Business

Know the Business — iShares Semiconductor ETF (SOXX)

SOXX is not a company; it is a packaged, modified-cap-weighted slice of US-listed semiconductor stocks, sold by BlackRock for a 0.34% annual fee. Two businesses are stacked on top of each other — the issuer's high-margin fee royalty on AUM, and the investor's NAV-level exposure to the chip cycle — and the most common analytical mistake is to value either one as if it were the other. The market most often underestimates how much of SOXX's long-run return comes from a single source (the chip cycle filtered through ~30 names) and overestimates how meaningfully methodology choices differentiate one semiconductor ETF from another over a full cycle.

1. How This Business Actually Works

There are two revenue engines stacked in this product, and they answer very different questions for the analyst.

BlackRock's slice — $47.4M of advisory fees in FY2025 (implying ~$13.9B average daily AUM during the fiscal year, with period-end net assets at $10.8B), scaling toward roughly $70M run-rate at the FY2026-end AUM of $20.6B — is a near-pure royalty on the size of the chip-sector wallet. Authorized Participants create and redeem 50,000-share Creation Units in-kind; BlackRock never touches a buy or sell decision because the index does the work. That makes the wrapper economics genuinely high-incremental-margin: each new dollar of AUM adds 34 bps of fee revenue at near-zero variable cost. The investor's slice is the inverse of that scaling — when AUM grows because chip stocks rallied 76% in twelve months, the investor experienced the rally and paid 34 bps for the experience; when chip stocks fall 35% (FY2022), the investor still paid 34 bps and got the drawdown.

The bottleneck for the issuer is flows + price, not operations; the bottleneck for the investor is the chip cycle, not BlackRock's execution. Bargaining power on the wrapper sits with whichever issuer can offer the lowest credible fee at scale (BlackRock has scale; Invesco's SOXQ at 0.19% has the price). Bargaining power inside the index sits at the leading edge of chip technology — TSMC in foundry, ASML in EUV, NVIDIA in AI accelerators — but SOXX's US-only mandate caps its exposure to two of those three.

2. The Playing Field

The competitive set is six US-listed pure-play semiconductor ETFs, sorted along three axes — fee, methodology, and what the investor is really buying.

Three things stand out. First, scale is bimodal — SMH and SOXX hold roughly 90% of pure-play US-semi-ETF assets, and the four sub-scale rivals each carry a fee penalty (PSI, FTXL, USD) or a methodology niche (XSD's equal-weight, USD's leverage). Second, methodology genuinely matters in any single year, but tends to wash out over a full cycle — SMH's 20%-cap rule lets it ride NVDA harder when mega-caps lead, while SOXX's 8%-cap rule makes it the better breadth vehicle when the rally widens; over a decade, both clear ~28% per year with similar shape. Third, the only structurally threatening competitor is SOXQ — same legacy index SOXX used to track, 15 bps cheaper, scaling from a small base. SOXQ has not yet hit the brand inflection point at which BlackRock would be forced to cut, but it is the "good enough cheaper alternative" that fee-sensitive allocators eventually find.

What "good" looks like in this arena is unusually concrete: high AUM (liquidity + scale economies), tight median bid-ask (SOXX is at 0.01%), fee at or below 35 bps for cap-weighted exposure, and tracking error inside the expense ratio. SOXX clears every bar except the fee bar, where SOXQ now sets the price.

3. Is This Business Cyclical?

Profoundly. The wrapper itself is steady — BlackRock collects 34 bps regardless of weather — but the AUM the wrapper sits on is one of the most cyclical asset bases in public markets, and that cycle is how investors experience SOXX.

The pattern that matters: every five-year window contains at least one drawdown of 25–50% on the underlying index, followed by a recovery that more than restores the prior peak. The 10-year CAGR through 3/31/2025 was 20.9%, but that path included a -16% year, a +109% year, two -35%-class drawdowns measured peak-to-trough, and a 17-day +42% winning streak in spring 2026. SOXX is not a "growth allocation" in the buy-and-hold sense; it is a high-volatility cyclical that has happened to compound at extraordinary rates since 2016 because the AI cycle layered on top of the secular cycle.

4. The Metrics That Actually Matter

For an operating company an analyst tracks revenue growth, ROIC, and FCF; for a sector ETF the right metrics are entirely different. Five matter — three about the cycle the fund tracks, two about whether the fund is doing its job.

These dominate the usual ratios because none of yield, P/B, or expense ratio in isolation contain information about the chip cycle. A 30 bps yield is meaningless because semiconductor companies pay almost no dividends; the return is capital. A holdings-weighted P/B of 6.9x looks rich but says little when the underlying earnings are mid-cycle. The single most important number for a SOXX investor is the holdings-weighted P/E in context of cycle position — at 42x today, the fund is priced for AI-cycle continuation, not for a normal-cycle year. The single most important number for the wrapper is tracking error, not fee, because a 0.34% fee is irrelevant if the fund drifts 80 bps per year from its index.

5. What Is This Business Worth?

The right valuation lens for SOXX is the underlying chip-cycle earnings power, adjusted for where in the cycle you are buying — not a price target on the ETF itself. The wrapper has no independent value to the investor; it only has a cost (34 bps) and a delivery quality (tracking error of 30 bps in FY25). Every dollar of SOXX value comes from the holdings, so what is being underwritten is the weighted P/E and earnings trajectory of about 30 chip companies.

Sum-of-the-parts is the right lens, but not in the usual way. The "parts" are not segments of one company — they are roughly 30 listed equities each with its own valuation, and the fund's NAV is the daily mark-to-market of those parts. There is no information edge in valuing SOXX itself; the edge sits in deciding whether the weighted P/E of 42x on the underlying constituents is reasonable for where in the cycle you are. If you believe NVIDIA, Broadcom, and Micron are fairly valued, SOXX is fairly valued. If you believe consensus AI capex projections are too high, SOXX is roughly 15–25% expensive. The wrapper does not change that calculation; it only collects 34 bps from it.

What would make this fund cheap: a chip-cycle correction that takes holdings P/E back to 22–28x while AI capex normalizes rather than collapses, leaving the fund as a high-quality re-entry point at the bottom of a cyclical drawdown. What would make it expensive: continued multiple expansion above 50x without earnings catching up, the kind of late-cycle setup that historically precedes 30%+ drawdowns within 18 months.

6. What I'd Tell a Young Analyst

Five things to internalize before writing a memo on SOXX.

First, never confuse the wrapper with the asset. The fund is an excellent wrapper around a violently cyclical asset. The wrapper has a small cost (34 bps) and a reliable delivery (-30 bps tracking); the asset can move 50% in either direction in twelve months. Most analytical errors come from analyzing the wrapper carefully and the asset lazily.

Second, watch the SMH/SOXX rolling spread as a breadth signal. When SOXX leads SMH, breadth is widening — historically a constructive mid-cycle signal. When SMH leads materially, the rally is mega-cap-narrow, which has historically been late-cycle. This is a cheaper telltale than counting hyperscaler capex announcements.

Third, track SOXQ's AUM growth, not its price. Invesco's 0.19% fee is the existential risk to SOXX's pricing power. SOXQ at $1.65B today is a rounding error; at $10B+ it would force a BlackRock fee response and compress the issuer's economics meaningfully. The inflection point is the watch.

Fourth, the holdings-weighted P/E is the single best valuation discipline. At 42x, the fund prices in continued AI-cycle earnings expansion. At 25x, it prices in a recession. At 18x, it prices in a depression. The history of semis is that the multiple range is wide and mean-reverting; entries below 25x have produced exceptional 5-year returns, entries above 40x have produced ordinary ones with much worse drawdowns along the way.

Fifth, the catalyst that would change the thesis is structural, not cyclical. A cyclical drawdown is the base rate; you can plan for it. The thesis-changers are: (a) loss of AI capex as the dominant marginal demand driver without something else replacing it, (b) a Taiwan-strait disruption that re-rates TSM and the entire foundry chain, or (c) a fee-driven flow shift from SOXX to SOXQ large enough to break BlackRock's 34 bps royalty. None is the consensus view today, which is exactly why each is worth the weekly half-hour.