Chip stocks slide as strong earnings fail to calm AI concerns

Strong earnings fail to calm AI concerns

Markets were not handed one clean reason to sell. They were given a collection of smaller problems and, in the thin air of a summer session, decided there was no reason to lean against any of them.

US equities weakened, the Nasdaq again carried the heavier load, Treasury yields edged higher, the dollar firmed, and gold slipped below $4,000/oz. Bitcoin retreated toward $64,000, while oil briefly surged amid another round of escalation in the Middle East before giving back most of the move. It was not panic. It was a market losing traction at low speed, with liquidity thinner, conviction lighter and the marginal buyer nowhere in a hurry to step forward.

The macro data were mixed but offered little reason for alarm. Earnings remained broadly solid, inflation pressure continued to ease, and the AI capital-spending cycle stayed intact. Yet nearly every major asset class struggled to advance, suggesting the market was no longer arguing with the fundamentals so much as wrestling with the weight of its own positioning.

Nowhere was that clearer than in semiconductors.

ASML delivered a strong quarter and lifted its outlook. Micron produced a blowout result. TSMC reported record profit, with net income rising 77% from a year earlier to T$706.6 billion, or roughly $22 billion. Gross margin reached 67.7%, management raised its 2026 revenue-growth forecast to slightly above 40%, and capital-expenditure guidance climbed to $60 billion to $64 billion.

Those are not the numbers of an industry running out of road. They are the numbers of an expansion still moving at speed.

Yet semiconductor and memory shares continued to fall. The sector is now roughly 22% below its mid-June high, firmly in bear-market territory even as the earnings continue to confirm powerful AI demand. The chips are not being sold because the factories are empty. They are being sold because the theatre was packed, every seat was occupied, and the audience had already priced in the encore.

That is the problem with a trade built on constant acceleration. Once investors have paid in advance for an extraordinary future, strong results are no longer enough. TSMC did not merely need to prove that demand remained robust. It needed to show that an already exceptional buildout was becoming even more exceptional. ASML did not simply need to beat estimates. It needed to beat the version of the future already embedded in the share price.

When the earnings bar is hanging from the rafters, even a record jump can look like a stumble.

The price action therefore looks less like a collapse in the AI thesis and more like a factor unwind. Weakness has spread beyond one company or one corner of the chip chain into optical networking, data centres, AI infrastructure suppliers, high-beta momentum and popular hedge-fund longs. At the same time, parts of the broader market outside the AI complex have performed better, suggesting that money is not abandoning technology altogether. It is moving away from the most leveraged and crowded expression of the AI capital-spending cycle.

Once volatility began to rise as prices fell rather than climbed, the machinery of the trade shifted into reverse. Leveraged ETFs had to reduce exposure, bullish option positions were cut, momentum models began selling yesterday’s winners and hedge funds trimmed crowded holdings before everyone reached the same door.

That process can become self-reinforcing. Falling prices lift volatility, higher volatility reduces the amount of risk systematic and leveraged strategies can carry, and those strategies then sell more. At that point, the market is no longer conducting a calm valuation debate. It is pulling chairs away while the music is still playing.

South Korea has become the most extreme version of the same story. Samsung Electronics and SK Hynix sit near the centre of the global AI memory trade, making Korean equities one of the purest regional expressions of the hardware boom. That worked beautifully while prices were rising and leverage was being added. It becomes much less elegant when the trade turns.

Realized volatility in the KOSPI has surged toward levels rarely seen outside major historical disruptions. Foreign investors are selling at the fastest pace in decades, while leveraged and derivative-based products have amplified the move. Margin calls are now forcing investors to sell not because they have suddenly changed their long-term view of AI, but because their brokers have changed the terms of the conversation.

Korean authorities have tightened rules around leveraged products, but regulation can only slow the rush toward the exit. It cannot immediately rebuild confidence once leverage begins unwinding. Korea is where the market’s positioning fire found dry timber, and it offers a useful warning for the wider AI complex: once leverage becomes involved, falling prices stop being only an expression of opinion. They become instructions.

The macro backdrop did little to settle the argument because it offered something for both bulls and bears.

Housing remained the weakest link. Homebuilder sentiment disappointed and pending home sales came in below expectations, showing that elevated borrowing costs are still pressing on the most rate-sensitive areas of the economy. Retail sales were respectable rather than spectacular, with consumers still spending but the pattern becoming less uniform and previous revisions muddying the headline.

The labour market remained firmer. Initial jobless claims fell to 208,000, the lowest since early May, providing little evidence that employers are suddenly preparing for a downturn. Then came the Philadelphia Fed survey, which surged to 41.4 against expectations of 12.5, its strongest reading since 2021. The survey is notoriously erratic and deserves a large pinch of salt, but a move of that size cannot be dismissed entirely, particularly with its price components also firming.

The economy is therefore still walking a narrow ridge. The harder data are softening around the edges, especially in housing, while employment and survey momentum remain resilient. Recent inflation numbers have been more helpful, with softer CPI and PPI reinforcing the disinflation story and reducing the immediate risk of a July rate increase. Second-quarter GDP tracking has also moved toward 2.4% annualized, with domestic final sales around 2.3%.

That should ordinarily be a favourable mix for risk assets: growth holding up while inflation cools. But the same resilience that lowers recession risk also prevents the Federal Reserve from turning fully dovish. The July tightening risk has faded, yet a less hawkish September reaction still requires weaker payrolls and another round of softer inflation data. For now, the market appears to be back in the familiar cycle of kicking the hawkish can down the road rather than removing it altogether.

The Fed has stepped back from the trigger, but it has not left the room.

Treasury yields ended only modestly higher, although the long end remains the more important signal. The 30-year yield continues to hold above 5%, acting like a toll booth for every long-duration asset whose valuation depends on earnings far into the future. That matters enormously for AI because the further investors must travel to justify today’s price, the more expensive the journey becomes when discount rates remain elevated.

A company can deliver excellent growth and still see its shares fall if the cost of waiting for those profits rises. Positioning may have lit the match, but the bond market has kept the wood dry.

Away from semiconductors, the earnings season has not sent a broad signal of corporate distress. Industrials were generally solid, although higher fuel prices are beginning to weigh on airlines. Healthcare results pointed to improved cost control and operational recovery, while financial companies benefited from stronger fee income, capital-markets activity and improved revenue guidance..

Those are not the fingerprints of a corporate-profit recession. The market is simply becoming more selective about what it is prepared to pay for growth.

For much of the AI rally, investors were willing to forgive almost anything as long as capital spending, demand and margins kept moving higher. Now they are asking harder questions about the return on that spending, the cost of financing it and whether every company attached to the theme deserves the same premium. That is not the death of the AI trade. It is the end of the period when every shovel seller was treated as though the gold rush could never slow.

Geopolitics added another layer of friction. US operations against Iran intensified, with reports of strikes near important export infrastructure and continued disruption around the Strait of Hormuz. Tehran again raised the possibility of widening the conflict through the Bab el-Mandeb route if Iranian power infrastructure were attacked.

Yet oil’s response remained contained. WTI climbed toward $81/bbl before reversing below $79/bbl, while Brent held around $85/bbl. The market retained a geopolitical premium without moving into a full shortage panic.

That is because the physical market has not yet confirmed a severe loss of supply. Vessel traffic remains impaired and escort arrangements through Hormuz remain uncertain, but barrels have not disappeared on a scale large enough to force disorderly buying. Oil is still trading the threat of disruption rather than the arithmetic of missing supply.

The more revealing warning may be further down the barrel. Refined products and European natural gas remain firmer than crude, suggesting that processing capacity, transport routes and regional energy security could become the more vulnerable pressure points. Crude may be marking time, but the product market is keeping one eye open.

There are also narrow diplomatic off-ramps. The release of an American detainee was acknowledged as a gesture of goodwill, even as military pressure continued. The conflict may still become more uncomfortable before it becomes calmer, but the market’s refusal to chase oil much higher suggests traders continue to see a difference between escalation designed to improve bargaining power and escalation designed to destroy the negotiating table.

The broader cross-asset message was caution rather than panic. The dollar strengthened as US resilience and higher yields offered support. Gold slipped below $4,000/oz under pressure from the firmer dollar and real rates, while Bitcoin retreated toward $64,000 after failing to sustain its rebound.

More striking was the continued weakness in implied equity correlation. Index volatility remains relatively restrained even while individual sectors and factors are being pulled apart. The index wears a composed expression while the machinery beneath it shakes.

Low correlation allows gains in a handful of large companies to conceal substantial damage elsewhere. Investors watching only the S&P 500 may miss the intensity of the liquidation in semiconductors, momentum factors, and leveraged regional markets. This is not yet broad capitulation. It is concentrated pain, but concentrated pain can spread if the market’s leadership fails to stabilize.

There are reasons to believe the semiconductor unwind may be moving toward the later stages of its mechanical phase. Hedge funds have already reduced substantial AI exposure over the past five to six weeks. Leveraged ETF positions have fallen. Momentum has been damaged, and the most crowded trades are less crowded than they were in June.

Eventually, forced selling runs out of inventory.

That does not mean the market automatically returns to its highs. Cleaner positioning removes one obstacle, but the next group of buyers will still demand evidence. The fundamental AI capital-expenditure story remains intact, yet the buildout is also becoming more expensive, politically sensitive and operationally difficult.

Data centres require enormous amounts of power and water. Communities are becoming more resistant to new construction. Financing needs are rising, and credit markets are absorbing increasing amounts of AI-related issuance. Investors are beginning to ask when the returns from hundreds of billions of dollars of annual spending will become visible.

The four largest US AI operators are expected to spend more than $725 billion this year. At that scale, the market no longer asks whether demand is real. It asks whether the returns can justify the bill.

The next major test will come from hyperscaler and neocloud earnings. Those companies must show that capital spending is still rising fast enough to support the hardware ecosystem and provide a credible path to revenue and profit. Too little spending and the chipmakers suffer. Too much spending without visible returns and the hyperscalers come under pressure.

The market wants acceleration and discipline at the same time.

The AI train has not left the tracks. But after running at full speed for months, investors have started checking whether the bridge ahead was built to carry this much weight.