The trade that defined the first half of 2026 — buying anything with proximity to a GPU — broke apart in the holiday-shortened week before Independence Day. The PHLX Semiconductor Index, which had surged more than 80% in the first half, sank 6.3% on Wednesday and 5.4% on Thursday, a two-session decline of roughly 12%. This sharp reversal caught many investors off guard, as semiconductor stocks had been the market’s darlings for months, fueled by insatiable demand for AI chips from hyperscalers like Microsoft, Amazon, and Google.
While chip stocks cratered, the Dow Jones Industrial Average closed at a record 52,900 on Thursday, lifted by a 5% surge in Apple after Bloomberg reported the company had instructed suppliers to prepare 10 million foldable iPhones for a launch this autumn. All three major indices finished the week higher despite the rotation, underscoring the narrowness of the selloff. The Dow’s gain was broad-based, with sectors like healthcare and consumer staples also contributing, but the spotlight remained on the dramatic shift out of semiconductors.
The numbers behind the crack
Micron Technology led the decline, falling more than 10% on Wednesday alone. SanDisk, Applied Materials, and Lam Research all dropped roughly 10%, while Intel and Marvell each lost about 9%. The selling extended across the entire semiconductor value chain, from memory makers to equipment suppliers. The sheer magnitude of the two-day plunge erased nearly $500 billion in market capitalization from the PHLX index constituents, a stark reminder of how concentrated gains had become in the AI trade.
The selling intensified after reports that SK Hynix was slowing its expansion of high-bandwidth memory (HBM) production, a signal that the supply side of the AI infrastructure boom may be catching up with demand. HBM is a critical component for NVIDIA’s GPUs, and any indication of oversupply spooks investors who have been betting on perpetual shortages. Equipment makers ASML, KLA, and Applied Materials all fell between 5% and 6%, suggesting investors are pricing in a slowdown in chip-factory orders. ASML, which manufactures the lithography machines essential for advanced chip production, saw its stock drop 5.8% despite announcing a new order from TSMC the previous week. The market’s reaction indicates that future orders may be at risk if hyperscalers reduce their capital expenditure plans.
The damage to chip stocks happened against a backdrop of broader macro uncertainty. June nonfarm payrolls came in at just 57,000, roughly half the 110,000 consensus estimate, and revisions to April and May cut a combined 74,000 jobs from previous tallies. The unemployment rate dipped to 4.2%, but only because the labour force participation rate fell to 61.5%, its lowest level since March 2021. This suggests that the job market is weakening, not tightening, as more workers leave the labor force entirely. For the Federal Reserve, this complicates the inflation fight, as a soft labor market could allow the central bank to hold rates steady without worrying about overheating.
Where the money went
The rotation was not a flight from AI but a repricing of where the returns will come from. Enterprise software stocks, led by ServiceNow, Snowflake, and Palantir, have been the primary beneficiaries, with the iShares Expanded Tech-Software ETF up 35% from its April low. This shift reflects a growing conviction among institutional investors that the AI infrastructure buildout has already been priced in, and that the next wave of gains will come from companies that actually use AI to generate revenue. The logic is straightforward: spending on GPUs and data centers is a cost, not a profit center, unless the applications built on top of that infrastructure produce measurable returns.
Snowflake surged 36% in late May after reporting strong earnings, adding 616 net new customers and lifting its count of million-dollar accounts to 779. ServiceNow, Oracle, and Palantir each rallied 6% to 8% in the session that followed. Snowflake’s data cloud platform has become a key beneficiary of AI workloads, as enterprises need to store, manage, and analyze vast amounts of data before they can train or deploy models. Its customer additions indicate that demand for data infrastructure is accelerating, even as the hardware side faces headwinds.
Palantir’s Q1 revenue hit $1.63 billion, up 85% year over year, driven by its AI platform (AIP) that helps clients integrate large language models into their operations. The company’s commercial revenue grew 95% in the quarter, while government revenue rose 67%. ServiceNow has projected $30 billion in subscription revenue by 2030, with roughly a third attributed to its AI product, Now Assist. This product automates IT workflows and customer service, and early adopters report up to 40% reduction in ticket resolution times. Such concrete metrics are precisely what investors want to see: evidence that AI is not just a cost center but a profit driver.
The valuation question
The first half left all three major indices in strong shape. The S&P 500 gained 9.6%, the Nasdaq rose more than 12%, and the Dow climbed 8.9%, its best first-half performance since 2021. However, these gains were heavily concentrated in a handful of mega-cap tech stocks, raising concerns about market breadth. The Shiller CAPE ratio sits at 38 to 40, second only to the dot-com peak of 44, and market concentration in the largest technology stocks has exceeded year-2000 levels. As of mid-2026, the top five companies (NVIDIA, Apple, Microsoft, Alphabet, Amazon) account for over 28% of the S&P 500’s market cap, compared to about 18% at the dot-com peak.
Proponents argue that this time is different because these companies are among the most profitable in corporate history. NVIDIA alone reported net income exceeding $120 billion for fiscal 2026, with gross margins above 75%. The company’s data center revenue grew 145% year over year, driven by sales of its new Blackwell architecture GPUs. Apple and Microsoft each generated over $100 billion in net income, and their cash reserves provide a cushion against downturns. In contrast, during the dot-com era, many of the largest companies were unprofitable and reliant on speculative funding.
The counterargument is that profitability at the top of the supply chain does not guarantee profitability in the middle. Hyperscalers are on track to spend more than $650 billion on AI infrastructure in 2026, with Amazon alone projecting $120 billion in capital expenditures. The question that spooked markets in the holiday-shortened week is whether anyone below NVIDIA in the stack will earn returns that justify those prices. Companies like AMD, Intel, and Marvell have seen their stocks rise on the AI hype, but their actual revenues from AI chips remain a fraction of NVIDIA’s. If demand for AI infrastructure slows, these companies could face severe margin compression.
What the Fed changes
The weak jobs report reshaped the interest-rate picture overnight. The probability of a Fed rate hike at the 29 July meeting collapsed to about 22%, with a hold now the overwhelming favourite at 78%. Earlier in the quarter, the market had priced in a 50% chance of a hike after stronger-than-expected inflation data. The sudden shift in expectations reflects the Fed’s dual mandate: maximum employment and price stability. With the labor market showing clear signs of softening, the Fed is unlikely to risk a recession by raising rates.
Fed chairman Kevin Warsh called the jobs picture “steady” and continued to emphasise his commitment to the 2% inflation target, without offering forward guidance on the rate path. Warsh, who succeeded Jerome Powell in 2025, has maintained a cautious stance, noting that while headline inflation has fallen to 2.8%, core services inflation remains sticky at 3.5%. A Fed that stays on hold gives equity markets one less reason to sell, but it also removes the catalyst that had been supporting bank stocks and the dollar earlier in the quarter. The dollar index fell 1.2% on Friday after the payrolls report, boosting commodity prices and emerging market equities.
For the AI trade specifically, a steady Fed reduces the urgency for yields to rise, which supports high-multiple stocks like enterprise software companies. If yields had surged on a hawkish Fed, growth stocks would have been penalized. The current environment — low yields, steady rates, and a soft labor market — favors the rotation into software stocks, as their future cash flows are discounted at a lower rate. However, the risk is that if inflation reaccelerates, the Fed may be forced to act, disrupting the entire equity market.
What to watch
The market reopens on Monday with Q2 earnings season approaching. The results that matter most will come from the AI software layer: whether Snowflake’s customer additions accelerate, whether Palantir’s commercial pipeline converts, and whether ServiceNow’s AI attach rate holds at the levels it projected. Snowflake is expected to report revenue growth of 30% year over year, but investors will scrutinize net revenue retention rates, which have fallen from 170% to 140% over the past three years. Palantir’s government contracts provide visibility, but its commercial segment remains the wild card. ServiceNow has guided for subscription revenue of $2.5 billion in Q2, with Now Assist contributing 12% of new deals.
If the software companies deliver, the rotation will look prescient. If they disappoint, the AI trade will face a harder question: what happens when the infrastructure is built and the applications do not come. Past technology booms — from railroads to the internet — have shown that infrastructure builders often enjoy temporary monopoly profits, but the biggest long-term winners are the companies that use that infrastructure to disrupt existing markets. Whether AI follows that pattern is the central uncertainty for the second half of 2026. The next few weeks of earnings will provide the first crucial clues.