In this article
← All articles

Thematic ETFs after the 2026 chip sell-off: memory chips, quantum and the cost of buying a theme late


Defiance's memory-chip ETF lost 49.2% of its launch NAV by 13 November 2026; Amundi's lost 20.4% from its pre-rename NAV to 12 November. Their holdings overlap with a world ETF, so adding a theme can concentrate an existing exposure just as it becomes expensive. A separate net-flow model suggests that purchase timing worsened returns, but does not measure investors' actual accounts. Amundi's fund switched to a memory-chip index on 17 July; Defiance's launched on 18 June, four days before the Philadelphia Semiconductor index set its record.

Quantum, battery, solar and clean-energy ETFs fell 28% to 41% from their 2026 peaks in euros. MSCI World, with net dividends, ended 13 November 0.9% below its October record. What follows measures what these themes added to a world-index portfolio, using issuer NAVs, holdings files and exchange closes. It asks which part of a theme an investor already owns, how purchase timing changes outcomes and what a loss does to the whole portfolio.

Key takeaways

  1. NAV losses are observed; investor returns need actual cash flows. Amundi's assets grew sharply near its peak. Our net-flow model estimates a worse money-weighted result, assuming NAV execution, but does not observe investors' exchange trades.
  2. A late purchase needs a large recovery. Defiance's memory fund needs a 97.0% gain to regain its launch NAV, while MSCI World net rose 0.8% in dollars over the same dates to 13 November.
  3. Most themes moved with the world index they were meant to complement. The VanEck quantum and iShares semiconductor ETFs had weekly correlations of 0.69 and 0.67 with a World ETF, with betas of 2.13 and 2.45.
  4. Expected return matters alongside losses and overlap. An 18-month covariance sensitivity put a 10% sleeve's break-even excess return at roughly 0.0 to 1.8 points a year, depending on the theme, and US research found new specialised ETFs lagged by about 6% a year, risk-adjusted, in their first five years.
  5. What a household can check is overlap and size. A 90/10 mix of a World ETF and a memory ETF held about 11.6% in the memory funds' companies, against 1.75% in the World ETF alone.

What happened to thematic ETFs in the 2026 chip sell-off?

The ETFs that led the 2026 rally gave back between 12% and 49% from their peaks by the latest observations on 12–13 November, while MSCI World stayed within 1% of its record. The Philadelphia Semiconductor index, the usual gauge of chip stocks, closed at a record 14,634.72 on 22 June. On Nasdaq's published closes it fell 23.4% from 23 June to 2 November and stood 30.7% below the record on 13 November.

FE fundinfo's Trustnet reported on 4 November that the index "dropped 21.3% between 23 June and 2 November" while MSCI AC World "lost just 1.2%" in dollars. MSCI's levels give the world figure only from the 22 June close, and we could not reproduce the 21.3%, so we use Nasdaq's closes. Trustnet named quantum, blockchain, battery and clean-energy UCITS ETFs among the 100 hardest-hit funds in its sterling universe.

Change in euros from end-2025, or launch or rename for the memory funds, to each fund's 2026 peak, then from that peak to 13 November 2026, except Amundi memory to 12 November. Semiconductors (iShares) plus 122.9% then minus 29.1%. Cybersecurity (L&G) plus 77.3% then minus 12.5%. Quantum computing (WisdomTree) plus 57.0% then minus 34.4%. Solar (Invesco) plus 49.3% then minus 41.0%. Clean energy (iShares) plus 43.7% then minus 29.6%. Batteries (L&G) plus 39.9% then minus 39.6%. Blockchain (Invesco) plus 33.9% then minus 22.6%. Memory chips (Amundi, from 16 July) plus 21.5% then minus 32.2%. Memory chips (Defiance, from 18 June) plus 3.7% then minus 49.2%. MSCI World net index plus 19.0% then minus 0.9%.

From the 2026 peak to 13 November (Amundi memory: 12 November), in euros: semiconductors −29.1%, cybersecurity −12.5%, quantum computing (WisdomTree) −34.4%, solar −41.0%, clean energy −29.6%, batteries −39.6%, blockchain −22.6%, memory chips (Amundi, to 12 November) −32.2% and (Defiance) −49.2%; MSCI World net −0.9%. Source: Xetra and Borsa Italiana closes via Yahoo Finance, HANetf DRAM NAV history, Amundi fund page, MSCI World 990100; exchange paths (CSV), memory NAV paths (CSV), daily closes (CSV).

An investor who held the semiconductor ETF since January was still 58.1% up on 13 November. One who bought quantum, solar or battery exposure in late May or early June was 28% to 41% down. The difference lies in the purchase date, which is the subject of this article.

The sell-off continued after these observations. On 17 November Defiance's NAV was $3.7865, 53.8% below its launch NAV, and Amundi's $15.3375, 40.5% below its 1 October peak. MSCI World net in euros closed 3.4% below its October record that day. These later figures are not in the charts or tables, which end on 12 or 13 November.

Themes did not all move together that autumn. In the week of 2 to 6 November, Trackinsight data reported by ETF Express gave solar ETFs a gain of 12.92% and cybersecurity ETFs a loss of 10.03%. Blockchain had the largest thematic outflow that week, €126.8 million.

Why did memory-chip ETFs arrive in June and July 2026?

Two issuers brought memory-chip ETFs to European investors within a month, just as chip stocks peaked. On 17 June 2026 the board of Multi Units Luxembourg gave notice that Amundi MSCI Disruptive Technology would track the Solactive Global Memory Chips Index from 17 July and be renamed Amundi Global Memory Chips. Its SFDR classification moved from Article 8 to Article 6, and holders who disagreed could redeem within 30 days without charges other than divestment fees.

The next day, 18 June, the Defiance Memory UCITS ETF started with a NAV of $8.20 on HANetf's platform, at a total expense ratio of 0.69%. Its index selects companies "deriving at least 50% of revenues from memory and storage activities", according to its September factsheet. Amundi's fund charges 0.45% according to its July factsheet.

Both funds hold the same core: Micron, SK Hynix, Samsung Electronics, Western Digital, Sandisk, Seagate and Kioxia. On 12 November the three largest holdings made up 35.9% of the Defiance fund and 46.3% of the Amundi fund, and the ten largest 85.4% and 87.8%. These are concentrated portfolios of one industry, priced on a memory cycle that had already run for months.

What can fund flows tell us about purchase timing?

A model of daily net creations and redemptions estimates worse money-weighted returns than the NAV path, but it cannot measure actual investor account returns. We value changes in units outstanding at that day's NAV and calculate an IRR using calendar dates. For Amundi, we additionally assume its reported fund assets all belong to this share class. These are model assumptions, not observed investor trades.

NAV returns and modelled net-flow IRRs in US dollars, not annualised. Defiance Memory, 18 June to 13 November 2026: NAV minus 49.2%, modelled IRR minus 54.2%, MSCI World net plus 0.8%. Amundi Memory, 16 July to 12 November: NAV minus 20.4%, modelled IRR minus 33.4%, MSCI World net plus 0.7%. The model assumes flows at NAV; it does not measure investor brokerage returns.

Defiance: NAV −49.2%, modelled net-flow IRR −54.2%, MSCI World net +0.8% (18 June–13 November 2026); Amundi: NAV −20.4%, modelled IRR −33.4%, MSCI World net +0.7% (16 July–12 November); in dollars, not annualised. Source: HANetf DRAM NAV history, Amundi fund page, MSCI World 990100; Defiance flows (CSV), Amundi flows (CSV).

Amundi's fund held $174.6 million on 16 July. By 5 October, four days after its NAV peak, assets reached $666.2 million and units in issue had roughly tripled. The model sums $583.1 million of positive daily net flows and $74.2 million of negative daily net flows to 12 November. Those sums are not gross subscriptions and redemptions; opposite flows within each day are unobserved.

Dividing the positive net flows by the additional units gives an implied average NAV of $22.03, 2.3% above the rename level. The 12 November NAV of $17.16 was 22.1% below that average. This indicates how net fund flows were timed; it is not the average execution price paid by investors on an exchange.

Defiance's NAV had already lost 29.9% by 24 July, when 17.68 million units were outstanding; between 29 and 31 July the count fell from 17.42 million to 6.24 million. It then rose to 25.35 million units by 13 November. The NAV implied by positive daily net flows was $6.11, compared with the final $4.1628, about 32% lower.

An investor can buy from another investor without changing units outstanding. Our model therefore misses exchange-trade timing, spreads, dealing prices and account-level costs or taxes. Its IRR compounded over the study window is also not the percentage of all contributed dollars lost. For your own result, use dated broker cash flows, dividends and final account value in a personal money-weighted return calculation.

What does a memory or quantum ETF add to an MSCI World ETF?

Part of it is already in the world fund, and most of what it adds moves with the world fund. On 12 November, 45.9% of the Defiance fund and 61.0% of the Amundi fund sat in companies that iShares' Core MSCI World UCITS ETF also held on 13 November. The rest was largely in South Korea, Taiwan and China, which MSCI World excludes: 51.8% and 37.1% of the two funds.

Holding on 12-13 November 2026Share of Defiance MemoryShare of Amundi Global Memory ChipsShare of Core MSCI World ETF
Companies also held by the World ETF45.9%61.0%1.75% (all companies held by either memory fund)
South Korea, Taiwan and China51.8%37.1%none
Micron Technology12.0%16.0%0.98%

A portfolio of 90% World ETF and 10% memory ETF holds about 11.6% in the memory funds' companies, against 1.75% in the World ETF alone. Its Micron weight rises from 0.98% to 2.1% or 2.5%, depending on the fund. Source: iShares Core MSCI World holdings, Defiance holdings, Amundi composition; overlap (CSV).

The memory funds are too young for a correlation estimate, so we measured themes with longer histories. Over 76 weekly returns in euros, the VanEck quantum-computing and iShares semiconductor ETFs had correlations of 0.69 and 0.67 with the World ETF. Their betas of 2.13 and 2.45 measure the average sensitivity to weekly World ETF moves in a linear fit over this sample. They reinforce the overlap concern explored in our S&P 500 and MSCI World comparison.

Correlation matrix of weekly euro returns over 76 weeks from 30 May 2025 to 13 November 2026, one UCITS ETF per theme. Correlation with the MSCI World ETF: semiconductors (iShares) 0.67; quantum computing (VanEck) 0.69; blockchain 0.56; batteries 0.61; clean energy 0.35; solar 0.19; cybersecurity 0.49. Solar with clean energy 0.84; semiconductors with blockchain 0.69.

Correlation with the MSCI World ETF, weekly in euros, 30 May 2025 to 13 November 2026: semiconductors (iShares) 0.67, quantum computing (VanEck) 0.69, blockchain 0.56, batteries 0.61, clean energy 0.35, solar 0.19, cybersecurity 0.49. Source: Xetra and Borsa Italiana closes via Yahoo Finance for the iShares, VanEck, Invesco and L&G ETFs named in the method; correlation matrix (CSV), volatility and beta (CSV).

Blockchain moved with semiconductors (0.69) more than with the world fund. Trustnet gave a reason: "Many former bitcoin miners now provide data centres for AI". Solar and clean energy had low correlations with the world fund, 0.19 and 0.35, but 0.84 with each other, and they fell 41.0% and 29.6% from their peaks.

A lower correlation can reduce dependence on world equities, but does not make a holding safe. The solar ETF's annualised volatility of weekly returns was 34.1%, against 9.8% for the World ETF, and it carried its own deep losses. Our independent-bets analysis explains why the number of funds says little about the number of return sources.

What did buying a theme after a rally do in the past?

On average it lowered returns: research on launch timing finds that new themed funds start near the top of their theme's valuation. Ben-David, Franzoni, Kim and Moussawi studied US equity ETFs from 1993 to 2019 in the Review of Financial Studies (2023). They found that "over their first five years, specialized ETFs lose about 30% in risk-adjusted terms", about 6% a year, and attribute it to "the overvaluation of the underlying stocks".

Their portfolio of all specialised ETFs earned risk-adjusted returns of −3.1% a year after fees, while broad-based ETFs were statistically indistinguishable from zero. The study covers US-listed funds, so it describes the average fund rather than memory chips in 2026. Morningstar's 2025 Global Thematic Fund Landscape, with data to 30 September 2025, put the odds of picking a thematic fund that both survives and beats global equities over longer periods at "less than 10%".

The previous cycle shows the same pattern in euros. The iShares clean-energy ETF peaked on 7 January 2021; to 13 November 2026 it lost 45.5%, or 9.9% a year, after a deepest fall of 64.2%. The World ETF gained 114.7% over the same dates, 13.9% a year.

UCITS ETF, from its 2020-2021 peak closeChange to 13 Nov 2026World ETF, same dates
Clean energy (iShares), 7 Jan 2021−45.5%+114.7%
Solar (Invesco), 22 Nov 2021−56.8%+64.9%
Batteries (L&G), 22 Nov 2021+24.1%+64.9%
Blockchain (Invesco), 8 Nov 2021+17.3%+69.1%
Cybersecurity (L&G), 10 Nov 2021+71.3%+69.0%

Cybersecurity is the counterexample: bought at its 2021 peak, it matched the World ETF five years later. Source: exchange closes in euros, adjusted for distributions; previous-cycle table (CSV). Our study of Cathie Wood's ARK Innovation ETF follows an actively managed theme fund through the same cycle.

What does a 20% to 40% fall in a thematic sleeve do to compounded growth?

At a 10% weight, a 40% fall in the sleeve costs the portfolio 4% if everything else holds still, but the sleeve then needs a 66.7% gain to recover. That recovery arithmetic, explained in our guide to loss maths, is what turns a fall into lost compounding. Defiance's memory fund needs 97.0% to regain its launch NAV as of 13 November.

Fall in the sleevePortfolio change at a 5% / 10% / 20% sleeveGain the sleeve needs to recover
20%−1.0% / −2.0% / −4.0%25.0%
30%−1.5% / −3.0% / −6.0%42.9%
40%−2.0% / −4.0% / −8.0%66.7%
50%−2.5% / −5.0% / −10.0%100.0%

The table holds the rest of the portfolio flat, which is a scenario, not a forecast; with correlations of 0.6 to 0.7, a world fund often falls at the same time. In 2026 it did not: 10% of a portfolio placed in Defiance's memory fund at its 22 June NAV peak left a 90/10 mix 1.1% down on 13 November, while the World ETF alone gained 4.3%. Source: flat-core scenarios (CSV), historical portfolio mixes (CSV).

For geometric growth, the useful question is how much extra return the sleeve must earn to pay for the volatility it adds. Taking log growth as roughly the arithmetic return minus half the variance, our weekly data put the extra drag of a 10% sleeve at 0.18 points a year for semiconductors and 0.14 for quantum. Under that approximation, the sleeve breaks even if it beats the world index by 1.8 and 1.4 points a year.

These are sensitivities to a short historical covariance sample, not expected returns, optimal weights or validated long-run hurdles.

The modelled volatility drag is small at this illustrative weight, so it alone is a weak case against a small theme. Joint losses, liquidity needs and an uncertain future return still matter. The decisive input is the sleeve's expected return after a 0.45% to 0.69% fee, and the launch-timing research above found that new themed funds, on average, earned less than the market. A theme bought after a 40% to 120% rise needs a reason to expect the next years to beat the world index, beyond the rally itself.

What can you check before adding or keeping a thematic ETF?

These checks use the issuer's documents and your own statements; none of them predicts whether a theme has bottomed.

  1. Look through the holdings. Compare the theme's holdings file with your world fund's and add any direct shares; count what is new, such as Korean or Taiwanese memory makers.
  2. Measure your own return with cash flows. Your broker's money-weighted return shows what your purchase dates did, which a fund factsheet cannot.
  3. Compare the correlation and the size of past falls. A correlation near 0.7 and a beta near 2 means the sleeve amplifies the market; a low correlation still needs a tolerable worst case.
  4. Stress the theme and the core together. Multiply each portfolio weight by its scenario return and add the results. For illustration, 10% falling 40% plus 90% falling 20% gives a 22% portfolio loss; compare that loss and planned spending with accessible savings.
  5. Read the index notice. A rename such as Amundi's changes what you own; the notice gave holders 30 days to redeem without charges other than divestment fees.
  6. Write down the hurdle. State what return the theme needs above the world index after fees, and check it against what you expected when you bought.

Sources and method

Checked on 16 November 2026, with market data and Defiance NAVs to 13 November, Amundi NAVs to 12 November 2026; the 17 November NAVs and MSCI World level were added on 18 November from HANetf, Amundi and MSCI. Each benchmark comparison uses the same dates as its fund.

Press and research: Gary Jackson, "Semiconductor stocks drop 20%: The funds hit hardest by the AI sell-off so far", Trustnet, 4 November 2026; "European ETF market wrap: week 45 2026", ETF Express, 9 November 2026, thematic section; Morningstar, Global Thematic Fund Landscape 2025, pages 3 and 15.

Ben-David, Franzoni, Kim and Moussawi, "Competition for Attention in the ETF Space", Review of Financial Studies 36(3), 2023; quotes and the −3.1% and −6% figures from NBER Working Paper 28369, revised July 2021, abstract and page 4.

Fund documents: Multi Units Luxembourg, notice to shareholders of Amundi MSCI Disruptive Technology, 17 June 2026; Amundi Global Memory Chips monthly factsheet, 31 July 2026, page 2; NAV, fund assets and composition from Amundi's fund page (LU2023678282).

HANetf: launch release, 19 June 2026; factsheet, 30 September 2026; NAV history (IE000CEUZ052) through 13 November and holdings as of 12 November 2026. World ETF holdings: iShares Core MSCI World UCITS ETF (IE00B4L5Y983), 13 November 2026.

Index and price data: Nasdaq PHLX Semiconductor (SOX) daily closes; MSCI World 990100 net return in EUR and USD and MSCI ACWI 892400 price in USD, from MSCI's index tool; ECB reference rates. ETF prices: Yahoo Finance daily closes in euros, adjusted for distributions.

The ETFs are iShares Core MSCI World (EUNL, Xetra), iShares MSCI Global Semiconductors (IE000I8KRLL9), WisdomTree Quantum Computing (IE000W8WMSL2), VanEck Quantum Computing (IE0007Y8Y157, used in the correlations), Invesco CoinShares Global Blockchain (IE00BGBN6P67, Borsa Italiana), L&G Battery Value-Chain (IE00BF0M2Z96), iShares Global Clean Energy Transition (IE00B1XNHC34), Invesco Solar Energy (IE00BM8QRZ79) and L&G Cyber Security (IE00BYPLS672). One fund per theme is an illustration; other funds in the same theme returned differently.

Our calculations: all paths, peaks, correlations, betas, overlaps, flows, money-weighted returns and scenarios. Peaks are the highest close in 2026; correlations use 76 weekly returns between the last closes of each week, and volatility is annualised with the square root of 52. Estimated daily net flows are the change in units outstanding times that day's NAV. For Amundi, units are fund assets divided by NAV, assuming those assets belong to this share class. Positive and negative daily net flows can hide offsetting creations and redemptions within a day.

The model discounts these flows by elapsed calendar days and treats terminal assets as a withdrawal. We compound the resulting daily IRR over each study window, without annualising it. Initial holdings are valued at the starting NAV. This period-equivalent IRR is neither an observed brokerage return nor the fraction of total contributed dollars lost; secondary-market trades do not change units outstanding.

The memory funds use NAV converted at ECB rates, because Xetra closes of the new Defiance listing differed from NAV by several percent in its first days. The geometric break-even uses log growth ≈ arithmetic return − variance/2 with 18 months of weekly covariances, which describe the past, not future risk. Index and NAV returns exclude investor trading costs, spreads and taxes.

The AI sell-off article covers the same weeks for world-index and S&P 500 investors. Every derived figure is in the CSV files linked above and in the research ledger.

This article is educational, not personalised investment advice.

FAQ

How much did the memory-chip ETFs lose in the 2026 sell-off?

The Defiance Memory UCITS ETF, launched on 18 June 2026 at a NAV of $8.20, had a NAV of $4.1628 on 13 November, a fall of 49.2% in dollars and 47.4% in euros. The Amundi Global Memory Chips UCITS ETF, which switched to a memory-chip index on 17 July, rose 19.7% in dollars to its 1 October peak and then fell 33.4%, ending 20.4% below its 16 July NAV on 12 November.

Do fund-flow estimates measure thematic ETF investors' actual returns?

No. Our model estimates calendar-day money-weighted returns from changes in units outstanding valued at NAV. It gives −54.2% for Defiance to 13 November and −33.4% for Amundi to 12 November 2026, versus NAV falls of 49.2% and 20.4%. Exchange trades, execution prices and individual account cash flows are unobserved; Amundi also assumes the reported assets belong to this share class.

Does a memory-chip ETF overlap with an MSCI World ETF?

Partly. On 12 November, 45.9% of the Defiance fund and 61.0% of the Amundi fund sat in companies that iShares' Core MSCI World ETF also held, led by Micron. The rest was mostly in South Korea, Taiwan and China, which MSCI World does not cover. Those companies weighed 1.75% in the World ETF.

Do thematic ETFs diversify a world-index portfolio?

Few did in our data. Over 76 weeks to 13 November 2026, weekly euro returns of the VanEck quantum-computing and iShares semiconductor ETFs had correlations of 0.69 and 0.67 with an MSCI World ETF, with betas of 2.13 and 2.45. A solar ETF had a correlation of 0.19, but it fell 41.0% from its May peak and 56.8% from its November 2021 peak.

Do thematic funds beat global equities over the long run?

Rarely, in the studies we checked. Morningstar's 2025 Global Thematic Fund Landscape, with data to 30 September 2025, put the odds of picking a thematic fund that both survives and beats global equities over longer periods at less than 10%. A study of US equity ETFs from 1993 to 2019 found that specialised ETFs lost about 30% risk-adjusted over their first five years.

Optional cookies help us measure visits and advertising. Choose each purpose, or decline both. You can change your choice using Cookie settings in the footer. Cookie policy.