The AI wobble is a valuation reset, not the end of the buildout
This week’s chip selloff looks like a repricing of crowded AI winners, not proof that the capex cycle has broken. The real debate now is which parts of the stack still have pricing power, backlog visibility, and earnings that justify the multiple.
The market is finally asking the right AI question, and it is not whether the buildout is real. It is which companies still deserve premium valuations now that the trade has become crowded and expectations are no longer rising in a straight line. An 11% weekly drop in the semiconductor index is sharp enough to trigger bubble talk, but the underlying operating data still points to an industry sorting winners from passengers. That is a valuation reset, not an obituary for AI infrastructure spending.
The cleanest way to read this selloff is that fundamentals and multiples are no longer moving together.
NVIDIA
still posted 65.5% revenue growth with a 63.0% net margin, and recent reported results showed revenue up 66% year over year with gross margin above 73%. That is not what a demand collapse looks like. The stock can wobble because the market no longer wants to pay any price for the obvious winner, but that is very different from saying the winner has stopped winning.
The same distinction matters across the stack. Broadcom is a good example of why the "AI is over" argument is too blunt: the company said AI semiconductor revenue reached $10.8 billion, up 143% year over year, and guided fiscal third-quarter revenue to about $29.4 billion, up 84%. Yet AVGO also trades at 46.76x earnings and 23.93x sales, which means even excellent execution can run into a harder valuation ceiling. When a stock with those multiples sells off on nuance rather than on broken demand, the message is not that the buildout is dead. The message is that the market is becoming less forgiving.
That is why we think the next phase is about selectivity, not surrender. Some names still have the combination investors should care about now:
NVDA: 35.26x P/E, 65.5% revenue growth, 63.0% net margin
AVGO: 46.76x P/E, AI semiconductor revenue up 143% year over year
AMD: 112.60x P/E, 34.3% revenue growth, data center revenue up 57% year over year
QCOM: 14.47x P/E, 13.7% revenue growth, EPS growth at -44.3%
That list makes the new argument obvious. AMD and Vertiv still have strong growth narratives, but they also carry the kind of multiples that invite punishment if execution slips even slightly. Qualcomm, by contrast, looks optically cheap at 14.47x earnings and 4.07x sales, but cheap is not the same thing as strategically advantaged. Its negative EPS growth and weaker position in AI infrastructure make it hard to argue that the market is missing a core buildout winner. In other words, this is no longer a simple growth trade; it is a conversion trade. Investors want to see that AI demand turns into durable margins, backlog visibility, and earnings power.
Yes, bulls can still point to the breadth of spending and say the cycle is broadening beyond one company. They are not wrong: AMD showing 57% data center revenue growth matters because it proves AI capex is flowing through more than one supplier. But the counterpoint is that broad demand does not protect every stock from de-rating. In crowded themes, the market eventually distinguishes between companies with real pricing power and companies that are merely adjacent to the spending wave.
That is also why the old bubble framing misses the structure of this moment. Bubble collapses usually come with broken orders, collapsing estimates, or margins rolling over fast. What we have instead is a group of companies still posting unusually strong growth while investors compress the multiple they are willing to pay. Reuters' reporting that NVIDIA was trading around 19x forward earnings earlier this week, its lowest level in more than a decade, is the tell. If the market were truly calling the end of the AI buildout, the debate would be about falling demand. Instead, it is about whether even elite businesses should still command elite premiums.
The better historical parallel is late-cycle infrastructure repricing, when the underlying technology trend keeps advancing but leadership narrows and valuation discipline returns. That is where names like Arista fit into the conversation. ANET has a 54.17x P/E, 28.6% revenue growth, and a 38.3% net margin, which is a strong operating profile by any standard. But once a theme becomes the market's most crowded trade, strong no longer guarantees upside if expectations were already near perfect. The stack is still being built; the market is just becoming choosier about who gets paid for it.
The bottom line is that this week did not kill the AI capex story. It forced a more serious conversation about where the economics are strongest and where valuation had drifted ahead of proof. We think the winners from here are the companies that can show three things at once: pricing power, visibility into demand, and earnings conversion that survives scrutiny when the multiple contracts.
What would change our mind? A real break in order commentary, a visible slowdown in data center growth across more than one major supplier, or margin erosion at the top of the stack would argue this is more than a reset. Until then, we see the selloff as the market moving from indiscriminate enthusiasm to discrimination. That is painful for momentum, but it is not the same as the end of the buildout.
Our take, not advice. This is opinion commentary — informational only, not personalized investment recommendations. Markets carry risk. Do your own research and consider your own situation before any trade.
▌The Daily Briefing · Free
A new stock idea, every evening.
One stock worth watching each weekday, plus the analysis behind it. Free, in your inbox.
