Major technology companies are projected to spend US$700 billion on AI infrastructure in 2026, up from US$410 billion in 2025, according to a World Economic Forum report. 

Singapore’s electricity demand from data centres is projected to nearly triple between 2025 and 2030, with the island’s 70-plus facilities collectively exceeding 1.4 gigawatts of capacity.

An AI ETF covers a very different slice of that spend depending on where it sits in the value chain. The Roundhill Memory ETF (DRAM) launched in April 2026 and reached US$24 billion in assets by end-June 2026; it holds nine memory stocks and nothing else. 

The Corgi Lithography & Semiconductor Photonics ETF (EUV), launched in May 2026, targets companies making the extreme ultraviolet lithography equipment that enables every advanced chip. 

These are not interchangeable with a broad AI ETF like XAIX or AIQ. Each occupies a distinct layer.

The AI value chain runs from chip design and lithography equipment through memory, networking, data centres, electricity, cloud platforms and eventually into the enterprise software and robots that apply AI commercially. Investing across that chain requires knowing which layer each fund targets and what it leaves out.

Choosing funds for this theme involves decisions across several dimensions:

• Which layer of the value chain the ETF targets, from compute and memory through to applications

• Whether the sub-theme is genuinely distinct from an existing semiconductor or broad-tech holding

• UCITS (Ireland or Luxembourg-domiciled) versus US-listed funds, and the tax implications of each

• Accumulating versus distributing share classes for long-term compounding

• Which exchange the ETF trades on and which brokerages in Singapore provide access

• Total cost of ownership beyond the headline expense ratio

Invest in AI-related ETFs with StashAway ETF Explorer. Access UCITS and global ETFs across 90+ asset classes at a flat USD 1 per transaction. No annual platform fee. SRS-eligible. Explore now →

TLDR: AI value chain ETFs by layer

How the AI value chain works

The AI investment opportunity extends far beyond Nvidia and other chipmakers. Building and commercialising AI depends on a much wider ecosystem of companies supplying computing power, physical infrastructure, electricity, cloud services and end-user applications.

The value chain can be divided into five broad layers.

The first is compute, which covers the semiconductors, manufacturing equipment and memory needed to train and run AI models. This includes GPU designers, chip foundries, lithography-equipment makers and producers of high-bandwidth memory.

The second layer is physical infrastructure. AI chips need to be connected, cooled and housed, creating demand for networking equipment, optical components, specialised servers, liquid-cooling systems and data-centre facilities.

The third is electricity and power infrastructure. AI data centres consume substantial amounts of electricity, which links the theme to power generators, regulated utilities, grid-equipment manufacturers, nuclear operators and uranium suppliers. These companies provide more indirect AI exposure, as their earnings are also influenced by electricity prices, regulation and broader energy demand.

The fourth layer is cloud and data infrastructure. Cloud platforms, databases and developer tools provide the software environment in which AI models are trained, deployed and integrated into business operations.

The final layer is applications, where companies attempt to turn AI investment into revenue or productivity gains. This includes enterprise software, industrial automation, robotics, machine vision and cybersecurity platforms.

Layer 1: Compute, semiconductor manufacturing and memory

The compute layer sits at the start of the AI value chain. It supplies the processors, production capacity and memory required to train and run AI models.

This layer can be divided into three sub-layers. The first covers companies designing GPUs, CPUs and other specialised AI processors. The second covers the foundries and equipment manufacturers that produce those chips. The third covers high-bandwidth memory, DRAM, NAND and data-storage products that allow processors to access and retain large volumes of data.

The tables below cover the main unleveraged US-listed and UCITS ETFs with direct exposure to these areas. Broad technology funds, single-stock ETFs, leveraged products and inverse funds are excluded.

1.1 Core computing and chip design

Core computing covers GPUs, CPUs, AI accelerators, custom chips, networking processors, semiconductor intellectual property and chip-design software. Nvidia, AMD, Broadcom, Arm, Marvell, Synopsys and Cadence are among the main companies in this part of the value chain.

Fabless semiconductor ETFs provide the most targeted exposure to chip designers. Broader semiconductor funds combine these companies with foundries, memory producers and equipment manufacturers.

1.2 Semiconductor manufacturing and equipment

Chip designers rely on foundries to turn their designs into physical semiconductors. The manufacturing process also requires lithography machines, wafer-fabrication equipment, testing systems, advanced packaging and electronic design tools.

Companies in this sub-layer include TSMC, Intel, ASML, Applied Materials, Lam Research, KLA, Tokyo Electron and Advantest. There are few ETFs dedicated solely to foundries or semiconductor equipment, so exposure normally comes through global semiconductor or supply-chain funds.

1.3 Memory and data storage

AI processors need fast access to large quantities of data. This has made high-bandwidth memory, or HBM, a critical component of AI servers, alongside DRAM, NAND flash, solid-state drives and enterprise storage.

The main listed companies include Samsung Electronics, SK Hynix, Micron, Kioxia, SanDisk, Western Digital and Seagate. Dedicated ETF exposure remains limited. Most semiconductor ETFs hold memory producers as part of a much broader portfolio.

Roundhill’s DRAM is the only dedicated memory ETF in this comparison. The remaining funds provide memory exposure alongside processors, foundries and semiconductor equipment, so the size of their allocation to companies such as Micron, SK Hynix and Samsung can differ considerably.

Holdings and portfolio weights are based on the latest issuer or fund data available between 30 April and 26 June 2026. Holdings change with market movements and index rebalancing. Cash positions may cause issuer-reported fund counts to differ by one from the number of underlying companies.

Layer 2: Networking, servers, cooling and data centres

AI processors need more than chips. They must be connected to other processors, housed inside high-density facilities, cooled efficiently and supported by digital infrastructure that can move large volumes of data.

This layer focuses on the physical infrastructure surrounding AI compute: data-centre REITs, colocation operators, telecom towers, communications equipment and AI-infrastructure suppliers. It excludes pure semiconductor ETFs, which belong in Layer 1; power-grid and nuclear ETFs, which belong in Layer 3; cloud-computing ETFs, which belong in Layer 4; and robotics, cybersecurity and AI-application ETFs, which belong in Layer 5.

2.1 AI infrastructure suppliers

These ETFs target companies supplying the infrastructure used to scale AI computing systems. They may still hold some semiconductor names, but their mandate extends beyond chip design into networking, data-centre capacity and supporting infrastructure.

2.2 Data-centre and digital infrastructure ETFs

These ETFs focus on the companies that own, operate or provide the physical infrastructure behind data storage, processing and connectivity. The portfolios are more exposed to data-centre REITs, tower companies, communications equipment and digital-infrastructure operators than to software or cloud platforms.

This layer is useful for investors who want AI infrastructure exposure beyond semiconductor designers. However, the fund choice matters. AI-infrastructure ETFs such as AINF and WAGI still have meaningful technology and semiconductor exposure, while DTCR, VPN, SRVR and IDGT are closer to real estate and digital-infrastructure plays.

Layer 3: Electricity, grids and nuclear power

AI data centres require large and reliable electricity supply. That has made power generation, grid upgrades, nuclear energy and uranium part of the AI infrastructure discussion.

This layer is still indirect. These ETFs are not AI software or semiconductor funds. Their returns are driven by electricity demand, regulation, power prices, grid investment, uranium prices and nuclear policy. The AI link is that larger data-centre workloads may increase the need for generation capacity and grid infrastructure.

3.1 Smart grid, electrification and AI power infrastructure ETFs

These ETFs focus on the electricity infrastructure needed to support higher power demand, including grid equipment, transmission systems, electrical components, energy-management software and power-infrastructure suppliers. This table excludes data-centre REITs and digital-infrastructure ETFs, which sit in Layer 2.

3.2 Uranium and nuclear energy ETFs

Nuclear and uranium ETFs provide exposure to a different part of the electricity chain. Uranium-mining funds are more sensitive to uranium supply and prices, while broader nuclear funds may include utilities, reactor developers, engineering companies and nuclear-component suppliers.

Layer 4: Cloud, data platforms and AI infrastructure software

Cloud computing is where AI models are trained, hosted and deployed. This layer includes cloud infrastructure, platform services, databases, developer tools, observability software and SaaS companies that support enterprise AI adoption.

This section excludes semiconductor ETFs from Layer 1, data-centre infrastructure ETFs from Layer 2 and cybersecurity ETFs from Layer 5. Some cloud ETFs may still hold a small number of cybersecurity or networking companies, but their primary exposure is cloud software and infrastructure platforms.

Cloud-computing ETFs

Layer 5: AI applications, robotics and cybersecurity

The final layer captures companies that commercialise AI or use it inside products, workflows and physical systems. It includes enterprise AI adopters, robotics and automation companies, and cybersecurity businesses using AI to detect, prevent and respond to threats.

This layer excludes cloud ETFs from Layer 4 and data-centre infrastructure ETFs from Layer 2. It also avoids repeating semiconductor ETFs from Layer 1, even though some robotics and AI funds may still hold chip companies as part of broader portfolios.

5.1 AI applications and adopters ETFs

These ETFs focus on companies expected to benefit from applying AI to products, services and business processes. They sit further downstream than semiconductor or AI-infrastructure funds.

5.2 Robotics and automation ETFs

Robotics ETFs hold companies involved in industrial robots, machine vision, warehouse automation, surgical robotics, factory automation and autonomous systems. They are part of the AI adoption layer because AI can improve how machines perceive, decide and act in the physical world.

5.3 Cybersecurity ETFs

Cybersecurity is an adjacent AI beneficiary rather than a direct AI-infrastructure play. AI can increase both cyberattack volume and defensive capabilities, which makes cybersecurity a separate downstream theme.

Data as of July 2026, based on the latest available issuer, exchange or ETF database disclosures. Holdings and weights change with market prices, index rebalancing and fund flows.

Broad AI ETFs: cross-layer exposure in one fund

Broad AI ETFs combine several parts of the AI value chain in a single portfolio. Instead of targeting only semiconductors, data centres, cloud software, power infrastructure or robotics, these funds typically hold a mix of AI hardware enablers, cloud platforms, software companies and businesses applying AI across different industries.

They are most useful for investors who want one thematic AI allocation rather than assembling the value chain layer by layer. The trade-off is less precision: a broad AI ETF may still be heavily exposed to semiconductors or mega-cap technology stocks, while another may lean more towards software, automation or AI adopters.

The funds below are broad AI ETFs only. Specialist funds already covered in Layers 1 to 5 are excluded.

Ireland-domiciled broad AI UCITS ETFs

US-listed broad AI ETFs

The primary layers column is an approximation. Broad AI ETFs can change their layer exposure after index rebalancing or active portfolio changes. The practical test is to compare the ETF’s largest holdings with the investor’s existing S&P 500, Nasdaq-100, semiconductor or global equity ETF exposure.

Where to buy AI ETFs in Singapore

AI ETFs trade mainly on the London Stock Exchange, NYSE Arca, Nasdaq and other European exchanges. Ireland-domiciled UCITS AI ETFs such as XAIX, WTAI, AIAI, AINF and AIAA usually require access to the LSE or European exchanges, while US-listed AI ETFs such as AIQ, ROBT, THNQ, CHAT and BOTZ require access to US exchanges.

Investors generally have three routes: local bank brokerages, global or fintech brokerages, and simplified investment platforms.

Platform fees as reported in published brokerage rate cards. Verify current rates directly with each platform before placing an order, as fees are subject to change.

Platform choice depends on which exchange the AI ETF is listed on and how frequently the investor plans to trade. London-listed UCITS AI ETFs such as XAIX, WTAI, AIAI, AINF and AIAA require access to the LSE or other European exchanges. US-listed AI ETFs such as AIQ, ROBT, THNQ, CHAT and BOTZ require access to US exchanges. Investors comparing both UCITS and US-listed AI ETFs should also account for foreign-exchange spreads, bid-ask spreads, custody fees, dividend withholding treatment and any platform fees.

Invest in AI ETFs with StashAway

StashAway ETF Explorer gives investors flat-fee access to UCITS and global ETFs covering semiconductors, AI infrastructure, robotics, and 90-plus other asset classes, at USD 1 per transaction with no annual platform fee. SRS accounts are supported.

How AI ETFs can fit into your portfolio

AI ETFs have become more relevant because AI is no longer just a software theme or a single-stock story. The investment cycle has expanded into chips, high-bandwidth memory, data centres, electricity, cloud platforms, cybersecurity, robotics and enterprise software. As companies spend more to build and commercialise AI, the opportunity now sits across the full value chain rather than only in the most visible mega-cap names.

A broad AI ETF can be used as a simple satellite allocation for investors who want exposure to the theme without choosing individual companies. Funds such as XAIX, WTAI, AIAI, AIQU and AIQ span several parts of the chain, including semiconductors, cloud platforms, software and AI adopters. They are easier to hold than a basket of layer-specific ETFs, but the trade-off is less control over which part of the AI economy the fund emphasises.

Layer-specific ETFs serve a different role. Semiconductor and memory ETFs are closest to the AI infrastructure spending cycle. Data-centre and networking ETFs focus on the physical assets needed to house and connect AI compute. Power, grid and nuclear ETFs provide more indirect exposure to the electricity required by data centres. Cloud, robotics and cybersecurity ETFs sit further downstream, where companies attempt to turn AI investment into commercial products, automation and security tools.

The main portfolio issue is overlap. Investors already holding the S&P 500, Nasdaq-100 or a global equity ETF may already own large positions in Nvidia, Microsoft, Amazon, Alphabet, Broadcom and other AI leaders. Adding an AI ETF without checking the underlying holdings can increase concentration rather than improve diversification.

After deciding on the role, compare the ETF’s value-chain layer, top holdings, domicile, income treatment, trading currency, exchange access, fund size, bid-ask spread, brokerage cost, FX conversion and dividend withholding treatment. For Singapore-based investors, the choice between US-listed ETFs and Ireland-domiciled UCITS ETFs also affects available share classes, platform access and estate-tax considerations.

Used with a defined allocation, an AI ETF can complement a diversified equity portfolio by adding targeted exposure to one of the biggest capital-spending and productivity themes of the decade. It should not replace a broad global ETF, but it can be useful when the exposure is intentional, appropriately sized and clearly different from what the investor already owns.