Everything About AI
NextFin News - Short sellers wagering against semiconductor stocks have racked up roughly $32 billion in gross paper profits over the past seven days, as the Philadelphia Semiconductor Index fell more than 12% and headed into Friday's close on track for an eighth straight losing session — its longest streak of declines since 2006. The S&P 500 fell 1.1% to 7,659.44 and the Nasdaq composite sank 2.1% to 26,140.03, capping the worst week for both indexes since June: the S&P 500 ended down 1.9% and the Nasdaq down 4.5%.
For most of this bull market, the AI trade was a one-way bet: buy the chips, buy the builders, buy the power. This week, that consensus cracked. Short sellers are no longer just picking off weak credits — they are betting against the entire complex, from memory makers to the debt-fueled data-center operators that buy their silicon. And for the first time in this cycle, they are being paid for it.
The Numbers Behind the Rout
Short sellers recorded $32 billion in gross paper profits over the seven-day span, according to data from S3 Partners. The Philadelphia index, known as the SOX, fell more than 12% in that window, and 20 of its 30 components have slid by more than 10% since the rout began. Coherent dropped close to 30%. Astera Labs, Macom Technology Solutions and Credo Technology each lost more than 20% in the eight-day slide. The VanEck Semiconductor ETF fell 7.4% for the week, with Micron Technology, Advanced Micro Devices and SanDisk among the biggest drags.
The reversal follows a steep climb. The SOX more than doubled between the end of March and a record high in June, adding roughly $6.5 trillion in market value. That run-up is precisely what gave the shorts their opening. "The run up in chips was very aggressive so invariably there was going to be some consolidation or pull back," said Dec Mullarkey, managing director of investment strategy and asset allocation at SLC Management.
The selling was broad, not isolated. On Friday alone, Micron fell 5.6% and Intel lost 5.8%. The Dow Jones Industrial Average fell 175.28 points, or 0.3%, to 52,002.41, while the Russell 2000 index of smaller companies dropped 1.7% to 2,721.37, extending its correction. For the year, the S&P 500 remains up 11.9% and the Nasdaq up 12.5% — gains built in the first half, before inflation re-accelerated and rates began to bite.
The trigger was a string of bad news that arrived at once. September consumer prices rose 0.6% month over month and 3.7% year over year, up from 3.4% in August, reinforcing bets that the Federal Reserve will keep lifting rates. The dollar hit its highest level since May 2025. The yield on 10-year Treasuries, after briefly pricing in two Fed rate increases by the end of January, settled at 5.097% on Friday, down from 5.19% a day earlier, as the equity retreat kindled demand for bonds.
At the center of the week's rout sat CoreWeave, the Nvidia-backed cloud builder. Its stock was headed for a weekly decline of more than 20% after its debt fell to levels usually associated with financial distress; traders betting against the shares were positioned for gains of more than $1 billion on a mark-to-market basis by week's end. Some of the short positions in CoreWeave stem from convertible-bond arbitrage trades rather than purely bearish wagers, according to Matthew Unterman, a managing director at S3. Credit-default swaps on the company are pricing in roughly a 60% probability of default within five years, up from about 50% two and a half months earlier, after five-year spreads blew past 1,000 basis points. On Wednesday alone, CoreWeave fell 15.5%.
Why the Chip Trade Broke
The immediate mechanism is the discount rate. Chip stocks, especially the AI-linked names that led this market higher, are long-duration assets: much of their value sits in cash flows expected years from now. When the 10-year Treasury yield climbs and the dollar strengthens, the present value of those distant earnings falls. That is basic valuation arithmetic — and it hit the most extended part of the market first.
But the deeper channel is financing. The AI buildout is not being paid for out of retained earnings alone. CoreWeave and its peers in the "neocloud" cohort are borrowing heavily to buy GPUs, betting that hyperscaler demand will cover the interest. Deutsche Bank estimated that CoreWeave faces roughly $45 billion in required capital expenditure over the next 12 to 18 months against about $10 billion in next-twelve-month operating cash flow — a funding gap of $35 billion that must be raised in a market suddenly less forgiving of leverage. Broadcom chief executive Hock Tan compared certain AI investors to gamblers earlier in the week and cautioned that the sector may run into funding constraints.
"While increased competition within AI, rising financing requirements, and higher rates may periodically add to volatility, we view this latest bout as more of a recalibration of expectations than a fundamental break in the AI investment story," said Ulrike Hoffmann-Burchardi at UBS's Chief Investment Office.
The recalibration, however, is hitting at the moment when earnings must prove the spending thesis. TSMC reported a sixth straight quarter of record profit on Thursday, with revenue up 51% year over year to NT$1.49 trillion and gross margin of 66.4%, yet the stock fell anyway — a sign that even flawless execution no longer moves the needle when the market is questioning the durability of demand. The 2-nanometer process now accounts for 8% of revenue, up from 3% a quarter earlier, but the ramp weighed on margins, which fell 1.3 percentage points sequentially. High-performance computing accounted for 68% of revenue, up from 57% a year earlier, while smartphones declined to 21% from 30%.
Memory is sending the same mixed signal. Micron reported record fiscal fourth-quarter revenue of $54.23 billion — evidence that demand is still real — even as its shares joined the weekly decline. That tension, between strong current earnings and a market pricing in a slowdown, is the defining feature of this selloff.
Short Covering Versus Structural Damage
Here is the question that will decide whether this is a buying opportunity or the start of something worse: is the selling cyclical profit-taking by shorts cashing in, or is it the first real crack in a structural bull market?
The evidence for a cyclical read is strong. The Philadelphia index's eight-day losing streak matches a pattern last seen in 2006 — sharp, fast drawdowns that were followed by new highs. Positioning has become as one-sided as it has been all year. Hedge funds have been net sellers of global AI stocks for five consecutive sessions, wiping out all net buying accumulated since the start of 2025 and pushing the long/short ratio to the seventh percentile of the past year, according to Goldman Sachs data. When positioning is that crowded on the short side, even a modest positive catalyst can force covering and ignite a rally. Nasdaq-100 futures printed a hammer candlestick reversal pattern this week, the semiconductor index touched a key technical support level, and South Korea's KOSPI reached its most oversold level since October 2022 — conditions that often precede a bounce.
But the structural counter-argument is equally forceful. The AI capital cycle may be running into a wall of economics that no amount of positioning can fix. Will Sommer, a senior director analyst at Gartner, noted that AI providers are expected to spend nearly $1 trillion on infrastructure in 2026 and an additional $6 trillion between 2027 and 2030 — while enterprise and consumer AI revenue is insufficient to support profitable returns on all planned capital expenditure during that period. In other words, the demand side of the equation may never catch up to the supply side.
"It's clear as ever that inflation is stubborn and moving in the wrong direction," said Skyler Weinand at Regan Capital. "While we expect the Fed to pause on any rate changes in October, we expect rate hikes to resume in December."
This is the crux. If the Fed resumes hiking in December, the discount-rate pressure does not lift — it compounds. A cyclical bounce in that environment is a selling opportunity, not a bottom. The market is not pricing that risk lightly: the CME FedWatch tool showed an 85% probability that the Fed leaves rates unchanged at its October 27-28 meeting, down from 28% odds of a hike a day earlier. Several Fed officials have signaled in recent weeks that they are comfortable waiting, and markets are pricing the next increase in December.
The Second-Order Trade Nobody Is Discussing
The first-order effect of this selloff is obvious: chip stocks fall, shorts profit. The second-order effect is what happens to the companies that were counting on cheap capital to fund the buildout. Higher financing costs do not just compress valuations; they force capital rationing. Projects that cleared the hurdle rate at 4% do not clear it at 6%. That means order cancellations, delayed deployments, and ultimately lower revenue for the equipment makers and foundries at the top of the chain — the very companies reporting record earnings today.
The third-order effect is the expectation gap. The market has priced a soft landing with steady AI demand. If capital rationing slows the buildout, earnings estimates for 2027 fall even as valuation multiples compress — a double hit that 2006-style technical bounces do not capture. This is why the divergence between the indexes matters: for the week the Nasdaq fell 4.5% while the Dow rose 0.7%. Money is rotating out of long-duration growth and into shorter-duration value — a regime shift in positioning rather than a simple pullback.
Not everyone agrees the damage runs that deep. Dan Ives argued the sell-off is "way overdone," contending that while 5% to 10% of data centers may not end up happening, deals are still accelerating and the demand-supply ratio stands at 13-to-1. CoreWeave's chief executive said the market is acting "irrationally." Fed Chair Kevin Warsh, speaking at the IMF and World Bank annual meetings in Bangkok, declined to signal the Fed's next move but said the global economy has shown surprising resilience despite elevated energy prices linked to the war in the Middle East. Warsh has vowed not to steer markets with his own near-term outlook for monetary policy, a departure from the forward guidance investors grew accustomed to under his predecessor.
The strongest counter-thesis, then, is that this is a liquidity event, not a solvency event — that the AI demand is real, the hyperscalers are still spending, and the shorts are simply harvesting a technical overshoot. That view has history on its side: every major drawdown in this bull market has been bought, and the earnings backing the AI trade are real, not hypothetical.
But there is a falsifying signal. Watch CoreWeave's credit spreads and the next round of hyperscaler capital-expenditure guidance. If credit-default-swap spreads widen beyond current distress levels and at least two of the major cloud providers cut 2027 capex guidance, the "recalibration" thesis breaks and the structural-downgrade case takes over. Conversely, if capex guidance holds and spreads tighten on any Fed pause signal, the cyclical-bounce case wins.
What Comes Next
In the short term, expect volatility to remain elevated into the October 27-28 Federal Reserve meeting. The 85% probability of a hold is not a certainty, and any hint of a December hike will keep pressure on duration-sensitive names. Technical support matters: the semiconductor index touched a key support level this week, and breadth is stretched — conditions that often precede a bounce, but do not guarantee one.
Over the medium term, the earnings season ahead is the real test. If hyperscaler spending holds and memory pricing stays firm — Micron just reported record fiscal fourth-quarter revenue of $54.23 billion — the fundamental case survives and the selloff looks like a cyclical gift. If guidance wavers, the multiple compression has further to run.
In the long run, the structural question dominates: does AI generate enough revenue to justify the capital being deployed? The answer will not come from a technical bounce. It will come from the income statement.
Base case: the Fed holds in October and signals a December pause, and the semiconductor complex stages a technical rebound that recovers roughly half of this week's losses. Upside case: hyperscaler capital-expenditure guidance comes in firm, inflation cools, and the AI trade resumes with shorts forced to cover. Downside case: December hike odds rise, CoreWeave-style credit stress spreads to other neocloud borrowers, and the SOX retests its spring support levels.
The week's lesson is simple: short sellers did not break the chip trade by themselves. They cashed in because the market finally gave them a reason to.
Explore more exclusive insights at nextfin.ai.
