I was on broadcast TV this week, and the woman on before me was pretty confident the momentum unwind, particularly in the semiconductor sector-related names, was done. Things I had been lightly studying myself showed me it was done-ish.

Let's look at the tape.

The run-up:

In the the chart below of the PHLX Semiconductor Index (SOX)—the benchmark most investors use as a proxy for "chips"—you can see a steady climb through 2024, a couple of sharp-but-survivable drawdowns in early-to-mid 2025 (a January AI scare, then April's tariff shock knocked the index almost back to where it started) and then, a nearly vertical move from April 2026 onward, after the conflict in Iran cooled off. In numbers, it was up 88% in Q2 2026. Zoom out further and the index was up nearly 97% for the first half of the year. The index roughly tripled in about 10 weeks.

But nothing goes up in a straight line forever, and this one didn't either. After one of the best runs in market history in Q2, a pullback is often what naturally happens. I’ve heard John Roque, Technical expert, quote Bob Farrell, a long-time Merrill Lynch head of Technical Analysis, saying: “Exponentially rapidly rising markets or stocks usually go further than you think, but they do not correct by going sideways."

July

Since peaking in mid-June, the SOX has pulled back roughly 20%, erasing more than a third of the sector's 2026 gains in about 3 weeks. A few things lit the fuse:

• Guidance jitters. Broadcom's AI chip sales guidance came in a bit below expectations, and doubts have crept in about whether AI capital spending will pay off as fast as investors hoped. This set a course of small doubts.

• A less friendly Fed. New Fed Chair Kevin Warsh, confirmed in May after arguing for tighter inflation discipline, had markets pricing in less rate-cut help than they wanted.

• Meta's compute announcement. Reports that Meta plans to build more of its own AI infrastructure in-house rattled the "everyone needs Nvidia's chips forever" thesis, and the SOX fell nearly 7% in a single session on July 2—its sharpest one-day drop since the post-pandemic normalization. This was exacerbated by lower volumes and gamma (see my note from last week).

• A historic SK Hynix drop. SK Hynix, the South Korean memory chip maker, suffered the biggest 1-day share decline in its history, and it didn’t just stay on its shores, hitting chipmakers globally.

What the momentum gauge says

This is one gauge for figuring out whether this is a gut-check or a trend change: relative strength index (RSI).

RSI sits on a scale of 0 to 100 and traditionally, an RSI reading of 70 or above has indicated an overbought condition, while it’s been considered oversold when it's below 30. RSI is currently sitting around 47. So it's well off the overbought territory the index was flirting with in June, but not  in the oversold zone either . The froth has come out, but this isn't (yet) a "panic, everyone's selling" reading (hence my done-ish assessment).

Worth noting from the 1st chart, the SOX broke below its 50-day moving average, but it's still sitting roughly 30% above its 200-day moving average (which is still upward sloping). That's a distinction between a pullback and a trend reversal, and right now it still looks like the former.

Some life in Mag 7?

Here's the part that isn’t getting talked about enough. While the chips themselves went vertical, the Magnificent 7—the hyperscalers and mega-caps actually buying and deploying all that hardware—did the opposite.

The Mag 7's forward P/E currently sits around 25.5x, which is below its own roughly 2-year median of about 29.4x and below 1 std deviation from the last 2 years. Despite AI capex driving skepticism, consensus earnings estimates for these companies have stayed consistent despite lower stock prices. This means on a forward basis, the "demand side" of the AI trade is trading at a discount to its own recent history, even as the "supply side" (chips) is still digesting one of the biggest rallies ever. 

Semiconductors just had a historic run, and July's pullback, with the technical picture (RSI cooling, price still well above the 200-day), supports "healthy correction" over "trend change,". A decisive break below the 50-day is the level to watch next. Meanwhile, the mega-cap side of the AI trade has gotten less expensive relative to its own history, making it worth selectively considering given how much attention the chip drama soaked up.