FEATURED

The Philadelphia Semiconductor Index capped a 101% surge for the first half of 2026, the steepest climb since the peak of the dot-com bubble in 1999. The second quarter alone produced an 87.75% gain, the largest quarterly move on record for the index. Then this week happened.

Semiconductors fell for a second day in a row on Thursday. The VanEck Semiconductor ETF dropped 4.5%, led by a 13.6% decline in Teradyne and an 11.5% slide for KLA. Nvidia shares pulled back 1.4%, while Micron lost 5.5%. The Roundhill Memory ETF is tracking to end the shortened holiday week down nearly 15%.

Here’s where I’m at on this. The sell-off is real. The rotation is real. The AI thesis is not broken, but the leverage inside the trade is. That is a different problem than people are treating it as.

What happened

The surface-level explanation is valuation fatigue after a historic run. The VanEck Semiconductor ETF had climbed over 80% year-to-date through June 30, and the pullback reflected investors locking in gains after that extraordinary run. The Philadelphia Semiconductor Index closed at 14,246.96 on June 30, capping a first half that dramatically outpaced the Nasdaq Composite’s 12.79% gain and the S&P 500’s 9.55% rise over the same period.

But the bigger driver is structural, not fundamental. The semiconductor trade became the most leveraged, most crowded sector position in recent equity market history. When sentiment turned even slightly, forced selling took over. What should be a normal pullback turns into something that looks much more alarming than the fundamentals justify.

Levered ETFs like SOXL compound the move on the way down. When the index drops, these products have to sell into weakness to maintain their leverage ratios, which pushes prices lower, which triggers more selling from the next fund. The businesses haven’t changed. The plumbing is what’s amplifying the noise.

Where the money went instead

What’s interesting is where capital moved. Jeff Kilburg of KKM Financial told CNBC: “The ‘Great Rotation’ trade persists into the third quarter as the blue boring names of the Dow Jones Industrials continue to attract inflows directly from recent profit-taking money from tech stocks. This is extremely healthy and underscores the broadening breadth of equities for this continued bull market in its fourth year.”

The Dow scaled to a record close on Thursday. The 30-stock average added 594.83 points, or 1.14%, for an all-time closing high of 52,900.07, touching an intraday record of 52,903.85. The catalyst: a June jobs report that missed badly, with only 57,000 jobs added versus a consensus expectation of 115,000. The unemployment rate dipped to 4.2% from 4.3%. Soft labor data pushed rate-cut expectations higher and sent money into cheaper, rate-sensitive names.

The S&P 500 was essentially flat on Thursday, while the Nasdaq dropped 0.8%. Most of the market is fine. The AI chip trade is being re-evaluated on valuation. Those are not the same thing.

What the AI thesis actually requires

The four largest hyperscalers, Google, Microsoft, Amazon, and Meta, are collectively guiding to roughly $725 billion in combined capital spending for 2026. That is up 77% from approximately $410 billion in 2025, with the overwhelming majority going to AI data centers, Nvidia GPUs, custom silicon, and power infrastructure.

That number has not changed. The capex cycle that drove the semiconductor rally is not turning. What is turning is the willingness to price in that cycle at 80-100x earnings multiples with no margin of safety. As Anshul Sharma, CIO at Savvy Wealth, put it: “This is a rotation potentially out of a sector that’s been red hot for the last few months and into other areas, but I also do think that there’s a little bit of a revaluation of the AI trade in itself.”

Those two things, a healthy spending cycle and a necessary valuation reset, can coexist. They usually do.

The part people are skipping

Software stocks are regaining favor as capital rotates out of semiconductors after an extended run, with fundamentals remaining robust despite recent sell-offs. This is the second-order trade. The SaaSpocalypse punished software names earlier this year on AI disruption fears. Now some of those same fears are showing up as cost concerns inside the hyperscaler capex model. If AI inference efficiency improves faster than raw compute demand grows, the chip-to-software ratio shifts. The dollars don’t disappear. They move.

Slight tangent, but it matters: Micron’s most recent quarter, reported in late June, came in at $41.5 billion in revenue with gross margins of 84.9% and EPS of $25.11, blowing past Wall Street’s consensus of $35.9 billion in revenue. The company said memory demand will exceed supply beyond 2027. That’s not a collapsing cycle. That’s a cycle with a bid underneath it.

Bull / Base / Bear

• Bull: Chips stabilize around current levels, the leverage flush completes by end of July, and Q2 earnings from Nvidia and AMD confirm the H2 data center order book is intact. SMH retraces toward prior highs by August.
• Base: Semis trade in a wide, choppy range through July earnings season as the market waits for HBM demand confirmation and Fed rate clarity. Rotation into industrials and healthcare continues but doesn’t become a rout in tech.
• Bear: Hyperscaler capex guidance softens in Q2 earnings calls, inflation proves sticky enough to delay Fed cuts further, and the leveraged ETF unwind extends. The Philadelphia Semiconductor Index gives back 30% or more from its June highs.

Memory and storage companies led the first-half rally, with SanDisk soaring 857%, Micron jumping 300%, and Intel climbing 257%. Those gains are not coming back in a week. The question is whether investors treat this as a needed exhale inside a structural spending cycle, or the beginning of something worse.

The spending cycle is real. The leverage inside the trade was not sustainable. Those are two different sentences, and keeping them separate is how you avoid the wrong trade on both sides of this.