 2018-style late-cycle wobble or 2022-style valuation reset.
Sometimes the most aggressive thing you can do is nothing.
The Bottom Line
The AI trade isn’t dead. But the “set it and forget it” phase probably is.
We’ve moved into the chapter where discipline beats conviction and risk management beats narrative. The companies that convert today’s billions in capex into tomorrow’s hundreds of billions in free cash flow will mint generational wealth. Many others will become cautionary tales with football-field balance sheets and empty data centers.
Your job isn’t to predict which is which. Your job is to position yourself so that you’re still in the game when the answer becomes obvious.
For now, that probably means smaller sizing, higher cash balances, and a willingness to look stupid for a little while if the melt-up comes anyway.
Because in this business, surviving the periods when everyone sounds smart is how you get to compound when they finally sound scared again.
Stay nimble out there. The next move will tell us a lot more than the last one did.
