AI has entered a full-scale infrastructure build-out phase. We expect annual AI-related infrastructure spending to exceed US$400–500 billion by 2026, driven by accelerated data-centre construction, higher-density compute requirements, and rising power and cooling needs. At this level, AI infrastructure investment approaches ~2% of US GDP, placing it alongside past general-purpose technology cycles such as cloud computing and telecommunications.

However, this remains a front-loaded capital cycle. Cash outflows precede revenue, and monetisation remains uneven across sectors. For current equity valuations to be sustained, the AI ecosystem must ultimately generate US$1.7–2.5 trillion in incremental annual revenue by the end of the decade.

As infrastructure spending accelerates into 2026, balance-sheet discipline becomes increasingly important. Large cloud and infrastructure providers operate with average debt-to-equity ratios around 0.5x, higher than pre-AI levels. A further ramp-up in capex raises sensitivity to funding costs, utilisation rates, and monetisation timelines. The risk is no longer whether AI adoption happens, but whether capital intensity runs ahead of returns at the margin.

While monetisation remains uncertain, consensus is building that AI represents the next general-purpose technology capable of lifting productivity by 1–3% per annum. With adoption still at an early stage, this supports sustained optimism around the theme.