Semiconductors are on track for their deepest monthly drawdown since the global financial crisis, and the loudest bull on Wall Street is refusing to blink.
The iShares Semiconductor ETF (SOXX) is down 18.6% month-to-date — on pace for its worst month since November 2008, according to Tradingview data.
The unwind follows one of the most violent rallies in the sector's history. The Philadelphia Semiconductor Index — the benchmark the SOXX ETF tracks — surged 87.8% in the second quarter, a move that left almost no room for disappointment.
Two things spoiled it.
Cost inflation is creeping through the supply chain, especially in memory chips, and investors have started to question whether cloud giants can keep writing bigger checks for artificial intelligence infrastructure.
Chinese models climbing the global performance rankings added a competitive worry on top.
But the pain has a pattern.
Is This Semis Correction Actually Unusual?
Bank of America's Vivek Arya, the firm's semiconductor analyst, counts nine separate drawdowns of more than 10% in the chip index since ChatGPT launched in November 2022.
The average decline was about 14% and lasted roughly 31 days.
The current one is 19% over 24 days — deeper, but faster.
Seasonality explains part of it. The third calendar quarter is historically the sector's weakest, with the chip index underperforming the S&P 500 in 10 of the last 16 years, and by a median 280 basis points between 2010 and 2025.
"We see correction as a summer reset, not a fundamental reversal," Arya said in a Thursday note to clients.
Is Hyperscaler Spending Slowing Down?
Quite the opposite, according to the bank's calculations.
Ahead of second-quarter earnings from Alphabet, Microsoft, Meta and Amazon, the bank now models global hyperscale capital expenditure — the money cloud companies spend building data centers — at roughly $851 billion in 2026 and $1.15 trillion in 2027.
That is 78% and 35% growth, up from the 68% and 25% BofA penciled in after first-quarter results in May.
Crucially, a growing share of that is contracted.
Remaining performance obligations, the accounting term for revenue a company has signed but not yet delivered, now exceed $2 trillion across the top four cloud providers.
Microsoft alone disclosed $627 billion, with only a quarter of it recognized within the next 12 months.
Demand on the ground is not cooling either. BofA's tracker of global token usage — the units of text AI models process, and the closest thing the industry has to a meter reading — has grown an average 9% per week since June.
Where Is the Value?
Valuation has quietly flipped.
According to Bank of America, the chip index now trades at 17.1 times forward earnings against 17.6 times for the S&P 500, a discount of roughly half a turn. Since 2024 it has averaged a premium of about 0.8 times.
Memory is the sharpest edge. It now absorbs 35% to 40% of cloud AI budgets, two to three times historical levels, with third-quarter DRAM contract prices tracking 20% to 30% higher quarter over quarter and spot prices up eight straight weeks.
Arya reiterated Buy ratings across leading compute names Nvidia Corp. (NVDA) and Broadcom Inc. (AVGO), memory maker Micron Technology Inc. (MU), networking plays Marvell Technology Inc. (MRVL) and Credo Technology Group Holding Ltd. (CRDO), and equipment suppliers Lam Research Corp. (LRCX) and KLA Corp. (KLAC).
The bank's $1.7 trillion 2030 AI spending forecast is unchanged.
Fourth quarter and first quarter are historically the two strongest stretches for the chip index.
The bulls are betting the calendar, not the thesis, is what broke.
