Semiconductor ETFs: SMH vs. SOXX — Buy the Global Chip Leaders in One Ticker

Don't want to bet on TSMC alone, but want a stake in the whole global chip industry? SMH and SOXX are the two purest shovels.
But behind the high returns lie the inventory cycle's boom-and-bust, a structural cycle AI can't erase, and a leveraged ETF's volatility decay — this piece also shows you how to use it correctly as a satellite.

2026.07.18 ｜ Shiba the Disciplined ｜ ProfitVision LAB ｜ Overseas ETF Series · Part 6

1. Taiwan has semiconductor ETFs, but these two are "pure global semiconductor industry"

Taiwan of course has semiconductor-related ETFs, but they lean toward the Taiwan-stock supply chain with a different weighting. If what you want is exposure to the "global chip leaders" — NVIDIA, TSMC (ADR), Broadcom, ASML, AMD… — then SMH and SOXX are the purest and most liquid shovels.

These two let you hold, under one ticker, the most critical companies in global design, manufacturing, and equipment. In other words, you're not just betting on TSMC, but on the "whole chip ecosystem" — from NVIDIA's GPU design, to TSMC's foundry work, to ASML's EUV equipment, all in one.

Data: StockAnalysis / Morningstar. AUM, avg daily volume, 1-year return, and holdings weights are as of 2026-07-17; 3/5/10-year same-period annualized figures are as of 2026-06 (comparison logic unaffected). 1-year is total return. We deliberately use 3/5/10-year same-period windows, not "since-inception annualized": because SMH starts in 2011 (an all-bull window) and SOXX in 2001 (including the dot-com crash + GFC), the since-inception gap (SMH ~30%, SOXX ~15%) is almost entirely inception-date bias and not directly comparable; over the same period the two are actually close (consistent with r > 0.95). Approximate; verify current figures.

2. The cyclical structure of semiconductors — why so boom-and-bust?

Semiconductors are one of the most cyclical of all major tech industries. To understand this, you first have to grasp the logic of the "inventory cycle" — the root cause of semiconductor ETFs' high volatility. It's not bad luck; it's a structural pattern.

The cycle is most extreme in memory chips (DRAM, NAND Flash): peak-to-trough prices can differ by more than 80%. Take DRAM: in 2021 a shortage drove prices sharply higher; by 2022–2023, as cloud providers and phone makers worked off stockpiled inventory, DRAM spot prices fell more than 70%. These swings aren't an exception — they're a structurally inevitable cycle.

The logic is clear: building a new fab takes 3–5 years, and the chips produced are highly standardized — demand overheats a little, every maker ramps at once, and by the time the wave of supply arrives demand has already faded, followed by oversupply and order cuts. This "bullwhip effect" is especially severe in semiconductors, because the supply side's response time is very long and its costs very high.

Does AI lengthen the cycle? Or remove it?

The AI wave of recent years drove an explosion in GPU and HBM-memory demand, and many ask: is it different this time? AI demand is structural — is the cycle gone?

The answer: AI may lengthen the upcycle, but it can't remove the cycle. Keep two things separate:

• AI is structural demand: large language models, inference acceleration, data-center expansion — these are 10-year-plus trends, and NVIDIA's moat is real.
• But the cycle is another matter: even with the long-term trend up, there will still be "over-order → inventory digestion → order cuts" loops along the way. By late 2024–2025 there was already discussion of AI-server GPU over-ordering: cloud giants, racing to position early, bought ahead of actual demand, and subsequent order momentum may slow in phases.
• The AI wave will also have deep drawdowns mid-trend: even if the final direction is right, a holder can be halved on paper at some cycle reversal — then shaken out, missing the rebound that follows.

"Structural demand" and "no short-term drawdowns" are two completely different things. E-commerce was the most certain structural growth of the past 20 years, yet Amazon fell 95% in the 2000 dot-com bust. Chips' long-term trend is no get-out-of-jail card; the reality of the inventory cycle won't change because of a narrative.

3. SMH's three holding models — you actually hold three kinds of company at once

Many assume semiconductors are a homogeneous industry. They aren't. SMH's and SOXX's holdings span three completely different business models: Fabless, IDM, Foundry. Each behaves very differently across the cycle.

Grasp this taxonomy and you understand SMH's risk structure: you're simultaneously betting on Fabless's high-growth high-volatility, IDM's turnaround risk, and Foundry's geopolitical sensitivity. TSMC's ADR is usually SMH's #1 or #2 holding — and almost all of TSMC's leading-edge manufacturing is on the island of Taiwan.

4. That scary one-year number hides the most dangerous misunderstanding

Seeing the 1-year return, many get fired up — stay calm. This is an AI-data-center-mania (NVIDIA-led) "extraordinary year," not the norm, and definitely not extrapolatable.

Semiconductors are a classic high-beta, high-volatility cyclical industry: explosive on the way up, brutal on the way down. The same-period 5/10-year annualized is impressive (both in the 33%–37% range), but it was "bought with large volatility" — the drawdowns along the way far exceed a broad-market ETF's. Reminder: don't compare SMH and SOXX by "since-inception annualized" — SMH counts from the 2011 bull, SOXX from 2001 (including two crashes); that gap is inception-date bias (see Part 4), and over the same period they're actually close.

A high long-run annualized doesn't mean "comfortable to hold." A semiconductor ETF suits someone who is bullish on the long-term chip trend and can stomach a single-year halving. Its role is to participate in the trend; how much to allocate depends on your situation — but don't treat it as a savings account and chase it, or sooner or later a cycle reversal will shake you out.

5. The SMH vs. SOXX decision framework — are they really that different?

SMH or SOXX? The bottom line first: their correlation is r > 0.95 and same-period (3/5/10-year) returns differ little — which one you pick matters far less to the long-run outcome than "when you buy and how much." But open up the top-10 holdings and you'll see two different design philosophies.

What do they actually buy? Top-10 holdings laid out

Weights are an approximate snapshot that floats daily with market value; rely on StockAnalysis / the issuers' latest disclosures (as of 2026-07-17). SMH ~26 holdings, SOXX ~34. Versus a month earlier: SMH's NVIDIA weight rose from ~16% to ~20.7%, concentration up further; SOXX's NVIDIA weight also rose from ~5.9% to ~7.6%, and TSMC and Marvell newly entered its top 10 (displacing Texas Instruments, Qualcomm, and ADI).

The risk profile of the three: Beta, volatility, standard deviation

"High volatility" is a feeling; to compare it you need numbers. Put the three in one table, with the S&P 500 (VOO) as the baseline, and the risk differences are clear at a glance:

Beta is StockAnalysis's latest published figure (as of 2026-07-17); annualized volatility and the 2022 max drawdown are trailing 3-year, vs. S&P 500, from 2026-06 data, source StockAnalysis / Morningstar. The leveraged fund's (USD) Beta/volatility vary greatly by measurement window — for order-of-magnitude reference only.

6. Leveraged ETF (USD) — not a "guaranteed loser," but a path-dependent satellite tool

There's also USD (ProShares Ultra Semiconductors), billed as "2x semiconductors." Many think it's a "doubled SMH," but that's a fundamental misunderstanding.

What is "2x daily leverage"?

USD is designed to reset its leverage daily: at today's close, it recalculates so that tomorrow's move is exactly 2x the underlying's. This has one mathematically unavoidable consequence — volatility decay.

But "guaranteed to lose" is a textbook misreading — it only holds in a "choppy market." The decay above comes from the round trip of "rise then fall back to start"; if the underlying instead runs a low-volatility sustained uptrend, daily 2x compounding actually gives you "more than 2x" the cumulative return. A leveraged ETF is a path-dependent tool: eroded sideways, amplified in a trend. The real problem isn't "mathematically certain to lose," it's that you can't predict whether the future is choppy or trending — so it's a tool, not a faith.

Essentially, USD works the same as Taiwan's 00631L (Taiwan 50 2x), just on a more concentrated, more volatile underlying. Issuer ProShares positions it as a "short-term directional trading tool" — meaning not "untouchable," but "not for buying mindlessly and holding long, set-and-forget."

Note: all three are U.S.-domiciled ETFs, and distributions are withheld at 30% — but the yields are inherently tiny (SMH ~0.2–0.4%, SOXX ~0.3–0.5%, USD ~0.2%), so the annual drag is only ~0.1%, effectively imperceptible; their returns come almost entirely from price, not distributions. The full tax mechanics (including estate tax and AMT) are in Part 3.

Frequently Asked Questions