Themes · general research

AI infrastructure and chips

AI accelerators, memory, chip equipment, data-centre power and networking, GPU clouds.

High risk

Can lose most of the money. Typical volatility 66%, typical worst fall −65%; 2 of 14 companies are more than half below their high.

The only theme where most names are profitable large companies with real revenue growth and near their highs. GPU clouds (CoreWeave, Nebius) and AI servers are the speculative edge.

Typical volatilityMedian 1-year volatility of the theme's companies (annualised daily moves).
66%
Typical worst fallMedian of each company's worst peak-to-trough fall in five years (or since listing).
−65%
Typical distance from the high
−26%
Typical 1-year return
+29%

How to hold it in the EU

US-listed ETFs have no KID and cannot be bought by EU retail investors; these UCITS lines can. One fund line spreads the single-company risk.

VanEck Semiconductor UCITS ETF
ISIN IE00BMC38736TER 0.35%Assets 8,241 M € Volatility 1y 42%Worst fall 1y −25%Since 2020-12-01
LineWhere
VVSM Xetra · EURFreedom24IBKR
SMH London Stock Exchange · USDFreedom24?IBKR
SMGB London Stock Exchange · GBPFreedom24?IBKR
SMH Borsa Italiana · EURIBKR
SMH Euronext Paris · EURIBKR
iShares MSCI Global Semiconductors UCITS ETF Acc
ISIN IE000I8KRLL9TER 0.35%Assets 5,709 M € Volatility 1y 37%Worst fall 1y −29%Since 2021-08-03
LineWhere
SEC0 Xetra · EURFreedom24IBKR
SEMI London Stock Exchange · GBPFreedom24?IBKR
SEMI Euronext Amsterdam · USDIBKR
Xtrackers AI & Big Data UCITS ETF 1C
ISIN IE00BGV5VN51TER 0.35%Assets 8,554 M € Volatility 1y 22%Worst fall 1y −14%Since 2019-01-29
LineWhere
XAIX Xetra · EURFreedom24IBKR
XAID London Stock Exchange · USDFreedom24?IBKR
WisdomTree Artificial Intelligence UCITS ETF Acc
ISIN IE00BDVPNG13TER 0.40%Assets 1,337 M € Volatility 1y 30%Worst fall 1y −20%Since 2018-11-30
LineWhere
WTI2 Xetra · EURFreedom24IBKR
WTAI London Stock Exchange · USDFreedom24?IBKR

Green: available; dashed with “?”: the venue is offered, check the line in the terminal; crossed out: not available.

US theme ETFs (SMH) are used only to measure the theme's risk: not buyable by EU retail.

The companies

Research data of 2026-10-15; where we score the company, its live Smart Score and red flags are shown.

Volatility: 1 year; worst fall and best/worst: 5 years or since listing; “12 m” = the share of rolling 12-month windows that at least doubled / at least halved.
TickerMarket capRevenue (y/y)P/SVolatility 1yWorst fallFrom high1 year12 m: ×2 / −50 %WhereOur rating
NVDA
AI accelerators
5.34 T$303.0 B$ (+83%)18×37%−66%−7%+18%42% / 2%Freedom24IBKR8/10Caution. Accruals or the Beneish M-score are in the watch zone. More in the glossary →
AVGO
custom AI chips and networking
1.65 T$89.1 B$ (+49%)19×46%−41%−28%−3%22% / 0%Freedom24IBKR3/10Caution. Share count growing 3–10 % a year over three years. More in the glossary →Caution. Accruals or the Beneish M-score are in the watch zone. More in the glossary →
TSM
TSMC: chip foundry (ADR)
1.97 T$139.6 B$ (+31%)14×39%−57%−11%+42%11% / 0%Freedom24IBKR
AMD
CPUs and GPUs
931.9 B$41.3 B$ (+40%)23×68%−65%−12%+164%21% / 6%Freedom24IBKR5/10
MU
memory chips (HBM)
1.03 T$133.2 B$ (+256%)8×80%−58%−25%+374%27% / 0%Freedom24IBKR9/10Caution. Dividend Safety below 40 points. More in the glossary →Note. Dividend or fundamental sources disagree beyond tolerance. More in the glossary →
ASML
lithography machines (also Euronext Amsterdam)
652.8 B$40.3 B$ (+10%)16×46%−57%−17%+68%11% / 0%Freedom24IBKR7/10
VRT
data-centre power and cooling
85.8 B$11.5 B$ (+26%)7×65%−70%−41%+24%53% / 0%Freedom24IBKR4/10
ANET
data-centre networking
246.5 B$10.5 B$ (+33%)23×54%−50%−10%+33%12% / 0%Freedom24IBKR6/10Caution. Accruals or the Beneish M-score are in the watch zone. More in the glossary →
CRDO
Credo: high-speed connectivity
33.7 B$1.6 B$ (+165%)21×96%−62%−41%+20%53% / 0%Freedom24IBKR4/10Caution. Share count growing 3–10 % a year over three years. More in the glossary →
ALAB
Astera Labs: connectivity chips
52.1 B$1.2 B$ (+98%)43×99%−64%−38%+50%42% / 0%Freedom24IBKR2/10Serious warning. Beneish M-score above −1.78 together with very high accruals: earnings may be overstated. More in the glossary →Caution. Share count growing more than 10 % a year over three years. More in the glossary →
SMCI
AI servers
23.5 B$39.1 B$ (+78%)1×95%−85%−70%−35%40% / 8%Freedom24IBKR1/10Serious warning. Beneish M-score above −1.78 together with very high accruals: earnings may be overstated. More in the glossary →Caution. Negative free cash flow in at least two of the last three years. More in the glossary →+1
NBIS
Nebius: GPU cloud
51.0 B$1.4 B$ (+507%)37×107%−80%−35%+37%35% / 9%Freedom24IBKR1/10Caution. Negative free cash flow in at least two of the last three years. More in the glossary →Caution. Net debt above 4× EBITDA or interest cover below 2×. More in the glossary →
CRWV
CoreWeave: GPU cloud
36.3 B$7.6 B$ (+115%)5×94%−67%−64%−54%18% / 1%Freedom24IBKR1/10Serious warning. Altman Z in the distress zone and negative free cash flow. More in the glossary →Caution. Negative free cash flow in at least two of the last three years. More in the glossary →+1
PLTR
AI and data software
482.9 B$6.2 B$ (+79%)78×58%−78%−4%+13%52% / 7%Freedom24IBKR4/10Caution. Share count growing 3–10 % a year over three years. More in the glossary →Caution. Accruals or the Beneish M-score are in the watch zone. More in the glossary →

Survivorship: the list is of companies known in 2026; earlier names that went bankrupt or were taken private are missing, so the “×2” share is an upper bound.

The theme as a basket

Equal weights, monthly rebalance, 08.08.2022–14.10.2026, compared with the S&P 500.

Equal-weight basketCAGRVolatility 1yWorst fall1 year
NVDA AVGO TSM AMD MU ASML+64%38%−32%+78%
VRT ANET SMCI CRDO NBIS+83%52%−46%+20%
S&P 500 (SPY)+15%16%−19%+13%

The window starts near the 2022 low, which flatters every basket. Past results do not predict future ones.

Base rates: how such bets usually end

Research results to keep in mind before any upside.

IPOs, 3 years38.5% of IPOs lost more than half in 3 years; 16.1% doubled, 0.5% grew 11× or more.

Jay Ritter, IPO statistics, Table 16e (9,195 IPOs 1975–2021), 23.03.2026

SPAC mergers−51% average 1-year return after a SPAC merger (2012–2024).

Ritter / Gahng–Ritter–Zhang, Table 15c (625 de-SPACs 2012–2024), 05.10.2026

Thematic ETFs−30% risk-adjusted over the first 5 years of specialised (thematic) ETFs.

Ben-David, Franzoni, Kim & Moussawi, RFS 36(3), 2023

Lottery stocks> 1% a month: how much “lottery” stocks (the biggest one-day jumps) underperform.

Bali, Cakici & Whitelaw, JFE 99(2), 2011

Where the gains come from4% of US stocks created all the net wealth since 1926.

Bessembinder, JFE 129(3), 2018

Volatility drag≈ 46% a year is lost to volatility at 96% volatility: +40 % a year on average compounds to about −6 %.

standard approximation: geometric mean ≈ arithmetic mean − σ²/2 (our arithmetic)

Source: Nasdaq historical closes 15.10.2021–14.10.2026; stockanalysis.com fundamentals 15.10.2026; justETF profiles and quotes 15.10.2026 · Research data of 2026-10-15 · Methodology and sources