FEATURED

Bank of America strategists are using the kind of language nobody loves to hear while they’re up money: the Philadelphia Semiconductor Index (SOX) is pushing into “bubble-like” territory, with volatility they claim is worse than the dot-com era.

That’s not a call that chips are “bad.” It’s a warning that the market is paying up for a very specific future – and it’s starting to pay up in a way where the path matters more than the destination.

Scoreboard (what happened)

The simple version: semis have been the market’s engine, and the index construction makes the engine feel even bigger than it is.

• SOX total return in 2025: +43.5% – about 26 percentage points ahead of the S&P 500.
• After an ugly first quarter, SOX hit an April 2025 trough at -28% YTD before rebounding hard into year-end.
• Within the index, 2025 returns were wildly uneven: Micron +240%, Lam Research +139%, Nvidia +39%, while Texas Instruments was -4%.

Quick tangent that matters: whenever a sector becomes “the place to be,” people stop buying the sector and start buying the easiest access point. That’s usually an ETF. Then the ETF’s biggest names get the most incremental dollars. Not because the fundamentals changed that day – just because flows don’t read footnotes.

If you’re using SOXX as your proxy, here’s what you’re actually buying right now.

• SOXX net assets: about $20.59B (as of 03/31/2026)
• Expense ratio: 0.34%
• Portfolio P/E: 42.29x; P/B: 6.87x
• 3-year beta: 1.58; 3-year standard deviation: 26.87%
• Top weights: NVDA 8.40%, AVGO 8.27%, MU 6.99%, AMD 6.47%, AMAT 5.84%, MRVL 5.17%, INTC 4.13%, KLAC 4.12%, MPWR 4.08%, TER 3.95%
• Top 10 = 57.42% of the fund

(Those figures are from the iShares SOXX fact sheet through March 31, 2026.)

The real reason the “bubble” word shows up

“Bubble-like” usually isn’t about one metric. It’s about a cluster of conditions that tend to appear together:

• Prices move faster than fundamentals for a sustained period (not a week – months).
• Dispersion gets weird: a few names become the whole story while others in the same industry quietly lag.
• Volatility rises even as prices rise – the market keeps bidding for upside, but it also pays up for protection.
• Expectations compress the range of acceptable outcomes: companies can still beat… and sell off.

That last one is the killer. The market stops rewarding “better.” It demands “perfect.” And when perfection is the bar, even great quarters can feel like disappointments.

The BofA point (as summarized widely in finance press) is that semiconductor volatility readings are brushing levels associated with the late 1990s/2000 period. I can’t quote their proprietary note, but the risk framing is familiar: the more crowded the winners, the more violent the downdrafts when positioning flips.

Deep dive: what you’re actually investing in when you “buy semis”

Semiconductors aren’t one business. It’s a supply chain with different economics at each layer.

• Compute accelerators / GPUs: Nvidia (NVDA) – high margins, very high expectations, demand tied to AI infrastructure cycles.
• Networking + custom silicon + infrastructure software: Broadcom (AVGO) – partly a chips story, partly a software cash flow story.
• Memory: Micron (MU) – brutally cyclical, but AI has created a high-value pocket (HBM) that can temporarily change the earnings profile.
• Foundry manufacturing: Taiwan Semiconductor (TSM) – the capacity gatekeeper, economics tied to utilization and leading-edge nodes.
• Equipment “toll collectors”: Applied Materials (AMAT), Lam Research (LRCX), KLA (KLAC) – benefit when the industry builds, upgrades, and debugs factories.
• Analog / industrial semis: Texas Instruments (TXN), Analog Devices (ADI) – slower growth, often better durability, usually less “AI premium.”

It’s tempting to treat all of these as one trade. They’re not. Their profit drivers don’t even peak in the same year.

Data section (numbers that matter more than vibes)

Start with the index itself. A few underappreciated details help explain why this group can feel “unstoppable” for long stretches… and then suddenly not.

• Index construction: SOX is modified market-cap weighted with caps on the biggest weights (top three capped at 12%, 10%, 8%; others capped at 4% at quarterly rebalance).
• Concentration: the 10 largest constituents were 58.3% of index weight (very similar to SOXX’s top-10 share).
• Fundamentals did improve: SOX constituents’ sales per share rose 44% over the past three years (as of 12/31/2025).

Now, the part people skip: when valuations get stretched, you don’t need earnings to fall for stocks to fall. You just need earnings to rise… less fast than the market already paid for.

SOXX’s own portfolio-level valuation gives you a quick gut check. A 42.29x P/E for a sector ETF is not “cheap.” Even if you argue AI changes the long-run demand curve, a lot of that is already being capitalized today.

Is it cheap? (not the companies – the price you’re paying)

Let’s be blunt, bargain hunter: at the ETF level, semis are priced like a premium category right now. That doesn’t mean they’re doomed. It means your margin of safety is thinner and your timing risk is higher.

When I see SOXX at ~42x earnings, the first question I ask isn’t “will AI be big?” It’s “what’s the probability the next two years of earnings land above what’s already implied?”

And that’s where stock selection matters. Some names have a second engine (software, recurring services, consumables). Others are more exposed to a single spending cycle.

Specific stocks: who looks fragile vs who looks sturdier

I’m going to frame this the Cheap Investor way: not “good company/bad company,” but what has to go right for the current price to look reasonable.

• Nvidia (NVDA) – In SOXX it’s the biggest weight at 8.40%. The bull case depends on AI infrastructure staying in heavy-build mode. The risk is that growth stays strong but decelerates, and the stock responds poorly because expectations were sky-high.
• Broadcom (AVGO) – Also huge in SOXX at 8.27%. The appeal here (relative) is diversification: semis plus a meaningful software segment. Still, it’s a big “AI spend stays hot” beneficiary through networking/custom silicon.
• Micron (MU) – 6.99% weight in SOXX and up +240% in 2025. Memory can go from hero to zero fast if supply loosens. The key question is whether HBM tightness stays structurally tight or simply “tight for a while.”
• AMD (AMD) – 6.47% in SOXX. The stock is leveraged to gaining share in data center compute and GPUs/accelerators. The risk is competitive intensity plus customer concentration in hyperscalers.
• Applied Materials (AMAT) and Lam Research (LRCX) – AMAT is 5.84% of SOXX; LRCX was up +139% in 2025. Equipment tends to do well when fabs are being built and upgraded. The risk is capex digestion – a pause after a buildout can hit orders hard.
• Texas Instruments (TXN) – It was down -4% in 2025 in the SOX top-10 table. That underperformance can be a feature, not a bug, if you’re trying to avoid the most crowded AI exposure.

If you want one simple rule: the more a stock’s recent multiple expansion depends on “AI capex never cools,” the more sensitive it is to even a mild disappointment. Not collapse. Just disappointment.

Bull / Base / Bear

Bull: AI infrastructure spend keeps compounding, and the supply chain stays constrained enough that pricing remains rational. Earnings rise fast enough to justify today’s premium multiples. In that world, the “bubble” talk fades because fundamentals catch up.

Base: Demand stays strong, but the pace slows. Not a crash – just a step-down in growth rates. The sector can still do fine, but the leaders become more stock-picker territory: great quarters are required to keep prices elevated.

Bear: The industry hits a digestion phase: hyperscalers pause, inventory normalizes, or geopolitics creates real frictions in supply and end-demand. In a premium-valuation environment (again: SOXX ~42x P/E), price declines can happen quickly even if companies remain profitable.

Action plan (conservative, cost-conscious)

I’m going to keep this practical.

• If you own a broad semi ETF (SOXX/SOXQ/SMH): consider trimming to a size you can hold through a 25–35% drawdown without panicking. Semis can do that on a normal cycle.
• If you want exposure but hate buying peaks: scale in slowly (3–6 buys over time) instead of swinging all at once. Let the market’s mood swings work for you.
• If you own single names that ran hard (NVDA, MU, LRCX, AMAT): separate “I believe in the company” from “I’m comfortable with the price.” You can love the business and still reduce position risk.
• If you’re hunting for less crowded semis: look at the parts of the supply chain that benefit from long capex cycles but aren’t pure GPU sentiment magnets (select equipment, test, and analog). That won’t be as exciting. That’s the point.

And yes, I know. Sitting on your hands while chips rip higher is emotionally awful. But paying any price for a great company is how you turn a great company into a mediocre investment.

Cheap Investor checklist (track these 8 items)

• SOXX valuation drift: does the ETF P/E stay around the low-40s, or fall because earnings rise (good) vs prices fall (less fun)?
• Concentration risk: do top weights keep creeping up, or does breadth improve? (SOXX top-10 is 57.42% today.)
• Memory cycle signals: Micron’s pricing commentary and HBM supply tightness (watch for signs of supply catching up).
• Foundry utilization: TSM’s node ramp cadence and utilization hints (the whole chain feels it).
• Equipment orders: AMAT/LRCX/KLAC commentary on leading-edge vs trailing-edge demand.
• AI capex language: hyperscalers’ capex guidance (the demand source for the hot part of the chain).
• Relative performance inside semis: does everything rise together, or do only 3–5 names do all the work?
• Risk appetite signal: are investors still willing to pay premium multiples (like SOXX at 42.29x) or do they suddenly care again?

Bottom line

If BofA is right that the SOX has entered “bubble-like” conditions, the right response isn’t to swear off semis forever. It’s to stop pretending you’re buying a sector and admit you’re buying a set of very expensive expectations.

If earnings keep surprising higher, you can still do well here. If growth merely stays good (not mind-blowing), returns can get weird fast. That’s the deal.

Worth a look: pull up your semi exposure and tell me whether you own “diversified semis”… or whether you really own NVDA, AVGO, MU, AMD with extra steps.

– The Cheap Investor