NextFin News - Semiconductor earnings expectations are still climbing, yet the sector has struggled to keep pace. That gap is the story: investors are still rewarding the AI chip complex for years of growth, but recent trading shows they are no longer willing to assume every upward revision should translate into higher share prices. On July 15, the iShares Semiconductor ETF (SOXX) closed at $555.27, down 2.23% on the day after a prior close of $567.92; the VanEck Semiconductor ETF (SMH) finished at $590.77, down 1.59% after a previous close of $600.31; and the State Street SPDR S&P Semiconductor ETF (XSD) fell 5.68% to $506.54 from $537.04. Each ETF remains far above its 52-week low, but all sit below recent highs, a sign that the market is repricing the group rather than abandoning it.
The tension is visible in the numbers. SOXX still showed a year-to-date daily total return of 88.78%, SMH 66.69%, and XSD 67.14% on the quoted pages, which means the sector has already absorbed a large amount of optimism. At the same time, the price action has turned choppy enough to force a harder question: if estimates keep moving up, why are the stocks not holding their highs? The answer is not that AI demand has vanished. It is that the market may be moving from a phase of narrative expansion to a phase of proof, where each new capex plan, earnings beat, or supply update has to justify not only current growth but also the multiple attached to that growth.
That shift matters because semiconductors are now carrying more than their own earnings cycle. They have become a proxy for how long the AI infrastructure boom can compound before it runs into budget discipline, customer concentration, or margin pressure. If the biggest buyers of chips start pulling forward less demand than expected, the current estimate trend can slow even if the underlying industry remains healthy. In that sense, the market is testing whether the AI trade is still a growth story or whether it has become a crowded duration trade.
What Is The Market Actually Pricing?
The sector is still trading as a growth leader, but not with the same conviction as before. The quoted ETF data show that SOXX, SMH, and XSD all remain well above their 52-week lows — $232.33, $279.19, and $254.82 respectively — which confirms that no broad collapse has occurred. Yet the same quotes also show that the group sits below its 52-week highs of $655.95, $671.83, and $658.14. That combination is important: investors have not given up on chips, but they are no longer willing to price them as if the next leg higher is automatic.
That is why the phrase “disconnect” is useful. It does not mean earnings expectations are falling. It means the relationship between those expectations and share prices is loosening. In a sector where valuation is tied to future growth more than current profits, even a modest change in confidence can matter more than a clean beat on the next quarter. The market is effectively asking whether the same earnings growth can still command the same multiple when the AI spending cycle has already done so much of the heavy lifting.
There is also a clear second-order effect. When investors worry that semiconductor demand is no longer purely incremental, they begin to look through the supply chain rather than just at the chipmakers themselves. Memory vendors, foundries, equipment companies, and designers are all linked by the same customer budgets. If hyperscalers stretch out deployment schedules or become more selective about procurement, the initial hit shows up in order expectations, but the later hit shows up in estimates, margins, and capital spending assumptions across the chain.
That is why a strong print can coexist with a weak stock response. TSMC’s latest quarterly net profit surged 77.4% year over year to NT$706.6 billion, or about US$22 billion, while the company said it will add another US$100 billion to its Arizona investment, bringing total U.S. investment plans to US$265 billion. Those are the kinds of numbers that normally reinforce the bullish semiconductor case. Yet the broader sector has still failed to sustain its highs. The message is not that demand is weak. It is that the market is increasingly focused on the sustainability of the demand path.
“The point where rising memory costs for the biggest AI buyers could start moderating the demand signal that semiconductor stocks are built around.”
That warning captures the mechanism. The pressure is not simply on chip orders; it is on the behavior of the buyers generating those orders. If a handful of large cloud platforms control the pace of AI infrastructure spending, then rising memory costs, capacity bottlenecks, or capital discipline at those buyers can ripple across the whole chip complex faster than a normal end-market slowdown.
Why The Sector Can Look Strong And Weak At The Same Time
The first part of the answer is cyclical. Semiconductor stocks have always been vulnerable to inventory and capex swings, even when end-demand stays constructive. A company can report strong revenue growth while the stock weakens if investors think the best part of the cycle is already in the numbers. That is especially true in a crowded trade. When positioning is heavy, the market often sells the idea before the data confirm it.
This is what makes the current drawdown different from a clean demand collapse. There is no evidence here of a collapse in the AI buildout. TSMC’s profit, the sector’s still-elevated year-to-date returns, and the continued optimism embedded in earnings forecasts all argue against that interpretation. But the market does not need a collapse to reprice the group. It only needs slowing acceleration. If earnings expectations keep rising at a slower rate than they did earlier in the year, while prices are already sitting near elevated multiples, the result is a flatter path for the shares even if the businesses remain healthy.
The second part of the answer is structural. AI has changed the composition of semiconductor demand in a way that should outlast this cycle. Data-center chips, advanced packaging, networking, and memory tied to artificial intelligence are now a larger part of the story than consumer electronics or PCs. That is a permanent shift in end-market mix. But the spending rate attached to that shift is still cyclical. It can accelerate, plateau, or decelerate depending on the return on capital that hyperscalers see in the next wave of infrastructure. Structural demand growth can coexist with cyclical stock weakness when investors think the market has front-loaded too much of the good news.
That is why the most useful lens is not “is AI demand real?” It is “what rate of demand growth is already in the price?” The answer is probably a high one. SOXX has a beta of 2.24 and SMH a beta of 1.98 on their quoted pages, which is a reminder that these funds amplify conviction as well as doubt. When conviction is high, the sector can move violently even without a dramatic change in fundamentals. That is exactly what a repricing looks like: the underlying thesis survives, but the multiple attached to it becomes more selective.
The broader market context also matters. Semiconductor stocks are no longer isolated names with idiosyncratic earnings. They have become one of the market’s main expressions of the AI capex cycle. That makes them sensitive not just to chip demand, but to the tone of commentary from the biggest cloud and infrastructure buyers. If those buyers keep spending aggressively, the sector can recover quickly. If they start emphasizing efficiency, pacing, or discipline, the market will assume the supply chain has to absorb a slower growth rate.
This is why the disconnect can persist even when estimates remain in a rising trend. Earnings revisions are a lagging expression of capital allocation decisions made by a small number of large customers. Share prices, by contrast, move on the market’s first read of whether those decisions are likely to accelerate or slow. When the two drift apart, the stock market is usually signaling that the marginal buyer has become less enthusiastic than the consensus model.
What Would Prove The Skeptics Right?
The strongest counter-thesis is straightforward: the selloff is only a pause inside a secular AI uptrend. That view has legitimate support. The sector is still far above year-ago levels, TSMC continues to post record profits, and the chip supply chain is still being pulled along by data-center demand rather than pushed by a fading legacy market. On that reading, the weakness is a valuation reset after a huge run, not the start of a structural unwind.
That argument cannot be dismissed. Semiconductors often overshoot on the way up and on the way down. A sector that rallies as hard as this one can easily need a long digestion period before the next advance. The question is whether the current move is merely that digestion or the market’s first sign that the AI spending curve is flattening. The answer will come from the buyers, not the sellers.
The key falsifying signal for the bearish read is quantifiable: if the major hyperscalers continue to maintain or raise AI infrastructure spending plans, and that spending continues to feed through into chip orders and forward estimates without a material slowdown, then the current weakness will look like a cyclical reset rather than a regime change. If those capex plans flatten, the market is likely to treat today’s estimates as too high even before the earnings reports catch up.
The base case is that the sector is in a cyclical correction inside a still-structural AI uptrend. In the short term, sentiment and positioning can keep weighing on chip ETFs even if the underlying demand picture stays constructive. In the medium term, forward earnings can still move higher, but likely at a slower pace than the market assumed during the most aggressive part of the rally. In the long term, the semiconductor industry remains structurally better placed than it was before the AI buildout, but that does not mean every link in the chain deserves the same multiple.
The upside case is that AI spending broadens and remains strong enough to support continued estimate expansion across memory, foundries, equipment, and design. The downside case is that the biggest buyers begin to slow the pace of infrastructure buildout, which would force the market to reprice chip earnings expectations lower even if reported results remain solid for a while.
The market is not saying chips are broken. It is saying the burden of proof has risen. When earnings expectations keep climbing but shares stop confirming them, the message is usually simple: the trade has become more expensive to believe in.
Explore more exclusive insights at nextfin.ai.
