The Philadelphia Semiconductor Index ended July 17 more than 20% below its late-June record, crossing the usual bear-market threshold after gaining about 65% year to date. The drop changes the terms of the AI trade. Semiconductor investors now need more than exposure to data centers and advanced computing. They need evidence that each company’s earnings can hold up when the first rush of AI infrastructure spending gives way to slower growth.
This is not yet proof that AI spending has collapsed. Earnings expectations for the industry remain unusually strong, and projected cloud and AI infrastructure spending still points to a large buildout through 2027. The market reaction instead suggests that expectations had become expensive. When a sector has already risen sharply, a company can disappoint even while reporting growth if its results do not support the next leg of the profit story.
That makes this a stock-picking market. The key question is no longer whether chips benefit from AI. It is whether a company has the demand, pricing power, and profit outlook to justify what investors still expect from it.
The burden of proof has changed
A bear market can begin as a valuation reset, not a demand recession. Valuation is simply the price investors are willing to pay for future profits. When confidence is high, investors often accept a high price because they expect earnings to rise quickly. When confidence weakens, they demand more proof before paying that same price.
That distinction is especially useful in semiconductors. The group can fall even if aggregate industry earnings continue to rise. Investors may be reassessing how long exceptional growth can last, how much of it is tied to the first stage of AI infrastructure spending, and which companies have enough profit support if growth slows.
A company reporting results in line with expectations may therefore get little credit. The market will be looking beyond the headline numbers for signs that orders are holding, pricing is firm, and margins can stay healthy. Margin measures how much profit remains after costs. Rising revenue with weakening margins is a far less appealing result than the top-line figure alone suggests.
The sector’s sharp gain before the selloff left little room for small disappointments. A cautious view of demand, softer pricing, or excess customer inventory could cause investors to lower their assumptions for future profits. That does not require a broad collapse in AI spending. It only requires growth to fall short of a very high bar.
Near-term earnings are still unusually strong
LSEG expects earnings for the S&P 1500 semiconductor and equipment industry to more than double in 2026. That forecast is driven largely by Micron and Nvidia, which are closely tied to the early wave of AI infrastructure demand.
The forecast then calls for 46.1% earnings growth in 2027. That is still a fast growth rate. Yet it is a meaningful change from earnings more than doubling. The investment math gets tougher as the rate of improvement slows. A high valuation can look reasonable when profits are rising at an extraordinary pace. It becomes harder to defend when the market must look further into the future for the next source of growth.
This is where broad sector labels become less useful. Two companies can both sell chips into AI-related markets while facing very different earnings paths. One may have tight supply and rising prices supporting near-term profits. Another may need sustained order growth to support an above-average valuation. A third may be judged mainly on whether it can show demand beyond the first stage of data-center construction.
The market may be shifting from rewarding the AI connection to rewarding earnings durability. Investors are likely to put greater weight on management commentary about demand visibility into 2027. They will also be watching whether current demand reflects lasting customer needs or a concentrated period of infrastructure spending.
That does not make the earnings outlook weak. A forecast for more than doubled industry earnings in 2026, followed by 46.1% growth in 2027, remains a powerful backdrop. It does mean that the market can become less forgiving. Strong industry growth does not protect every company if its own results fail to support the valuation investors assign to it.
Memory has a clearer near-term support
Tight memory supply and rising prices support Micron and other memory suppliers. That creates a direct path from industry conditions to revenue and margins. When supply is limited and prices rise, the near-term earnings setup can improve quickly.
Still, memory requires a different kind of caution. It is cyclical, which means conditions can change as supply and demand move over time. A low valuation based on unusually strong earnings can mislead if those earnings sit near a pricing peak. If supply catches up with demand or demand cools, profit can fall quickly. The multiple may have looked modest only because the earnings figure was temporarily high.
The useful question for memory investors is not whether current pricing is favorable. It is how much of that favorable pricing can persist into the period when industry earnings growth is expected to slow. Tight supply supports the near-term case. It does not settle the longer-term earnings question.
The pressure is different for Intel, AMD, and Marvell. Their forward P/E ratios remain above their longer-term averages. A forward P/E compares a company’s share price with analysts’ expected earnings over the next year. Above-average forward P/E ratios indicate that investors are still paying more than usual for expected profit growth, even after the sector decline.
That does not prove those valuations are unjustified. Higher valuations can be supported by higher growth. But they leave less room for an ordinary setback. If orders slow, prices weaken, or margins come under pressure, investors may cut the premium they are willing to pay before any broader AI spending downturn is visible.
The market is therefore judging two distinct risks. Memory investors need to assess the durability of a favorable supply-and-pricing cycle. Investors in Intel, AMD, and Marvell need to assess whether expected earnings growth is strong enough to support valuations that remain above their longer-term norms. Those are not interchangeable tests.
Intel and Texas Instruments provide the next read
Upcoming results from Intel and Texas Instruments offer the next observable test of the selloff. Their reports may help show whether the market is mostly repricing elevated expectations or responding to a wider weakening in customer demand.
The most useful details will likely sit beneath the headline results. Management commentary on orders, inventory, pricing, and demand into 2027 carries more weight than a single quarter’s revenue figure. Orders can show whether customers are still committing to future purchases. Inventory comments can indicate whether customers are working through chips already on hand. Pricing helps reveal how much bargaining power suppliers retain as demand becomes less certain.
No one signal will settle the issue. A soft quarter could reflect normal swings in a cyclical industry. Similarly, a one-day rebound in semiconductor shares would not confirm that the demand outlook is secure. The more serious warning would be a combination of softer orders, pressure on pricing, and signs that customer inventories are rising or taking longer to clear. That pattern would point beyond a simple reset in valuation assumptions.
The strongest countercase is substantial. Global cloud and AI infrastructure capital spending is projected to approach $1.5 trillion by 2027. If hyperscalers continue funding data centers at that pace, demand for chips used across AI infrastructure could remain stronger than the selloff implies.
Even that outcome would not lift every semiconductor company equally. Large infrastructure spending can favor companies most closely tied to the strongest areas of demand, while firms with slower earnings growth may remain exposed to high valuation expectations. The sector’s decline has made that separation more visible.
The next test is whether Intel and Texas Instruments describe stable orders, manageable inventories, firm pricing, and credible demand visibility into 2027. If management commentary supports those conditions, the selloff may look more like a reset in what investors will pay for growth. If the commentary points to weakening demand as well, the market will have a stronger reason to question the industry’s current earnings path.
