The AI trade is not breaking — it is getting brutally more selective
The June chip washout did not break the AI trade; it exposed how little room is left for companies that cannot prove near-term monetization. After Broadcom’s near-miss and the $1 trillion-plus semiconductor drawdown, the market is rewarding supply control, platform leverage, and visible earnings power — not generic AI exposure.
The clean read on this month’s chip selloff is not that AI demand cracked. It is that the market finally started pricing AI stocks like a hierarchy instead of a theme basket. Broadcom’s June revenue miss was tiny in dollar terms, yet the reaction was violent because expectations had drifted to perfection even as AI semiconductor revenue still surged 143% year over year to $10.8 billion. That is the tell: the bar is no longer simply “has AI exposure,” but “can convert AI demand into durable revenue, margins, and control over the bottlenecks that matter.”
The June 5 drawdown — more than $1 trillion erased across U.S.-listed chip names, with the semiconductor index down about 10.26% in the session — looked dramatic enough to invite bubble talk. But the rebound that followed matters more than the selloff itself. Leadership snapped back first in the franchises with the clearest claim on AI economics, not in every semiconductor stock that can mention accelerators on an earnings call. That is not a broken trade. That is a market getting stricter about who deserves to keep premium multiples.
Start with NVDA, because it remains the easiest test of whether AI demand is real or rhetorical. Nvidia just posted revenue of $81.6 billion, up 85% year over year, with data center revenue of $75.2 billion, up 92%. In the comparative valuation set, it trades at 32.26x earnings with 65.5% revenue growth and a 63.0% net margin. That is expensive in absolute terms, but it is not indiscriminate exuberance when the company is monetizing AI at industrial scale and keeping extraordinary profitability. If anything, Nvidia is the case for why selectivity is increasing: the market is still willing to pay for visible demand, dominant platforms, and cash generation.
The same logic helps explain why AVGO was punished even though its AI business is still growing fast. Broadcom’s quarter was not bad; it was merely not perfect enough for a stock carrying a 68.22x P/E and 25.93x sales multiple. A revenue print of $22.19 billion versus a $22.27 billion consensus is not a thesis-breaker, especially when AI semiconductor revenue hit $10.8 billion and management pointed to supply visibility into 2026 and 2027. But that is exactly the point. Once a stock is priced for flawless execution, even strong AI fundamentals can be overshadowed by a tiny miss or an unchanged long-term target. The market is no longer paying the same premium for “AI growth” as for “AI growth with no ambiguity.”
That makes the next tier of winners easier to identify. TSM and ASML are not just AI beneficiaries; they are infrastructure toll collectors with direct leverage to the bottlenecks. TSMC expects AI accelerator wafer demand to rise 11-fold from 2022 to 2026, while advanced packaging capacity is projected to grow at more than an 80% CAGR from 2022 to 2027. ASML, meanwhile, is talking in hard capacity terms, with 60 low-NA EUV tool shipments in 2026 and 80 in 2027. Those are not story-stock metrics. They are evidence that the AI buildout is still happening in physical, constrained, monetizable ways.
The public-market numbers reinforce that distinction:
NVDA: 32.26x P/E, 65.5% revenue growth, 63.0% net margin
TSM: 39.74x P/E, 33.0% revenue growth, 47.0% net margin
ASML: 64.45x P/E, 15.6% revenue growth, 29.7% net margin
AMD: 179.72x P/E, 34.3% revenue growth, 13.4% net margin
MU: 53.46x P/E, 48.9% revenue growth, 41.5% net margin
The outliers tell the real story. AMD at 179.72x earnings is what a selective market should challenge, not because the company lacks AI relevance, but because the valuation leaves little room for anything short of sustained share gains and cleaner monetization. Earlier market commentary already highlighted how AMD’s forward multiple ran well above Nvidia’s despite Nvidia’s much larger AI footprint. That spread was always hard to defend. By contrast, MU shows why the market can still rerate a name aggressively when scarcity and earnings inflection are tangible. Micron’s memory exposure gives it a direct line into AI server demand, and its 48.9% revenue growth with a 41.5% net margin makes the rerating easier to justify than a generic “AI adjacency” story.
Yes, the bulls have a fair point that some of the June damage was macro, not thematic. Higher yields and a stronger jobs print can compress long-duration growth multiples across the board, and the AI capex cycle is plainly still alive if you look at Nvidia’s quarter, Broadcom’s AI revenue, or the capacity commentary from TSMC and ASML. But that counterargument actually strengthens the main takeaway: if the macro backdrop is less forgiving, then stock selection matters even more. In a zero-friction tape, almost every AI-linked semiconductor can levitate. In a harder tape, the market starts asking who controls supply, who owns the platform, and who can show the earnings now.
That is why we think the right analogy is not “the bubble burst,” but “the sorting phase began.” Late-cycle infrastructure booms often narrow before they end. The market keeps paying for the picks-and-shovels businesses with real bottleneck power while cutting down the names whose AI case is more derivative, more cyclical, or simply priced too richly for the proof on hand. The June tape looked brutal because the entire complex sold off together for a day. The rebound looked revealing because leadership quickly reconcentrated in the companies with the strongest claim on actual AI economics.
The bottom line is simple: the AI trade still works, but it no longer works evenly. Premium multiples are likely to hold best for companies that can point to visible monetization, constrained supply, and platform leverage — the businesses selling the compute, the manufacturing capacity, and the tools the ecosystem cannot easily replace.
What would change our mind? A real deterioration in demand signals from the infrastructure leaders, not just a valuation wobble in the broader chip complex. If Nvidia’s data center growth rolled over, if Broadcom’s AI bookings stopped converting, or if TSMC and ASML backed away from their capacity outlooks, then the thesis would be in trouble. Until then, this looks less like the end of AI leadership than a much harsher admissions process.
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.
