Investing in AI: The Best Ways to Get Exposure to the AI Economy
AI is the defining investment theme of the 2020s — but “invest in AI” is advice too vague to be useful. The AI economy has multiple layers, each with different risk/return profiles, different time horizons, and different exposure characteristics. Some of the biggest gains are already in the rearview mirror — Nvidia is up over 2,000% from its 2022 lows. But the opportunity set in AI remains enormous, and there are smart ways to position a portfolio for the decade ahead. This article, part of our AI and the economy series, maps the landscape practically.
- → AI investing has five distinct layers: chips, infrastructure, foundation models, enterprise software, and sector-specific applications — each at different stages of the investment cycle
- → The “picks and shovels” infrastructure approach — power, data centres, chips — tends to be lower-risk than betting on which AI software companies will dominate
- → ETFs offer broad AI exposure for investors who want the theme without single-stock risk; the main options span from pure-play AI to tech-heavy broad indices
- → Valuation is the key risk: many AI-exposed stocks trade at historically elevated multiples; a correction could be severe even if the underlying technology delivers on its promise
- → Dollar-cost averaging into AI exposure over 12–24 months is likely more prudent than lump-sum entry at current valuations
The Five Layers of AI Investment Opportunity
The most useful framework for AI investing is to think in layers — from the physical foundation up to the end applications. Each layer has a different competitive structure, revenue profile, and investment timing.
| Layer | What It Is | Key Names | Stage |
|---|---|---|---|
| 1. Chips | AI training & inference hardware | Nvidia, AMD, TSMC, ASML | Peak valuation — high risk/reward |
| 2. Power & Infrastructure | Data centres, electricity, cooling | Vertiv, Eaton, NextEra, uranium plays | Early-mid cycle — significant upside |
| 3. Cloud / Hyperscalers | AI compute delivery platforms | Microsoft, Google, Amazon, Meta | Mid cycle — diversified exposure |
| 4. Foundation Models | Core AI model developers | OpenAI (private), Anthropic (private), Google DeepMind | Mostly private — limited direct access |
| 5. Applications & Software | AI tools for specific use cases | Salesforce, ServiceNow, Palantir, sector-specific | Early cycle — highest uncertainty, highest potential |
The Picks-and-Shovels Logic
During the California Gold Rush, the merchants who sold picks, shovels, and denim trousers made more reliable fortunes than most of the miners. The same logic applies to AI: rather than betting on which AI software company wins the application layer — a genuinely uncertain question — investors can gain exposure through the physical and infrastructure layer that all AI requires regardless of which models or applications prevail.
“You don’t need to know which AI company wins. You need to know that all of them will need power, chips, and data centres. That’s the picks-and-shovels trade — lower variance, still significant upside.”
The power infrastructure opportunity is particularly compelling and underappreciated. As covered in our AI investment boom article, data centres are projected to consume 8–10% of US electricity by 2030. Utilities, electricity transmission infrastructure, and nuclear power operators are direct beneficiaries that many retail investors overlook while focusing on Nvidia and Microsoft.
ETF Approaches to AI Exposure
For investors who prefer diversified exposure without the risk of single-stock concentration, several ETF approaches exist:
Pure-play AI ETFs (e.g. BOTZ, ROBO, ARKQ) concentrate on robotics and AI companies directly. Higher beta, more volatile, higher fees. Broad tech ETFs (e.g. QQQ, VGT) provide AI exposure through the hyperscalers and chipmakers as a significant weighting. Lower fees, more diversification. Thematic infrastructure ETFs targeting data centres, power grids, or semiconductors capture the picks-and-shovels angle with sector-specific focus. Always compare expense ratios, holdings, and liquidity before investing.
The Key Risks to Manage
Valuation risk. Many AI-exposed stocks trade at price-to-earnings ratios that price in years of flawless execution. Nvidia at peak traded at 35x forward earnings — high, but justifiable given growth rates. Some smaller AI-adjacent companies trade at 50–100x revenues with no clear path to profitability. A market-wide derating of growth multiples — triggered by rising interest rates, disappointing earnings, or macro deterioration — could compress AI valuations sharply even if the technology continues to deliver.
Competition and commoditisation risk. The foundation model layer is particularly exposed to commoditisation. If AI models become interchangeable utilities — like cloud computing or internet bandwidth — the economics will favour the lowest-cost provider, compressing margins across the sector. Open-source models (Meta’s LLaMA, Mistral) are already exerting downward pricing pressure on proprietary model APIs.
Regulatory risk. The EU AI Act is already in force. US regulatory frameworks are developing. Sector-specific AI applications in healthcare, finance, and legal services face additional oversight. Regulatory restrictions could delay or limit monetisation in key verticals.
Practical Portfolio Approach
For most investors, a pragmatic AI allocation combines core broad-market index exposure (which already gives significant AI weighting through the Magnificent Seven), a modest thematic allocation to infrastructure plays (power, data centres, chips), and selective individual stock positions in high-conviction application-layer companies with demonstrated revenue traction. Sizing AI as 10–20% of a portfolio — rather than going all-in — captures the theme while managing the very real risk that current valuations overshoot. For broader investment strategy principles, see our overview of self-directed investing approaches.
AI is a genuine multi-decade investment theme — but the entry point matters enormously, and the choice of which layer to invest in matters even more. The biggest near-term risk is that current valuations for AI software companies price in outcomes that may take a decade to materialise. The most resilient opportunity remains infrastructure: power, data centres, and chips. Dollar-cost averaging into broad AI exposure over 12–24 months, rather than lump-sum allocation at peak enthusiasm, remains the most sensible approach for most investors.
Responses