iShares Semiconductor ETF
iShares (BlackRock)
Tracks the NYSE Semiconductor Index (~30 US-listed semiconductor design and manufacturing companies). Cap-weighted but capped, so it spreads exposure more evenly than SMH — no single mega-cap dominates.
Top holdings
Top weight on the memory/HBM rally
AI accelerators and CPUs
Capped far below its SMH weight
Custom AI silicon and networking
CPUs and foundry turnaround
Semiconductor equipment
NAND/memory — AI-driven storage demand (post-WD spinoff)
AI-cluster connectivity (active electrical cables / SerDes)
Performance
July 2026: 21.2% (adj close 640.76 on 30 Jun -> 504.89 on 31 Jul) — Sector's worst month since Nov 2008, and SOXX took it harder than peers: -21.2% vs SMH -17.6%, SOX -20.6% — the cap underweights NVDA (+0.3%) and AVGO (+3.1%) and overweights memory/WFE (MU -28.7%, AMAT -29.8%, LRCX -32.4%, KLAC -39.4%). Peak close 655.01 (22 Jun) to 29 Jul low 465.00 = -29.0%. Assets still ROSE to $44.69B on record inflows (~$6.13B MTD to 20 Jul).
Recent moves
Among the top broad semi ETFs in H1 2026, with outsized gains driven by memory names like Micron.
Our take
The diversified-semi counterweight to SMH — its capped construction means MU/AMD/INTC carry weights similar to NVDA, so it captured the 2026 memory leg better than SMH's NVIDIA-centric book. Lower single-name risk, still 100% one cyclical sector.
The capping is the entire story and it is under-explained by the issuer.
SMH's index lets NVIDIA run to 20.97% (18 Jul 2026); SOXX holds it at 8.49%, level with AMD and just above Micron at 7.72%. That is a rules-based underweight of the largest winner and an overweight of everything else — a breadth bet, not a risk reduction. When the memory shortage lifted Micron sharply in Q2 2026, SOXX carried MU at cap-scale rather than trailing weight and outran SMH by a wide margin. Nothing was selected; the rules did it. Reading 2026 as evidence SOXX is the 'smarter' construction back-fits a mechanism onto one favourable rotation. The reciprocal: when leadership re-narrows into a single accelerator franchise, the same cap becomes the drag. Second point marketing omits — 34 holdings with 60.23% in the top 10 is diversified only relative to SMH. It is one sector, one capital cycle; constituents share customers, fabs and end-demand, and correlations converge in a downturn.
Thesis
SOXX is not a better semiconductor fund than SMH — it is a different bet on the same sector. Its capped index systematically underweights whatever name is winning most, converting the fund into an anti-concentration tilt on chips. That mechanism drove a wide relative gain in H1 2026 because leadership rotated from NVIDIA to memory; it was a drag in 2023–24 when leadership was NVIDIA. Marketing calls this 'diversification'. It is more precisely a factor choice, and it cuts both directions.
Tracks the NYSE Semiconductor Index: the 30 largest US-listed chip designers, manufacturers and equipment makers, modified market-cap weighted with the top five capped at 8% and the remainder at 4%; ADRs capped at 10% cumulative. Rebalanced quarterly, reconstituted annually. The fund held 34 positions vs the index's 30 (20 Jul 2026). Unlevered, 1x, no daily reset.
Assessment
- Cap rule mechanically limits single-name dominance — largest position 8.49% vs NVIDIA's 20.97% in SMH (20 Jul / 18 Jul 2026).
- Wider net than SMH: includes equipment/WFE (AMAT 5.20%, KLAC 4.89%, LRCX 4.35%) alongside logic and memory (20 Jul 2026).
- Unlevered 1x exposure — no daily-reset path dependency or compounding decay, unlike its 3x sibling SOXL.
- 0.34% expense ratio and a 25-year operating history (inception 10 Jul 2001) spanning multiple full chip cycles.
- The 2026 outperformance is a rotation artefact, not a durable edge — the same cap rule underperformed when NVIDIA led 2023–24.
- 'Diversified' is relative only: 34 names, 60.23% in the top 10, 100% one cyclical sector with shared end-demand and shared fabs.
- Weights drift above the stated 8% cap between quarterly rebalances — AMD and NVDA both sat at 8.49% on 20 Jul 2026.
- Since-inception 14.02% annualized is the honest long-run anchor — far below the 113.43% trailing year dominating coverage.
- Holders often conflate SOXX with SOXL (3x daily-reset); volatility decay makes that a structurally different instrument.
Record
Trailing 1-year +113.43% with dividends (20 Jul 2026), YTD roughly +73–76% (17 Jul / 16 Jul readings) — but the fund had fallen more than 20% from its June peak of $655.95 into bear-market territory by 20 Jul. That round-trip matters more than the trailing figure: +113% describes a window that already contains a >20% drawdown. Attribution is largely mechanical — Q2 2026 was a record sector quarter on a memory shortage, and SOXX's cap held MU/AMD at NVIDIA-scale size rather than trailing size. Quarter-level precision (SOXX ~+94% vs SMH ~+70%, MU ~+240%) rests on a single source; read as approximate. Against all of it, since-inception average annual return is 14.02% from July 2001. The gap between 14% compounded and 113% trailing is the most useful pair here — it prices how much of a chip cycle is mean reversion.
Risks & fit
- Memory pricing is the marginal driver of the 2026 record; DRAM/HBM is the most violently cyclical part of semis.
- Leadership re-concentration into a single accelerator franchise turns the cap rule from tailwind to structural drag.
- Single-sector, single-capital-cycle exposure — constituent correlations converge sharply in a downturn.
- Export-control and geopolitical shocks hit designers, equipment makers and TSM exposure simultaneously.
- Valuation compression after a record sector quarter can outweigh earnings delivery, as the >20% drawdown off June highs shows.
Testable: if over the next 12 months NVIDIA outperforms the equal-weighted semi complex by more than ~15pp and SOXX still keeps pace with SMH, the capping explanation for its 2026 lead is wrong and something else drives it. Conversely, an NVIDIA-led leg in which SOXX lags SMH by a magnitude similar to its 2026 win confirms the mechanism is symmetric — neither fund structurally superior, just opposite sides of one concentration trade. Watch also whether SOXX's memory weight is trimmed at the next quarterly rebalance after a DRAM roll-over; that removes the 2026 engine.
Suits an allocator who already wants concentrated semiconductor exposure but prefers it spread across memory, logic and equipment rather than resting on one accelerator franchise. Neutrally: a higher-breadth expression of the same cyclical bet as SMH, not a hedged or defensive one. It does not fit those seeking diversification away from the AI capex cycle, nor those wanting SOXL's daily-reset leveraged exposure — different instruments, different mechanics.
0.34% net expense ratio (20 Jul 2026), roughly at parity with the main US semiconductor ETF cohort. At $45.06B AUM, fee is not the differentiator between SOXX and SMH — index construction is. Dividend yield 0.28%.