The first phase of the AI boom has funnelled an extraordinary amount of capital into a handful of US mega-caps. The early spoils have gone to a familiar trio, with NVIDIA, Microsoft and Alphabet dominating buy‑lists and performance tables alike, while the high‑profile IPOs of Anthropic and OpenAI look set to draw more money into an already crowded trade.
The concentration risk, however, is becoming harder to ignore. UBS estimates that hyperscalers will spend the equivalent of 100% of operating cashflow in 2026, up from a 10-year average of 40%. And that means those supposedly asset-light, cash-generative machines are starting to look distinctly more capital-intensive. With valuations leaving little margin for disappointment, betting solely on the US mega-caps is overlooking a much broader field of AI beneficiaries.
One alternative is to follow the money into the physical infrastructure required to train and run large language models, rather than betting on the winners of software arms race. Data centres have become the engine rooms of the AI economy, with McKinsey estimating that data centre capex could reach almost $7trn by 2030 (including enough fibre-optic cable alone to circle the Earth 120 times).
Energy supply is another potential constraint: data centres consume vast amounts of energy, with the cooling needed to manage heat-intensive AI chips often rivalling the power needed for the computing itself. The International Energy Agency forecasts that data centre electricity demand will more than double by 2030, which, for context, exceeds Japan's current consumption.
In short, the second wave of AI investment is taking place well beyond the mega‑cap software names dominating the headlines.
The commodities supercycle
One of the clearest beneficiaries is the commodities sector, with the AI build-out adding to soaring demand for copper, aluminium, silver, silicon and other industrial metals. Copper has emerged as the critical resource for electricity transmission, data centres and network infrastructure, not to mention electrification, renewable energy and defence spending.
This creates a bottleneck as physical supply can’t be quickly scaled like software, with new mines often taking more than 15 years from discovery to production. Added to this, companies have focused on paying down debt over investing in exploration and production in recent years, meaning that supply constraints are likely to persist in the near term.
While the long-term structural demand and supply dynamics may be attractive, it’s rather more challenging to capture it in a portfolio: single-commodity strategies come with high volatility, while individual miners carry operational and geopolitical risk.
One way to spread these risks is through a diversified portfolio of mining companies, which offers exposure to underlying commodity prices alongside operational leverage. BlackRock World Mining Trust (BRWM) operates as a quasi-virtual mining company, holding public and private mining assets across industrial and precious metals. The trust looks to capture the potential upside from higher-returning commodities with strong structural growth drivers, while managing the downside risk of cyclical and non-cyclical volatility.
The trust’s largest weighting is gold, which offers a potential hedge during periods of market volatility, and has benefitted from central bank demand, as well as a significant weighting to copper to capture the AI tailwinds. BRWM has achieved an 80%-plus return in the last year and currently offers a 3% dividend yield.
Feeding the beast
The race is on to build the data centres, fibre networks and power grids required to meet growing AI demand from both businesses and consumers, and the closed-ended structure of investment trusts lends itself well to holding less liquid infrastructure assets.
On the equity side, Cordiant Digital Infrastructure (CORD) owns a diversified portfolio of data centres, fibre networks and towers across the US, Ireland and Central Europe, including a flagship site in Prague with the potential to become the largest data centre in the Czech Republic. Its ‘buy, build and grow’ strategy combines in‑house operational and private‑equity expertise to drive earnings growth, with a five-year return of over 45%.
Another angle is infrastructure debt, which offers attractive yields in the aftermath of traditional banks pulling back from long‑term project lending post the global financial crisis. Sequoia Economic Infrastructure Income (SEQI) invests across more than 50 assets spanning hyperscale data centres, renewable energy and transport. It currently yields just under 9%, offering a healthy premium to high-yield bonds to compensate investors for taking credit risk.
Both trusts are currently trading at discounts to NAV of more than 10%, which may offer an attractive entry point if the downstream beneficiaries of the AI boom begin to re-rate.
Away from the crowd
The temptation is to focus on the headline‑grabbing winners of AI, but this comes at the expense of single-stock risk and the challenge of picking the eventual champions of the AI race. With concerns that the AI frenzy has stretched mega-cap valuations, the second-order beneficiaries doing the heavy lifting may prove a rich source of returns.
Jo Groves is an investment specialist at Kepler Trust Intelligence. The views expressed above should not be taken as investment advice.