SMH's 88% Year Finished Last: One Weighting Choice Decided the Chip Fund Race

The race just ended, and the chip fund retail investors know best brought up the rear.

At the U.S. close on August 26, 2025, five ETFs built to own the same thing — the semiconductor boom — each started at 100 paper points. One finish line: the close one year later. One scoring rule: published return across that window, no rebalancing favors, no substitutions. The scoreboard, as published for the twelve months that closed on August 26, 2026:

Fund	12 months through 8/26/26	2026 year to date
VanEck Semiconductor ETF (SMH)	+87.86%	+54.32%
iShares Semiconductor ETF (SOXX)	+106.41%	+71.33%
Invesco Semiconductors ETF (PSI)	+114.47%	+72.98%
First Trust Nasdaq Semiconductor ETF (FTXL)	+123.50%	+72.93%
Xtrackers Semiconductor Select Equity ETF (CHPS)	+134.83%	+74.47%

Read that first line again. An 87.9% year — your money up nearly 88% in twelve months — did not just fail to win. It finished last. CHPS beat SMH by about 47 points: $10,000 turned into roughly $23,500 in the winner and $18,800 in the loser, before taxes, on this return basis.

That gap is not luck. It is the price of a single decision — which rulebook, not which sector.

Race card. Five long-only chip funds, all trading in dollars, all aiming at the same economic question: how to own the AI semiconductor buildout. Start: 8/26/25 close. Finish: 8/26/26 close. Formula: published returns, net of each fund's own fee; yields across the five are under about 0.4%, so dividend treatment can't flip the order; the math is before your taxes and trading costs. Reconstructed handicap, not a pre-published line: concentration, in the form of SMH, entered as the historical favorite — the same bet returned roughly 321% across 2021–2025. The race's open question, which answered itself over the next twelve months, was the width of the rally.

Round 1 — SMH is not a sector, it's a bet on two stocks

SMH does not buy equal slices of the chip industry. It holds roughly 25 of the largest U.S.-listed semiconductor companies, weighted by market value — the biggest business mechanically gets the biggest share. Taiwan Semiconductor and ASML sit inside only because they list here; "the sector" is really "the biggest chip names you can trade in dollars."

As disclosed at this race's finish, NVIDIA is about 17.55% of SMH and Taiwan Semiconductor about 9.29% — a full quarter of the fund riding on two tickers. That was never a hidden flaw; it was the feature. When a handful of giant names lead a sector, a mega-cap-weighted fund is the cleanest way to own the leader. For five years that feature won: the same construction produced roughly 321% across 2021–2025. Then the twelve months that just closed turned the feature into the liability.

Round 2 — the year the lead changed hands

Nothing broke in NVIDIA the company. Revenue was still growing about 83% year over year with gross margins near 74%, according to market data through late August — a monster business. The problem was the stock that owned it. Over the four months through August 27, market data shows Micron gaining about 153%, Marvell 170%, AMD 148%, and Intel 112% — while NVIDIA gained about 28% and Taiwan Semiconductor about 26%. NVIDIA the stock was up only roughly 22% year to date. The two names SMH leans on hardest were the two slowest horses in the year's AI trade.

The money financing the whole trade kept growing: the big cloud providers committed close to $700 billion of spending for 2026; a separate tally by Futurum put the figure between $660 billion and $690 billion, nearly double 2025 levels. What changed was where the marginal dollar landed first. After two years spent cramming the obvious winners full, 2026's re-ratings moved to the chips the trade actually needs next — the memory that feeds every AI server, the custom chips hyperscalers design to reduce their dependence on one supplier, and the value names Wall Street had written off.

Each leg has a story. Memory became a supercycle: SK Hynix listed $28 billion of ADRs on Nasdaq in July, and its U.S. offering was reported about seven times oversubscribed. The memory boom was real earnings, not just multiples — Micron's revenue ran up about 167% year over year with gross margins near 73%, according to market data. NVIDIA's China revenue fell into export-control math after an estimated $8 billion quarterly hit last year, and the CEO is on the record saying the company now has zero market share in China. And Intel, long the sector's punching bag, rallied about 150% year to date, according to market data, after reporting better-than-expected results and foundry momentum.

SMH held Micron, Marvell, AMD, and Intel too. Just not enough of them.

Round 3 — one number explains the finish order

Watch how clean the pattern is. List each fund's disclosed NVIDIA weight against its twelve-month score:

Fund	NVIDIA weight	12-month return
SMH	~17.55%	+87.86%
SOXX	~9.05%	+106.41%
PSI	de minimis (no single-name dominance)	+114.47%
FTXL	~4.79%	+123.50%
CHPS	~3.40%	+134.83%

Over one year in one regime, the less NVIDIA a fund owned, the faster it ran. Three different rulebooks produced that ordering, because all three capped the mega-cap:

• CHPS caps NVIDIA near 3.4% and spreads 54 names across the U.S., Taiwan, Korea, Japan, Europe, and Israel. Its top bets at the finish: Micron 6.82%, SK Hynix 6.80%, AMD 6.46%. The most global, the most memory-heavy — and its expense ratio is just 0.15%.
• FTXL runs a factor screen and weights by cash flow — fundamental size, not market size. NVIDIA sits at a disclosed 4.79%; its heaviest bets are Intel 13.28%, Micron 12.14%, Marvell 7.91%. It charges 0.60%, the most on the card.
• PSI scores the industry on a multi-factor model — momentum, earnings trends, quality, management action, value — then applies modified equal weights so no single stock can dominate. It charges 0.56%, and Applied Materials was its largest holding.

Different machinery, one shared judgment: this rally was not going to be won by the two names everyone already owned.

Round 4 — the counter-evidence, and the rule it forces

The check that keeps this honest: diversification by itself did not win. The most equal-weighted fund in the category, the SPDR Semiconductor ETF (XSD) — 47 holdings, NVIDIA under 3%, near-equal slices — returned only about 84% over the same window, less than SMH.

Why? XSD spread across the whole industry: radio-frequency, analog, consumer, automotive chips — much of it untouched by the AI buildout. So the rule that won in 2026 was not "less concentrated." It was: the same AI trade, minus the mega-caps, plus the corners the market had barely started repricing.

That reframing is the whole lesson. SMH did not lose this year because it is a bad fund. Its construction is a bet on which chips lead, and in a broadening rally the bet on the leaders finished at the back of the field. Concentration is a two-sided ledger: the same rulebook printed 321% in 2021–2025 and gave up about 47 points in the year just closed. The condition changed, not the fund.

Round 5 — scoreboard vs. mechanism board

Final scoreboard: CHPS by about 47 points, with FTXL, PSI, and even the other giant SOXX all ahead of SMH. Mechanism verdict, kept on its own board: the winners' construction worked in this specific regime — a broad AI rally where the biggest names lagged and memory exploded. The mechanism board is also carrying warnings:

• Memory is the most violent cycle in technology, and its own executives are starting to say the obvious. In August, SK Group's chairman told CNBC prices went up too fast.
• Intel, FTXL's single largest bet, rallied on a turnaround while still posting negative operating margins. That leg of the win is priced on what might happen.
• CHPS, the scoreboard winner, is a roughly $90 million fund launched in 2023 and then switched its index methodology in July 2026 — so the printed result was largely produced before the current recipe existed.
• The winners were not cheaper: PSI costs 0.56%, FTXL 0.60%, versus 0.35% for SMH.
• The winners are not nearly as liquid as SMH, either — about $70 billion in assets and tight spreads versus CHPS's roughly $90 million. If you trade in real size, price impact is another cost.
• None of these returns are net of your capital-gains tax. Inside a taxable account, switching from SMH to the new leader hands the IRS part of the 47-point spread.

One crowd-level tell closes the round: U.S. semiconductor ETFs took in about $46 billion of inflows in 2026 — roughly 31% of the whole category's assets and more than double everything the group collected across 2017–2025 combined. Everyone is inside this trade at the exact moment the memory executives are telling you prices overshot.

Final bell. Scoreboard winner: CHPS. Thesis winner: the construction that capped the mega-caps and leaned into the AI second tier — for this regime, not for all regimes. Design lesson in one line: a semiconductor ETF is a rulebook, not a sector, and you are buying whichever rulebook the fund picked before you ever paid a fee. SMH is a concentrated bet on NVIDIA and Taiwan Semiconductor wearing a sector fund's jersey. It wins when leadership is narrow; it loses when leadership widens; neither result is a bug in the software.

If the race renews for the next twelve months, one question decides the card: does the rally stay broad, or narrow back to the mega-caps? Answering that is not predicting chips — it is predicting the width of the crowd. The scoreboard will keep score either way.