Eight Straight Losing Sessions for Chip Stocks Is the Longest Streak Since 2015. A Shorter One Last Year Preceded a 181% Rally.

Eight. That's the number of straight losing sessions now behind the iShares Semiconductor ETF (SOXX), the longest such streak since September 2015 and only the eighth time it's happened since 2001. The Dow Jones Industrial Average, meanwhile, closed Thursday near 52,238, up about 1%, lifted by strong bank earnings. Chip stocks and the rest of the market are no longer moving together, and the gap between them is the real story this week.

[[IMG1]]

The scale of the drawdown is real and still growing. Micron has fallen for six straight sessions and lost roughly $227 billion in value since October 2. SK Hynix, down 11% in Seoul on Thursday alone, has erased about $353 billion over the same stretch in its own seventh straight loss. Samsung has shed roughly $344 billion. Combined, six of the largest memory and storage makers have lost about $1.09 trillion since the close on October 2. The pressure wasn't confined to memory chips. Arm Holdings, Applied Materials and Intel were each down double digits over the comparable six-session window through Wednesday, and SOXX itself traded near $502 Thursday morning, down more than 3% on the day and approaching a 15% decline for the full streak. South Korea's Kospi slid toward a six-month low this week on sharp declines in SK Hynix and Samsung, both central to the memory chips that feed AI data centers.

Behind the number is a question that's been building for weeks: can AI infrastructure spending generate enough return to justify its cost? Broadcom CEO Hock Tan said Monday at a panel in Singapore that access to capital, not a shortage of chips or memory, could become the biggest constraint on AI's expansion, and he compared some of the speculative investing around it to casino gambling. The International Monetary Fund followed with a warning in its Global Financial Stability Report that weaker-than-expected returns on AI investment could trigger a broader technology correction, amplified by rising leverage and concentration, noting that hyperscalers now account for roughly a tenth of global investment-grade bond issuance, up from under 1% in 2024. That leverage concern isn't abstract. Credit insurance on Oracle and CoreWeave's debt hit a record high this week, and CoreWeave's own shares fell more than 15% Wednesday and another 9% Thursday.

This isn't the first crack in the story. OpenAI told investors its annualized revenue was running about $20 billion below what had previously been reported, and the chip stocks tied most closely to its buildout fell with it. SpaceX is reportedly seeking $40 billion in debt specifically to buy Nvidia chips, a bet placed while the 10-year Treasury sits near a 24-year high. Higher borrowing costs cut two ways for a sector that prices itself on profits still years away: they make the debt funding AI data centers more expensive, and they shrink the present value of the earnings Wall Street is betting on. Memory makers face an added squeeze from China's CXMT, the same competitive threat that weighed on Samsung and SK Hynix's shares earlier this week.

Thursday added a fresh data point: September's producer price index rose 6.3% year over year, the sharpest increase since 2022 and well above the 5.5% economists had forecast. That came a day after a stronger-than-expected consumer price report pushed up expectations for a Fed hike, and traders now put the odds of a hike at the Fed's meeting in two weeks at roughly a coin flip. Yet Treasury yields actually eased Thursday, with the 10-year down about 9 basis points to 5.10%. That's worth sitting with. Chip stocks kept falling even as the rate pressure supposedly driving the selloff loosened a notch, which suggests the AI capex debate has taken on a life of its own, separate from the rates story.

The divergence between chipmakers and the broader market has been building all month. Marvell's own bullish 2028 AI revenue forecast failed to lift the rest of the market in September, with only about a quarter of S&P 500 stocks confirming the rally at the time. Thursday looked similar in miniature: the Dow closed up about 1% near 52,238 and the S&P 500 gained roughly 0.35% to about 7,746, both lifted by bank earnings and old-economy names, while the Nasdaq barely budged as chip losses offset gains elsewhere. When a market's gains concentrate in a shrinking number of sectors, a stumble in the biggest one draws outsized attention, whether or not it reflects the health of the economy as a whole.

What History Says About Losing Streaks This Long

An eight-session losing streak is genuinely rare for SOXX. It's happened only seven times since 2001, most recently in September 2015, and the research available on this streak doesn't extend forward-return data to that specific threshold. The more commonly tracked marker is a shorter, seven-session losing streak, which has occurred 15 times since July 2001, including the run-up to this one. In the 14 completed seven-session cases, SOXX was higher a month later by an average of 3.24%, more than double the 1.45% average return across all one-month periods. Three months out, it averaged a 10.02% gain, versus 4.40% for a random three-month stretch. Twelve months out, the average climbed to 35.28%, with the index higher 85.7% of the time, compared with a 67.7% win rate on any given trading day.

The precedents cut in both directions. The seven-session streak that ended in May 2025 was followed by an 11.8% gain a month later, 39.6% after six months, and 181.4% after a year, as the AI investment boom accelerated. But a losing streak of the same length in January 2008 preceded a 43.9% loss over the following year, as the global financial crisis took hold. A similar stretch that October saw chip stocks fall another 13.4% over three months before rebounding 35.4% over the next year. Fourteen data points is a small sample, and most of them sit inside a 25-year period when chip stocks mostly went up, which flatters the buy-and-hold math. Still, the pattern is a reminder that a sharp pullback and the start of a structural decline aren't the same thing, and the difference usually only becomes obvious well after the fact.

SOXX isn't flashing the kind of extreme reading that often accompanies a market bottom. As of Wednesday, it sat just below its 50-day moving average near $539 but remained comfortably above its 200-day average near $466, with a relative strength index hovering around 40, above the level below 30 that traders typically associate with an oversold asset. It's also still solidly positive for the year even after this week's slide, ahead of most other major industry groups.

The Pulse24 Take

The AI capex debate isn't going away, and it shouldn't. Spending at this scale, with chipmakers, cloud providers and now even rocket companies all borrowing to build out compute, deserves the scrutiny it's getting from the IMF, from Broadcom's own CEO, and from credit markets pricing default risk into CoreWeave's bonds. Questioning whether the returns show up on schedule is healthy, not alarmist.

What this week's SOXX streak adds is a reminder that healthy skepticism and a healthy correction look a lot alike from the inside. UBS has described the pullback as a recalibration of expectations rather than a break in the AI investment story, and it's still forecasting roughly 28% earnings growth for the S&P 500's third quarter. The historical pattern on seven-session SOXX losing streaks points the same way, more often marking an entry point than a top, though both the 2008 precedent and the sheer rarity of an eight-session streak are reasons to treat this one with more caution than usual.

The number worth watching from here isn't the ETF's daily move. It's whether the companies actually spending the money, the hyperscalers, OpenAI, and the data center builders leaning on debt, start showing results that justify the spending on a normal earnings timeline rather than a hypothetical one five years out. That's the test the market is actually running right now, rates and headlines aside.