SMH Was the One of the Best-Performing Non-Leveraged ETFs of the Last Decade. Here Is the Concentration Risk Investors Miss

SMH posted a 31.34% annualized 10-year return through March 31, 2026, closely tracking its benchmark at 31.45%, but the article emphasizes growing concentration risk. Nvidia now represents 18.57% of the fund and Taiwan Semiconductor 10.63%, leaving nearly one-third of assets in just two names. The piece argues that XSD’s equal-weight structure may offer more balanced semiconductor exposure, though its 10-year return of 22.62% trails SMH.

Analysis

The key takeaway is that the ETF is no longer a clean “semis beta” vehicle; it is increasingly a leveraged proxy for NVDA/TSMC sentiment and positioning. That matters because when ownership becomes this top-heavy, the marginal buyer is really underwriting a very short list of earnings and capex outcomes rather than the breadth of the semiconductor cycle. In practice, that means headline index strength can mask much higher single-name factor risk, especially around earnings, guidance, and export-policy headlines.

The second-order effect is that concentration itself can become a volatility amplifier. If NVDA disappoints or TSMC signals a pause in AI-related capex, ETF flows can turn mechanically one-way as risk-parity, CTA, and retail holders all de-risk the same exposure at once. That creates a sharper path-dependent drawdown than the backward-looking return profile suggests, with the highest risk window concentrated over the next 1-3 earnings cycles rather than over years.

The contrarian angle is that equal-weight semis may outperform not because the leaders fail outright, but because the market has already priced an unusually narrow AI capex regime. If the buildout broadens into analog, equipment, packaging, memory, and design tools, breadth should improve and a cap-weighted basket will lag as winners are repeatedly re-rated to perfection. In that scenario, the “cheaper” way to own the theme is not the lowest fee but the structure with less embedded concentration risk.

For now, the asymmetric setup is to fade concentration while keeping the secular trend exposure intact. The best risk/reward is not a blanket short semis; it is a relative-value expression that benefits if leadership narrows less than expected or if the dominant names simply mean-revert on elevated expectations.

Key Decisions for Investors

• Long XSD / short SMH as a 3-6 month relative-value trade: thesis is breadth rotation and lower concentration risk; stop if NVDA/TSMC continue to outperform the group by >10% on a rolling quarter.
• Buy NVDA downside protection into earnings via 1-2 month put spreads rather than outright shorting: concentration makes the ETF vulnerable to one name, and vol is likely underpriced versus the index’s top-heavy exposure.
• If long semis structurally, prefer a barbell of XLNX-equivalent quality in equipment/design-adjacent names plus a smaller NVDA position, instead of an SMH core; this lowers single-name drawdown risk while preserving AI upside.
• For hedged portfolios, consider shorting SMH against SOXX/XSD or a broader tech basket over the next 1-2 earnings seasons; the trade benefits if leadership broadens or if NVDA/TSMC face any guidance reset.
• Set a tactical alert around NVDA earnings and major TSMC guidance: use any post-event selloff to initiate longs in equal-weight semis, since dispersion spikes usually create better entry points than chasing momentum.