Gazeta Lodz - AI's transatlantic debt trap

The borrowing behind artificial intelligence is creating financial obligations that reach far beyond Silicon Valley. A reversal could damage American growth, strain public finances and transmit losses to Europe. The danger deserves scrutiny, although claims of an inevitable US bankruptcy go beyond the evidence.

The most consequential feature of the artificial intelligence boom may ultimately be its financing. Behind the expanding capabilities of software models lies an enormous physical investment programme: data centres, processors, power infrastructure and the networks connecting them. These assets require money today. The revenues expected to justify them must arrive over many years.

That gap between immediate expenditure and future earnings is where technological optimism becomes financial risk. A business can develop useful products and still earn too little to service the obligations accumulated while building them. An economy can benefit from a new technology while suffering substantial losses on the investments made in anticipation of its success. By November 2026, the scale of borrowing had made that distinction increasingly important. Debt issuance associated with advanced computing infrastructure exceeded $680 billion through October, compared with $130 billion during the whole of 2023. Its share of nonfinancial investment-grade corporate debt issuance had risen from less than 6 per cent to approximately 31 per cent.

Those figures describe financing, rather than a direct measure of AI revenues or productive capacity. Nevertheless, they show how substantially credit markets have become involved in underwriting the expansion. If earnings disappoint, the consequences will extend beyond shareholders willing to accept the risks of a speculative technology investment.

The financial commitments extend beyond conventional debt.
The balance sheet offers only a partial view of the obligations surrounding the build-out. By the second quarter of 2026, seven major technology companies had accumulated more than $1 trillion in undiscounted commitments for leases that had yet to begin, alongside over $2 trillion in other obligations, including guarantees and contractual commitments. These amounts are neither immediately payable debts nor directly comparable with discounted balance-sheet liabilities. They nevertheless represent future claims on resources. Their significance becomes clearer when a project’s anticipated income weakens while its contractual obligations remain in place.

Consider the economics of a data centre financed against a long-term customer commitment. The contract helps the developer borrow and gives lenders confidence that the completed facility will have an occupant. But the arrangement can also bind the customer to payments long after its expectations about computing demand have changed. The financing distributes exposure among different entities without necessarily eliminating the underlying commercial risk.

Technological progress can complicate the calculation. More efficient models and cheaper computing could make AI services more accessible, benefiting customers and encouraging adoption. For an investor who financed expensive equipment on the assumption of sustained scarcity and premium prices, the same development could weaken returns. Technological success and investment success need not arrive together. Interlocking investments add another layer. Fourteen publicly disclosed transactions since 2023 linked approximately $300 billion in equity investment and $650 billion in debt with $1.4 trillion in future revenue commitments. These arrangements connect funding decisions with subsequent purchases of computing services or chips.

The analytical concern is that several apparently separate assets may depend on the same underlying customer. If that customer struggles, the supplier could face weaker sales, a declining investment valuation and exposure through a guarantee at the same time. Revenue commitments can support construction, but they cannot substitute indefinitely for commercially sustainable demand.

How a private investment reversal could reach the public finances
An AI downturn would initially affect companies, investors and lenders. Its wider economic impact would depend on the scale of cancelled investment, the distribution of losses and the ability of the financial system to continue extending credit. The first transmission channel would be business spending. A company that postpones a major computing project also reduces orders for equipment, construction and related services. Suppliers facing weaker demand may then defer their own investment or hiring. The effect can spread well beyond the firms that originally misjudged the market.

A second channel would run through household wealth. Falling equity valuations can weaken consumption, particularly when losses affect households with substantial stock-market exposure. A third would operate through credit: lenders facing losses or greater uncertainty may tighten financing conditions even for borrowers with little direct involvement in AI.
None of these mechanisms requires the technology itself to fail. A sufficiently large gap between expected and realised profitability would be enough.

The United States would enter such a downturn with an already substantial fiscal deficit. In the financial year ending on September 30, 2026, the federal budget shortfall was approximately $2 trillion, an increase of $217 billion from the previous year. That deficit cannot reasonably be attributed to AI. It does, however, define the fiscal starting point from which Washington would have to respond to a major economic shock.

A recession would tend to reduce tax receipts while increasing expenditure on income support and other automatic stabilisers. Emergency assistance, if policymakers chose to provide it, could create additional costs. Private losses would not automatically become public liabilities; that would depend on the institutions affected, existing guarantees and subsequent political decisions.

The claim that AI could “bankrupt America” therefore requires precision. The federal government is a sovereign borrower issuing debt in its own currency, with taxing powers and access to a monetary system fundamentally different from those of a company. A technology-sector collapse would not mechanically render it insolvent. The more credible danger is a deterioration in the relationship between growth, borrowing costs and public debt. Weaker economic activity could worsen the budget balance just as investors became more sensitive to fiscal risk. Conversely, a flight towards safe assets could lower Treasury yields and cushion part of the shock. The outcome would depend on whether the downturn primarily damaged growth expectations or also undermined confidence in public finances.

Europe is already connected to the financing cycle.
Europe’s exposure cannot be assessed simply by counting its own AI developers. European investors participate in American equity markets, purchase technology-company bonds and finance businesses throughout the infrastructure supply chain. A shock originating in US computing investment could therefore cross the Atlantic through portfolios and credit markets before appearing in trade figures.

American technology companies are also raising funds directly in European currencies. Alphabet launched a second euro-denominated bond sale of 2026 in early November, following a €9 billion issue in May. This is a concrete connection between American investment plans and European pools of savings.

Additional issuance can give investors access to a wider range of highly rated securities. It can also intensify competition for capital. If investors demand higher yields to absorb a larger supply of bonds, other borrowers may face less favourable financing conditions. That is a potential market effect, rather than proof that every technology-company bond displaces a European industrial investment.

There is also a domestic European credit exposure. Estimates for the first quarter of 2026 attributed approximately 0.9 percentage points of annual aggregate credit growth to the AI boom, roughly one quarter of the total. European AI developers and adopters had increased their leverage more than comparable businesses without similar exposure.
This does not establish that those loans are unsound. European AI firms also tend to hold larger cash buffers than comparable peers, and internal financing remains important. The figures nevertheless demonstrate that Europe is participating in the investment cycle through its own corporate borrowing as well as through overseas assets.

A deterioration in American demand could create a further channel. European suppliers serving the global computing industry might face weaker orders, while businesses elsewhere in the economy could be affected by a broader US slowdown. The severity would vary considerably across countries and sectors. Treating Europe as a single, uniformly exposed balance sheet would obscure those differences.

The crucial distinction is between a market correction and a systemic crisis.
A fall in technology shares, even a large one, would not automatically produce a banking emergency. Equity investors can absorb losses without interrupting payments, deposits or ordinary business lending. Strong corporate cash flows, adequate bank capital and stable funding could prevent an investment correction from becoming a general financial crisis.

The outcome becomes more dangerous when losses meet leverage and urgent cash demands. An investor required to post additional collateral may sell assets unrelated to the original problem. A lender unable to establish the value of its security may withdraw financing. Several institutions trying to reduce similar exposures simultaneously can push prices down further. The central question is therefore who ultimately bears the loss and whether that institution has the resources and time to absorb it. The nominal size of announced AI investment cannot answer that question. Neither can a headline valuation or the credit rating of the strongest company in a financing chain.

There are substantial reasons to resist a deterministic collapse narrative. Large technology groups have profitable established businesses. AI adoption can improve productivity and create revenue outside the infrastructure industry. Lower computing costs could expand demand enough to support continued growth, even if some early investments prove uneconomic.

But those possibilities do not remove the financing challenge. Useful infrastructure can be built at an excessive price. Strong long-term demand can coexist with weak near-term cash generation. A commercially promising project can become distressed if its repayment schedule leaves too little room for delays.

The warning for Washington and Europe is consequently specific. As AI investment becomes more dependent on debt, guarantees and long-term commitments, a disappointment in earnings can create obligations that are harder to absorb than a decline in share prices alone. Public finances could suffer through slower growth and crisis intervention, while European investors and businesses could incur losses through several connected markets. The decisive test will be whether durable customer revenues grow fast enough to support the financial promises already made. The technology may continue advancing even if parts of its investment boom unravel. America and Europe would still have to deal with the bill.