It’s Do-or-Die Time for the AI Infrastructure Trade
After the strongest earnings of the entire cycle, semiconductor stocks sold off anyway. Here is what the tape is telling you.
Listen to the audio version of this article (generated by AI).
Garry Kasparov, the greatest chess player to ever live, found his match not in a flesh-and-blood opponent, but in a machine.
Kasparov built his career on reading intentions accurately. But that’s a much-taller ask when the intentions you’re reading are completely sterile.
Under the scrutiny of a mass audience and blinding, hot television lights in a Manhattan office tower, the machine slid its rook into a corner square. Kasparov couldn’t read the move. The machine, Kasparov thought, was seeing too far down the board for him to keep up.
As a result, Kasparov was defeated.
Years later, one of Deep Blue’s designers explained that the machine wasn’t actually seeing into the future. Rather, it simply hit a bug, ran out of search time, and fell back on a random legal move. There was no genius internal plan. Only noise which Kasparov, the greatest human chess player, mistook for signal.
Wall Street is in Kasparov’s shoes right now.
The AI complex just delivered the cleanest fundamental results of this entire cycle. Alphabet Inc. (GOOGL) grew its cloud business 82% year-over-year and hiked capital spending guidance. Tesla Inc. (TSLA) reaffirmed its own spending and promised more for three straight years. And semiconductor stocks sit in a 25% bear market.
The move that spooked everyone? The 10-year Treasury yield ticked up about 20 basis points.
In other words, it is the proverbial rook in the corner.
Let me show you why the intention that Wall Street sees behind the move is noise, but also, where the bear case should earn your respect:
What the Big Spenders Said
Alphabet smashed it.
Cloud revenue up 82%. Record operating margins. Search came in a touch light, and I do not care about that at all.
What matters is the capital spending guide, which climbed to roughly $185 billion to $200 billion at the midpoints, with 2027 rising significantly from there.
Tesla’s quarter was messy, but the read-through was clean anyway.
The picture repeats outward with Taiwan Semiconductor Manufacturing Co. (TSM), which raised its 2026 capital budget from $54 billion to $62 billion and lifted its revenue growth target from more than 30% to more than 40%.
ASML Holding N.V. (ASML), for its part, raised guidance for a second consecutive quarter. Meanwhile, Morgan Stanley took its hyperscaler spending forecasts up roughly 10%, to $1.2 trillion in 2027, and OpenAI raised its compute budget through 2030 to $750 billion.
Six data points showing moves in one direction: up and to the right.
Twenty Basis Points Does Not Move Mark Zuckerberg
So why did the stocks get hit?
Because Wall Street stopped worrying about demand and started worrying about durability.
Alphabet’s free cash flow (FCF) went negative this quarter. So did Tesla’s… and Amazon, Microsoft, and Meta will probably follow. A buildout financed out of corporate cash now has to lean on debt, and debt is interest-rate sensitive. So the morning after those results, when the 10-year yield spiked to 4.71%, a trade that previously carried zero rate sensitivity was suddenly burdened with it.
I understand the fear. The math behind it, however, collapses on contact.
A 4.7% 10-year versus a 4.5% 10-year changes nothing about what Meta spends over the next three years. Nothing for Satya Nadella at Microsoft. Nothing for Elon Musk at Tesla or SpaceX. These men do not get knocked off their horses by 20 basis points, and they probably do not flinch at 100. Because they believe the AI race is existential, and I think they are right.
Alphabet is scared to death that a year of restraint hands ChatGPT the search business, hands Amazon Web Services and Microsoft Azure the cloud business, and leaves nothing standing. You do not slow a race like that with a bond yield. You slow it with a recession, 20% unemployment, and buildings on fire.
Rate sensitivity does exist here, though. Call it 2%, concentrated in the leveraged end of the neocloud complex. The other 98% goes through without a hiccup.
Valuations: Why the Math Doesn’t Add Up
The S&P 500 trades at about 21 times forward earnings against a decade average near 22 times, and the Nasdaq-100 sits around 24 times against an average near 26 times. The Philadelphia Semiconductor Index (SOX) trades near 25 times forward while semiconductor earnings per share (EPS) are expected to grow more than 100% this year and another 42% next year. Paying 25 times for compounded growth in the 40s is a bargain wearing a warning label.
The sharper argument attacks the estimates instead of the multiples. Fine. Those estimates are powered by AI capital spending, and the spenders just guided higher.
Furthermore, the “where are the profits” complaint died on Alphabet’s earnings call, as Google Cloud grew 25% to 35% in its normalized post-pandemic years. And that growth was decelerating. It just printed 82% on a business annualizing around $100 billion, carrying a backlog of more than $500 billion.
Fifty points of acceleration at that scale is about as cut-and-dried as evidence gets.
The Bear Case I Fear
Two words: the tape.
SOX is down 25% from its highs. Since the global financial crisis, semiconductor bear markets come in three flavors: the ones that bottom in the 20% to 25% drawdown zone, like late 2018 and summer 2024; the ones that grind into sloppy prolonged declines, like 2022; and the ones driven by black swans, like COVID-19 and Liberation Day. No black swan is attached here, so history says we bottom near here or this gets ugly and slow. The dot-com years argue for caution, too: between 1995 and 2000, SOX entered three bear markets and never once stopped at 20%.
Here is the line I am watching. SMH, SOXX, and SOX have all fallen to their 100-day moving averages. Every AI Boom sell-off since late 2022, excluding Liberation Day, bottomed at or just below that line.
One thing tilts me constructive. When the dot-com boom actually ended, the crash in infrastructure names dragged the whole market with it: by early 2000, SOX was down about 25% and the S&P 500 was down 10% alongside it. Today the S&P 500 is down 2%. The hot trade is getting hit while the rest of the market broadens underneath it. That looks like late 1995, and it reads as a recalibration.
The Stocks I’m Buying
The setups I am watching include Broadcom Inc. (AVGO), Intel Corp. (INTC), and Advanced Micro Devices Inc. (AMD). Down the market-cap chain, Amkor Technology Inc. (AMKR) owns real leverage to advanced packaging and Alphabet’s tensor processing unit ramp, though a break below its 200-day moving average argues for patience. Comfort Systems USA Inc. (FIX) and GE Vernova Inc. (GEV) attack the power bottleneck, and memory looks compelling, led by Micron Technology Inc. (MU).
Earnings and stock prices go hand in hand, and that relationship has held as long as markets have existed. It breaks down every so often, and it always corrects. Right now, earnings trends point higher and price trends point lower. That resolves one of two ways: prices correct sharply higher to meet earnings, or earnings correct sharply lower to meet prices. After what Alphabet, Tesla, TSMC, and ASML just told us, I see no path to the second one, barring a black swan.
When in doubt, follow the earnings. And the earnings are still going (say it with me) up and to the right.
Most notably, while chip stocks were sitting in a 25% bear market this week, Apple Inc. (AAPL) touched $5 trillion in market value, becoming only the second company in history to reach that milestone.
What is Apple doing differently?
Apple is not out-spending Alphabet, Microsoft, and Amazon on data centers. Its capital budget is a fraction of theirs. What Apple is doing instead is funneling money into a different set of assets entirely, the same assets I am walking through live this Thursday.
Apple has made 124 recorded acquisitions, some as large as $3 billion at a time. Apple has spent more than $206 billion on mergers and acquisitions in total. And this playbook is not new. It is the playbook that built the company. Back in 1996, a struggling Apple paid $429 million for a small, unprofitable outfit called NeXT. That acquisition brought Steve Jobs back to Cupertino and, with him, the operating system that eventually became the foundation for the iPhone, the iPad, and the Mac as you know them today. Buy small. Buy early. Buy before the world understands why.
Apple is looking to invest heavily again this year, and I think that decision tells you more about where AI money is actually made than any capex chart out of a hyperscaler earnings call.
This Thursday, July 30, at 1 p.m. Eastern, I am walking through the exact framework Apple runs on, the same one I use to find AI opportunities before the big checks arrive. I call it PPT – People, Product, and Timing. And instead of talking about it in the abstract, I am applying it live, in real time, to one specific AI company I am recommending for free.
You’ll see how this framework works on a real deal. You will also see the People question, the Product question, and the Timing question answered against real facts, with a real opportunity to act the same day, if you decide it is right for you.
The setup here is not complicated. The company with the most cash on Earth just spent decades and $206 billion proving where the real AI money gets made: in the deals that happen before the rest of Wall Street notices.
I want to show you how to get a seat at that table.