At 30 September 2026, NVIDIA was the largest holding in the Vanguard FTSE All-World UCITS ETF (VWRA), at 5.01% of a fund holding 3,821 stocks. In the iShares Core S&P 500 UCITS ETF (CSPX) it was 8.37% of the fund. Before weighing AI stocks vs AI ETFs, investors in Singapore holding a global or S&P 500 fund should know they already own a sizeable slice of the AI trade.
An AI stock is a company whose revenue depends in a measurable way on spending on artificial intelligence, from chip designers to cloud providers. An AI ETF holds a basket of such companies chosen by a thematic index, and the definition of "AI" varies widely from fund to fund. In the VanEck Semiconductor ETF (SMH), NVIDIA alone made up 20.92% of the fund on 22 October 2026.
Investors here can buy US-listed AI stocks and ETFs through a broker with US market access, and Ireland-domiciled UCITS AI ETFs on the London Stock Exchange. Whichever route you take, the same few mega-cap names keep appearing, so what matters is how much of your portfolio ends up riding on them.
The options run from single stocks, through semiconductor funds and broad AI and big-data funds, to narrower themes such as AI infrastructure, robotics and data centres.
This guide explains where AI revenue accrues, how to analyse a single AI stock, the main types of AI ETF and what they hold, how much they overlap with VWRA, ISAC and CSPX, a worked example of look-through exposure, costs and tax for investors in Singapore, and what to monitor each quarter.
AI stocks vs AI ETFs at a glance
The fund figures come from the five AI-related ETFs covered in this guide.
| Factor | Individual AI stocks | AI ETFs |
|---|---|---|
| Number of companies | One per position | 25 to 90 or more, depending on the index |
| Single-company risk | Full exposure to one company's results | Spread, though top holdings can exceed 10% to 20% |
| Ongoing fund cost | None | Expense ratios of 0.35% to 0.68% a year |
| Research workload | High: earnings, guidance, competition, valuation | Lower: index rules, holdings and fees |
| Theme purity | Depends on how much revenue comes from AI | Depends on index definitions, which vary widely |
| Overlap with a global core fund | Direct, if the stock is a large index member | Often heavy in the largest names |
| Rebalancing | Your decision | Set by the index schedule |
Source: VanEck SMH, Global X BOTZ, Global X DTCR, iShares AINF, Xtrackers XAID fact sheet (as of 25 October 2026)
Where AI revenue actually accrues
"AI stock" covers companies with very different business models. Mapping the value chain shows which layer each stock or fund depends on.
AI value chain from chips to cloud, software and data centres, mapped to SMH, AINF, XAID, BOTZ and DTCR
Hardware: semiconductors and memory
Chip designers, foundries, equipment makers and memory producers sell the processors and memory that train and run AI models. Their AI revenue is often the easiest to trace, because data-centre chip sales appear in segment reporting. They are also cyclical: orders rise and fall with customers' spending plans.
Cloud and hyperscale infrastructure
The largest cloud providers buy chips and build data centres, then rent computing power. Their AI exposure shows up as capital expenditure first and revenue later, so the gap between what they spend and what AI services earn is a key measure to track.
Models, software and applications
Model developers and software companies sell AI features to businesses and consumers. Revenue here is harder to isolate, because AI is often bundled into existing products and subscriptions.
Power, cooling, networking and data-centre property
Data centres need electricity, cooling, networking gear and physical sites. Data-centre real estate investment trusts and tower owners earn rent from the build-out, a rental income model with its own sensitivity to interest rates.
How to analyse an individual AI stock
Buying a single AI stock means taking a view on one company's results, so the research has to cover the business itself as well as the theme. Eight questions do most of the work.
Is AI revenue reported or estimated? Look for a segment line in the annual or quarterly filing. Management commentary and analyst estimates carry less weight.
How fast is that revenue growing? Compare growth against the prior year and against guidance.
What are the margins? Gross margin shows pricing power; a falling margin can signal competition.
How much capital does growth need? Compare capital expenditure with operating cash flow and free cash flow.
How concentrated are the customers? A supplier that depends on a handful of cloud buyers carries their spending decisions.
What is priced in? Price-to-earnings and price-to-sales ratios show how much future growth the market already expects.
How strong is the balance sheet? Net cash or manageable debt gives room to keep investing through a downturn.
How durable is the competitive position? Technology lead, switching costs and supply constraints can all erode.
Repeating this for several companies every quarter is the main cost of the single-stock route. An ETF hands that work to an index committee, at the price of an expense ratio and a fixed set of rules.
Types of AI ETFs
AI ETFs fall into five broad groups, and the label on the fund says little about what it holds.
| Fund | Ticker | Index | Domicile | Fee | Holdings | Largest holding |
|---|---|---|---|---|---|---|
| VanEck Semiconductor ETF | SMH | MVIS US Listed Semiconductor 25 | US | 0.35% | 25 stocks | NVIDIA 20.92% |
| Xtrackers Artificial Intelligence & Big Data UCITS ETF | XAID | Nasdaq Global AI and Big Data | Ireland | 0.35% | 90 in index | Microsoft 5.75% |
| iShares AI Infrastructure UCITS ETF | AINF | STOXX Global AI Infrastructure | Ireland | 0.35% | 52 | IT sector 89.76% |
| Global X Robotics & Artificial Intelligence ETF | BOTZ | Indxx Global Robotics & AI Thematic | US | 0.68% | 81 | Intuitive Surgical 10.22% |
| Global X Data Center & Digital Infrastructure ETF | DTCR | Solactive Data Center REITs & Digital Infrastructure | US | 0.50% | 25 | American Tower 14.95% |
Source: VanEck SMH (holdings 22 October 2026), Xtrackers XAID fact sheet (30 September 2026), iShares AINF (22 October 2026), Global X BOTZ and Global X DTCR (23 October 2026) (as of 25 October 2026). XAID's fee is a flat all-in fee. AINF's holdings breakdown is shown by sector.
Semiconductor funds
SMH tracks 25 US-listed chip companies, with NVIDIA at 20.92% and Taiwan Semiconductor at 9.72% on 22 October 2026. Its expense ratio is 0.35%, and 78.78% of the fund was in US companies at 30 September 2026. It is the most concentrated of the five and the closest to a bet on the hardware layer alone.
Broad AI and big-data funds
XAID is the London USD line of the Xtrackers Artificial Intelligence & Big Data UCITS ETF (ISIN IE00BGV5VN51), an Irish accumulating fund. At 30 September 2026 its top ten holdings made up 45.40% of the fund, led by Microsoft (5.75%), Meta (5.35%), Apple (4.93%) and NVIDIA (4.82%), for an all-in fee of 0.35% a year. It spreads across chips, platforms and software, which also makes it the fund that looks most like the top of a global index.
AI infrastructure funds
AINF is the iShares AI Infrastructure UCITS ETF (ISIN IE000X59ZHE2), an Irish accumulating fund. It held 52 companies on 22 October 2026, with 89.76% in information technology, for a TER of 0.35%. It targets the hardware and equipment that data centres need.
Robotics and automation funds
BOTZ leans towards industrial automation and medical robotics. At 30 September 2026, industrials made up 47.38% of the fund, and US and Japanese companies 33.96% and 31.94%. Its 0.68% expense ratio is the highest of the five, and its country mix differs most from an S&P 500 fund.
Data-centre and digital-infrastructure funds
DTCR holds data-centre REITs, tower owners and hardware makers. Its index requires companies to earn at least 50% of revenue from those areas, and real estate made up 56.87% of the fund at 30 September 2026. Singapore companies made up 4.34% of the fund. It behaves more like a property and infrastructure fund than a technology fund, with sensitivity to interest rates.
AI ETF overlap with VWRA, ISAC and CSPX
A global core fund already holds the largest AI names at market weight. The MSCI ACWI Index, which ISAC tracks, had information technology at 33.21% and NVIDIA at 5.11% on 30 September 2026. VWRA's top ten holdings made up 24.73% of the fund at the same date, and nine of those ten were classed as technology companies.
The six largest names, NVIDIA, Apple, Microsoft, Amazon, Alphabet and Meta, made up 20.28% of VWRA. In XAID, the same six made up 29.24%. Seven of XAID's top ten holdings also sit in VWRA's top ten, so adding XAID to VWRA mostly increases weight in companies you already own.
| Company | VWRA | CSPX | SMH | XAID |
|---|---|---|---|---|
| NVIDIA | 5.01% | 8.37% | 20.92% | 4.82% |
| Apple | 4.52% | 7.36% | Not held | 4.93% |
| Microsoft | 3.58% | 5.77% | Not held | 5.75% |
| Meta Platforms | 1.50% | 2.42% | Not held | 5.35% |
| Broadcom | 1.54% | 2.53% | 5.56% | Outside top 10 |
| Micron Technology | 1.13% | 1.82% | 4.59% | 3.85% |
| Taiwan Semiconductor | 1.78% | Not in index | 9.72% | Outside top 10 |
Source: Vanguard VWRA (30 September 2026), iShares CSPX fact sheet (30 September 2026), VanEck SMH (22 October 2026), Xtrackers XAID fact sheet (30 September 2026) (as of 25 October 2026). "Outside top 10" means the issuer publishes only the top ten holdings in that document. SMH holds only semiconductor companies.
Different tickers can still carry the same exposure. A portfolio of VWRA plus XAID plus SMH can hold NVIDIA three times over, and the combined weight is what matters for risk.
Holding weights of NVIDIA, Apple, Microsoft and other AI names across VWRA, CSPX, SMH and XAID
Worked example: S$20,000 with direct NVIDIA and an AI ETF
Take S$20,000 split four ways, add up NVIDIA across every holding, and work out what a 50% fall in NVIDIA's share price alone would cost each portfolio, with everything else unchanged.
| Portfolio | NVIDIA look-through weight | NVIDIA exposure | Loss if NVIDIA falls 50% |
|---|---|---|---|
| A: 100% VWRA | 5.01% | S$1,001 | S$501 (2.50%) |
| B: 85% VWRA, 15% NVIDIA shares | 19.25% | S$3,851 | S$1,925 (9.63%) |
| C: 70% VWRA, 15% SMH, 15% NVIDIA shares | 21.64% | S$4,328 | S$2,164 (10.82%) |
| D: 70% VWRA, 15% SMH, 15% XAID | 7.36% | S$1,473 | S$736 (3.68%) |
Example only, based on Vanguard VWRA (NVIDIA 5.0056% at 30 September 2026), VanEck SMH (20.92% at 22 October 2026) and Xtrackers XAID (4.82% at 30 September 2026) (as of 25 October 2026). Assumes fixed weights and no change in other holdings.
Portfolio C looks diversified on paper, with three holdings, yet more than a fifth of it rides on one company. Portfolio D adds two AI funds and still keeps NVIDIA below a tenth, because XAID spreads its weight across many names.
NVIDIA look-through exposure in four S$20,000 portfolios combining VWRA, SMH, XAID and direct shares
Risk, volatility and valuation
Thematic funds and single stocks both tend to swing more than a global index. A fund of 25 chip companies moves with one industry's order cycle, and a single company moves with one set of results. Returns for different AI stocks can also diverge sharply over the same period, so a good pick and a bad one can end up far apart.
Valuation is the second risk. CSPX's S&P 500 holdings traded at a price-to-earnings ratio of 30.28x at 30 September 2026, with its top ten holdings, mostly large technology companies, at 38.99% of the fund. High ratios leave less room for disappointment if AI spending slows or the returns on that spending fall short.
To compare performance fairly, use total returns over the same period for each stock and fund, measured in the same currency, and look at the deepest fall along the way as well as the end result. A thematic fund launched after a strong run has a short record that can flatter or mislead.
Costs, tax and access from Singapore
Single stocks carry no annual fund fee, only trading costs and currency conversion. AI ETFs charge expense ratios of 0.35% to 0.68% a year, on top of the same trading costs.
Domicile changes the tax picture. US-source dividends paid to nonresident aliens face 30% withholding by default, which applies to US shares and to US-domiciled ETFs such as SMH, BOTZ and DTCR. Irish funds such as XAID and AINF can receive US dividends at the 15% general treaty rate for Ireland at the fund level.
Estate tax matters more. US shares and US-domiciled ETFs are US-situs assets, and the estate of a non-US person must file Form 706-NA when those assets exceed US$60,000. Singapore generally does not tax gains from selling shares and financial instruments for individuals investing personally, though frequent trading can change that. This is general information, not tax advice.
Buying US stocks or US-listed ETFs needs a broker with US market access and a W-8BEN form; UCITS funds trade on the London Stock Exchange.
Which route suits your approach?
The choice comes down to how much research you want to do and how much concentration you can hold through a bad year.
Single AI stocks suit investors who will read quarterly filings, can explain why one company should outperform its peers, and keep each position small enough that a 50% fall is survivable.
Semiconductor funds such as SMH give you the hardware layer without picking one chipmaker, though the largest holdings still dominate.
Broad AI funds such as XAID spread across chips, platforms and software, and overlap most with a global core fund.
Narrower themes such as AINF, BOTZ or DTCR work as a small satellite aimed at one part of the build-out, such as equipment, robotics or data-centre property.
A global core fund alone is enough for investors content with the AI weight the market already gives them, which was 5.01% in NVIDIA alone through VWRA at 30 September 2026.
What to monitor each quarter
Whichever route you choose, a short quarterly check keeps the position honest.
- Reported AI revenue at the companies you hold or that dominate your funds.
- Cloud capital expenditure plans from the largest buyers of AI chips.
- Semiconductor sales and gross margins, which show whether pricing is holding.
- Order backlogs at chip and equipment makers.
- Your ETF's top holdings and weights, since index rebalances can shift concentration.
- Your total look-through weight in the largest names across every fund and stock you own.
Build AI exposure with StashAway
ETF Explorer lets you add AI as a satellite with one fund per part of the value chain: SMH for Semiconductors, AINF for AI Infrastructure, DTCR for Data Centres, BOTZ for Robotics and XAID for broad Artificial Intelligence exposure. VWRA (Global Equities (FTSE)) and ISAC (Global Equities) cover the core, so you can see how much NVIDIA you already own before adding a theme.
Each buy or sell order costs US$1 (excluding GST), with no management fee on top, and you can pay with cash or SRS.
AI funds are concentrated in a few richly valued companies, so you could get back less than you invest.
Frequently asked questions
Short answers on AI stocks, AI ETFs and their overlap with a global core fund.
Are AI ETFs safer than AI stocks?
An AI ETF spreads company-specific risk across 25 to 90 or more holdings, so one bad earnings report does less damage. The fund still moves with the AI theme as a whole and can have one stock above 20% of its assets, as SMH did with NVIDIA in October 2026.
Is NVIDIA already in VWRA?
Yes. NVIDIA was VWRA's largest holding at 5.01% on 30 September 2026, and it was also the largest constituent of the MSCI ACWI Index tracked by ISAC.
Which AI ETFs can Singapore investors buy?
US-listed funds such as SMH, BOTZ and DTCR need a broker with US market access. Irish UCITS funds such as XAID and AINF trade on the London Stock Exchange.
Are semiconductor ETFs the same as AI ETFs?
They overlap without matching. SMH holds 25 chip companies, many of which sell into AI data centres, but it excludes the cloud platforms and software firms that broad AI funds such as XAID hold.
What data shows whether an AI company can earn its valuation?
Start with reported revenue from AI-related segments, its growth rate and gross margin, then compare capital spending with free cash flow. Set those against valuation ratios such as price-to-earnings, and track whether results meet the company's own guidance over several quarters.
Can I hold an AI ETF alongside a global index fund?
Yes. A thematic fund often works as a small satellite next to a global core. Add up the look-through weight of the largest names first, because the same companies often sit at the top of both funds.
