The AI Boom Is Becoming So Huge It Could Threaten the Financial System

AI Investment Is Reaching Macroeconomic Scale

The Trillion-Dollar AI Boom Meets Its Biggest Warning Yet

The extraordinary global investment race to build artificial intelligence is becoming large enough to create risks for the wider financial system, according to one of the world's most influential central banking institutions. Bank for International Settlements General Manager Pablo Hernández de Cos warned on 10 September that soaring valuations, growing debt, opaque financing and an increasingly concentrated AI industry could magnify the damage if expectations surrounding the technology ultimately disappoint.

This is not a prediction that artificial intelligence is about to collapse. Hernández de Cos explicitly stopped short of that conclusion. The warning is more significant in another sense: AI has grown from a technology story into a macroeconomic force whose investment, financing and stock-market footprint are now large enough to matter to central banks and potentially to global financial stability.

AI Investment Is Reaching Macroeconomic Scale

The numbers explain why policymakers are paying attention. The five largest global technology companies are expected to spend more than $1 trillion on AI-related capital expenditure across 2025 and 2026, while industry expectations cited by the BIS suggest global AI-related investment could rise from roughly $500 billion today to between $3 trillion and $4 trillion by 2030.

Money is pouring into data centres, specialised semiconductor manufacturing, servers, networking equipment, cloud infrastructure and the electricity systems required to support increasingly powerful models. That spending is already affecting economic growth, trade and financial markets rather than operating as a small technology-sector investment cycle.

AI optimism has also helped support equity valuations and broader financial conditions. That creates an important feedback loop: investment raises expectations about future AI profits, rising valuations increase investor wealth and financial confidence, and those conditions can support additional spending and investment.

The cycle works remarkably well while expectations keep rising.

The harder question is what happens if they stop.

The Boom Is Increasingly Being Financed With Debt

One of the biggest differences between the early stages of the AI boom and the investment race now developing is how the infrastructure is being financed.

The largest technology companies initially had enormous cash flows available to fund data centres, chips and computing infrastructure. But the scale of required investment has become so large that spending is increasingly moving beyond internally generated earnings towards borrowing, including private credit.

That matters because debt can amplify both sides of an investment cycle.

Borrowed money allows companies to expand faster when revenues and valuations are rising. But debt still has to be serviced when expected revenues disappoint, assets fall in value or customers cut spending.

BIS research has argued that the AI build-out ranks among the largest technology-driven investment booms in US history. Its modelling suggests intense competition between companies attempting to secure dominant positions can encourage them to commit more capital than would otherwise be economically efficient.

That does not mean those investments will necessarily fail. AI could generate productivity gains large enough to justify an extraordinary amount of spending.

It means the required payoff is becoming extraordinarily large too.

Why Circular Financing Worries Policymakers

Debt is not the only concern. The relationships connecting companies inside the AI ecosystem are becoming increasingly complicated.

Hernández de Cos highlighted what is sometimes described as circular financing. Chip manufacturers and hyperscale technology companies can invest in AI businesses which, in turn, agree to purchase computing capacity, chips or other infrastructure from companies linked to their investors.

Individually, such arrangements can make commercial sense. Collectively, they can make it harder to determine where underlying economic demand ends and financially interconnected demand begins.

The danger appears if stress at one important company passes through the same relationships that helped finance expansion.

BIS research has examined this possibility using network analysis and found that financial connections between companies could allow stress at one participant to spread through chains of exposure elsewhere in the sector.

The AI economy is therefore becoming not only enormous but interconnected.

That is where a technology boom starts becoming a financial-stability question.

What Happens If AI Returns Disappoint

The central risk is straightforward.

Companies are investing vast amounts of capital today because they expect artificial intelligence to produce correspondingly vast economic returns tomorrow.

If those returns arrive, the infrastructure could support one of the most significant productivity transformations of the modern economy. Task-level research cited by Hernández de Cos has already found productivity improvements ranging between 10% and 65% in some settings, alongside substantial time savings, particularly in areas such as programming, consulting and professional writing.

But translating improvements on specific tasks into economy-wide productivity growth is much harder.

If commercial returns undershoot what investors have priced in, companies could reduce spending. Suppliers dependent on the infrastructure boom could then lose revenue, highly valued technology shares could fall and heavily indebted participants could face greater pressure.

Those effects could spread well beyond Silicon Valley.

Households hold considerable wealth through equities and retirement investments, while American shares make up a huge proportion of global stock markets. The BIS therefore argues that a substantial repricing of AI companies could hit household wealth and consumption while transmitting the correction internationally.

History Offers an Uncomfortable Comparison

Hernández de Cos pointed to a pattern that predates artificial intelligence by nearly two centuries.

The canal investment mania of the 1830s, British railway boom of the 1840s, electrification boom of the 1920s and dotcom expansion of the late 1990s were all associated with genuine technological breakthroughs. Yet each also attracted investment that eventually exceeded the financial returns some projects could support.

That distinction is important.

A financial bubble does not require the underlying technology to be useless.

Railways transformed the world.

Electricity transformed the world.

The internet transformed the world.

Investors could still lose enormous amounts of money building those revolutions.

Artificial intelligence could therefore become economically transformative while parts of the investment boom surrounding it simultaneously prove excessive.

The technology succeeding and investors overpaying for that success are not mutually exclusive outcomes.

AI Is Also Concentrating Economic Power

Another vulnerability comes from concentration.

AI infrastructure has extremely high fixed costs. Developing frontier models, manufacturing advanced semiconductors and constructing hyperscale computing infrastructure requires enormous amounts of capital, electricity and specialised expertise.

Those economics favour very large companies. The BIS says the dominant AI groups are concentrated mainly in the United States, China, Chinese Taipei, Korea and the Netherlands, with several leading companies operating across multiple stages of the AI supply chain.

That concentration creates spectacular winners when the market rises.

It can also concentrate financial consequences when expectations reverse.

The pattern is already visible in trade. Hernández de Cos noted that just five firms accounted for around 43% of South Korea's export earnings in the first quarter of 2026, compared with 27% only two years earlier, demonstrating how rapidly AI-related industries can reshape the exposure of entire national economies.

Central Banks Face a New Problem

Artificial intelligence poses another challenge even if the investment boom never crashes.

It is changing several parts of the economy simultaneously.

AI can increase productivity while replacing some jobs. It can create enormous investment demand while eventually reducing production costs. It can push up demand for electricity, construction and semiconductors while potentially lowering the cost of many digital services.

That makes inflation, growth and interest-rate signals harder for central banks to interpret.

Hernández de Cos said AI is reshaping demand, supply and financial markets at the same time, increasing uncertainty around monetary-policy transmission and making indicators such as potential economic output and the natural rate of interest more difficult to estimate.

The AI revolution could therefore make economies more productive while temporarily making them harder to manage.

AI Could Still Deliver the Productivity Boom Investors Expect

The warning should not be mistaken for an argument against artificial intelligence.

Evidence of productivity benefits is accumulating, while countries with stronger digital infrastructure and greater preparedness for AI have already shown indications of faster productivity growth. Advanced economies with large professional-services, financial and information sectors may be particularly well positioned to capture the early gains.

The BIS's concern is instead that technological potential does not eliminate financial risk.

The larger the investment boom becomes, the greater the amount of future income required to justify today's expenditure and valuations.

That creates perhaps the defining financial question of the AI age.

It is no longer simply whether artificial intelligence works.

It is whether AI can become profitable enough, quickly enough, to justify the trillions of dollars now being committed to building it.

What Happens Next

AI investment is unlikely to stop because central bankers have identified risks. Competition between the world's largest technology companies is itself one of the forces encouraging them to continue spending, because slowing down may mean surrendering ground to a rival.

That means attention will increasingly turn towards how much infrastructure is financed through debt, whether AI revenues catch up with capital expenditure, how concentrated financial exposures become and whether productivity gains escape individual businesses and begin appearing throughout entire economies.

Hernández de Cos did not declare that an AI bubble is about to burst. His warning is subtler and arguably more important: the scale and speed of the boom have become large enough that the possibility can no longer be treated as somebody else's technology problem.

Artificial intelligence may ultimately justify one of the greatest investment waves in modern economic history. But with trillions of dollars moving towards infrastructure, valuations concentrated among a small group of companies and more of the expansion financed through debt, the consequences of being wrong are becoming global too.