TL;DR

The head of the Bank for International Settlements has warned that AI infrastructure spending is now large enough to move the global economy, and that a growing share of it rests on debt rather than profits. Pablo Hernandez de Cos singled out stretched valuations, market concentration and financing structures nobody can see into. He stopped short of predicting a bust.

The numbers behind the warning

The BIS reckons the five biggest technology companies will commit upwards of $1 trillion to AI across 2025 and 2026. Industry projections stretch further, from roughly $500 billion annually today towards $4 trillion by the end of the decade.

His concern is less the amount than its source. Where this spending was once covered by corporate earnings, borrowing and private credit are taking over — and much of that funding, he said, remains “opaque and interconnected”. Those two words are doing considerable work: opacity prevents supervisors sizing the exposure, and interconnection is how a localised disappointment becomes a systemic one.

For central banks specifically, he argued AI does not alter the mandate but does make the economy harder to read, striking supply, markets and demand simultaneously.

Productivity, jobs and trade

Hernandez de Cos was not uniformly gloomy. He pointed to studies finding task-level gains between 10% and 65%, concentrated in professional writing, consulting and coding, while noting that economy-wide translation is the harder question — current estimates put the effect on total factor productivity growth at roughly 0.5 percentage points annually.

On employment, he said losses so far are contained but visible in administrative roles, programming and customer support. Trade patterns are shifting too, with Taiwan, Malaysia, Singapore and South Korea all gaining from firmer export prices on AI hardware.

Looking forward

The historical framing is the part worth keeping. He set current conditions alongside the railway expansion and the dotcom surge — episodes where the underlying technology genuinely mattered and the financing still went badly wrong. “I do not say that this is where the AI boom must lead,” he said, while arguing the pace warrants caution.

UK relevance is direct. British financial services sit downstream of this credit, and the country’s datacentre pipeline has reached 173 projects and 14.6GW. Those projects need funding on terms the BIS has just described as hard to examine.