The four phases of the AI investment theme: How investors think about AI adoption

Phase 1 – Core AI enablers

In this framework, the first phase begins with chips. Companies such as Nvidia and AMD design processors, TSMC manufactures advanced chips, and ASML supplies lithography equipment used in chip production. These companies are examples of the hardware layer that supports many AI workloads today.

The market treated them as growth stocks in 2023 and 2024, but they are also capital-intensive businesses tied to cyclical supply chains. Taiwan and South Korea dominate advanced chip fabrication, while the U.S. and Europe are racing to secure their own capacity. For investors, this creates exposure to potential growth drivers as well as geopolitical and supply-chain risk.

AI-related hardware demand is an important consideration for this phase, but chip demand and valuations can change quickly. More complex models may require more processing power, although demand varies by model, use case and efficiency gains. Some investors use semiconductor ETFs or individual suppliers for exposure, but both approaches carry sector and company-specific risk. Hardware demand may have structural drivers, but semiconductor markets remain cyclical.

Phase 2 – The infrastructure builders

Once the chips exist, they need power and data capacity. Large cloud providers such as Microsoft, Amazon, and Alphabet have increased spending on data centres and AI infrastructure. Energy producers and grid operators are also being monitored because data-centre growth can affect power demand and grid capacity. Utilities and data-centre REITs are examples of sectors investors sometimes monitor in relation to data-centre demand.

Power and grid connection capacity are increasingly important constraints in some regions. Large AI training runs can be very energy-intensive, but consumption varies widely by model, hardware and duration—so comparisons (such as ‘a small city’) depend on assumptions. That pressure may support investment in renewables, battery storage, and transmission upgrades, depending on regulation, financing and local grid constraints.

Some investors look to infrastructure or utilities for different risk characteristics than high-growth tech, but these assets can still be volatile and sensitive to rates, regulation and sector risks.

Phase 3 – The software monetisers

In this stage, investors look for evidence that AI spending can translate into revenue, productivity gains or margin improvement. Companies such as Adobe, Salesforce, Palantir and ServiceNow are embedding AI into their platforms, but revenue impact, customer adoption and margins vary by company. In general, companies that convert innovation into repeatable revenue may be better positioned.

Markets are beginning to separate noise from evidence. One important signal is customer adoption and whether firms are willing to pay for AI-enhanced tools. Some organisations report efficiency gains from AI in certain routine tasks, but results vary widely by use case, data quality, implementation and measurement method. Evidence of adoption may affect investor confidence, but it does not guarantee returns.

Lastly, broad software ETFs focused on automation and analytics may provide exposure to this theme while spreading company-specific risk, but they still carry market and sector risk.

Phase 4 – The productivity adopters

The fourth phase focuses on AI adoption across the wider economy. Banks, retailers, and healthcare groups are among the sectors testing or integrating AI into daily operations. Use cases include fraud detection, inventory logistics and patient analytics, although implementation, cost and results vary.

This stage is where investors assess whether AI will become part of routine business processes. Productivity gains may affect earnings over time, but the link is not automatic. McKinsey estimates generative AI could enable labour productivity growth of around 0.1% to 0.6% annually through 2040, depending on adoption and how work is reorganised. For investors, this may widen the range of sectors affected by AI adoption, but outcomes remain company-specific.

Exposure here may come through diversified equity funds or thematic ETFs focused on efficiency and automation, subject to holdings, costs, concentration and market risk. The story moves from building the technology to using it.

Market drivers behind the cycle

Valuation, liquidity and interest-rate expectations can affect how each phase is priced by markets. After the sharp rally in 2023, valuations and capital spending became more important parts of the AI investment debate. Capital spending by large technology firms on data centres and AI infrastructure has increased sharply, but estimates of the AI-linked share vary by company and methodology.

However, investor sentiment can shift quickly, especially when expectations for earnings, regulation or interest rates change. Markets may place more weight on evidence of earnings impact than on announcements alone. Regulation is tightening across regions, and interest-rate expectations continue to guide sector rotations. AI-related investment is increasingly tied to competition, regulation and capital spending, but this does not determine future winners on its own.

Building AI exposure across phases

AI is often treated as a multi-year investment theme rather than a single short-term trade. To stay invested through the full cycle, investors may layer exposure across the four phases.

Compare early- and later-stage exposure

Hardware, infrastructure, software and adopters may have different risk drivers, but none of these phases guarantees smoother returns.

Diversify by geography

The U.S., Asia, and Europe have different roles in AI software, chip manufacturing, and energy infrastructure. A global approach may provide exposure to different drivers and risks.

Manage concentration risk

The largest technology stocks now account for a significant share of several major indices. Broader exposure may help reduce reliance on a small number of mega-cap stocks, although diversification does not prevent losses.

Review ETF exposure carefully

Thematic and sector ETFs may offer targeted exposure, but holdings, concentration, fees and liquidity differ by fund. Combining semiconductor, infrastructure, and automation ETFs may also create overlap or concentration if holdings are similar.

Revisit allocations regularly

The AI cycle moves quickly. Hardware performance may weaken while adoption-related areas perform differently, or vice versa. Periodic review may help investors assess whether their exposure still aligns with their objectives and risk tolerance.

Outlook for the next stage

A key question for the next stage is whether AI investment delivers measurable returns on spending. Energy, infrastructure, and productivity metrics may become more important alongside model innovation. Countries able to secure power and chips may have strategic advantages, although policy, costs and supply chains can change.

Long-term investors may see volatility, and outcomes will depend on adoption, competition, valuation and policy—growth is not guaranteed. AI is increasingly affecting capital spending and policy discussion, but its effect on margins and valuations will vary. It is now a major market theme, but its future investment impact remains uncertain.

Closing thoughts

AI is being adopted across parts of the global economy, including software, infrastructure, supply chains, utilities and services. Thinking in phases may help investors understand where AI-related revenue, spending and risk may appear across the value chain. However, the theme can still entail high valuations, concentration risk, competition, regulatory risk, and volatile returns.

Rather than treating AI as a guaranteed growth cycle, investors should assess each company, fund or sector on its own fundamentals, valuation and risk profile.