Central banks face AI-driven growth and stability challenges, says ECB’s Hernández de Cos

SINGAPORE, 22 SEPTEMBER 2026 —

ECB board member Pablo Hernández de Cos warned that artificial intelligence is reshaping global macro-financial conditions, driving a projected surge in investment from approximately $500 billion today to between $3 trillion and $4 trillion by 2030.

Speaking at the Reserve Bank of India’s fintech conference, Hernández de Cos noted that AI has unleashed a broad investment boom in data centres, semiconductor manufacturing, and cloud infrastructure. He explained that optimism around the technology has helped keep global financial conditions accommodative, supporting household consumption through wealth effects.

According to BIS research cited in his remarks, the five largest big tech companies are set to spend over $1 trillion on AI-related capital expenditure between 2025 and 2026.

The central bank official highlighted that this investment boom is increasingly financed by debt and private credit rather than earnings. He pointed out that global trade is being reshaped as exports of goods with high AI content grow rapidly, largely due to rising prices. Hernández de Cos observed that this has created uneven terms of trade, with upstream exporters like Malaysia, Korea, and Singapore gaining as their export prices rise more than import costs.

However, he cautioned that these gains are narrowly concentrated. In Korea, for instance, five firms accounted for around 43% of export earnings in the first quarter of this year, up from 27% two years earlier. Conversely, economies investing heavily in their own digital infrastructure face higher costs for AI-related equipment, which worsens their terms of trade.

On the productivity front, Hernández de Cos cited empirical evidence showing generative AI delivers productivity gains of between 10% and 65% in specific tasks. These gains are particularly evident in coding, consulting, and professional writing, with less experienced workers seeing relatively larger benefits within narrowly defined tasks. The median estimate in the literature points to an increase of around half a percentage point per year in aggregate total factor productivity growth.

Hernández de Cos noted that preliminary cross-country evidence suggests economies better prepared for AI in 2023 have recorded stronger labour productivity growth since then. He identified two simultaneous forces in labour markets: AI complements human labour in roles requiring judgment but substitutes for workers performing routine cognitive work. In developing economies with large informal sectors, the balance between these risks and benefits remains tenuous.

Although actual displacement has been limited so far, early signs are emerging in customer support, programming, and administrative work. Recent earnings calls indicate that nearly 80% of firms discuss plans to automate production processes and increase labour substitution. Hernández de Cos emphasized that the priority is not to resist the technology but to prepare people for it through reskilling and retraining.

For Malaysia, the implications are mixed as an upstream exporter benefiting from rising prices for AI-linked goods, yet facing higher equipment costs for domestic digital infrastructure. The concentration of export gains among a few firms in similar economies like Korea suggests that Malaysia may see benefits clustered in specific high-tech sectors rather than spreading broadly across the economy.

The remarks highlight that central banks must monitor how AI adoption affects financial stability, particularly as leverage rises in the AI sector. Hernández de Cos argued that policy responses must address both the potential for significant productivity gains and the risks of uneven distribution of those gains across sectors and countries.

He concluded that the macroeconomic effects of AI will vary significantly depending on an economy’s sectoral composition and its preparedness, including digital infrastructure and regulatory frameworks. For policymakers, the challenge is ensuring that AI augments human labour rather than substituting for it, especially in economies where social protections for informal workers are limited.

Related: Pablo Hernández de Cos