THE AI BOOM IS NO LONGER A TECHNOLOGY STORY. IT IS BECOMING A STORY.

For the past several years, the artificial intelligence revolution has been presented primarily as a technology story.

The narrative has been compelling: unprecedented computing power, rapidly improving models, massive productivity gains, new business models and a small group of technology companies positioned to dominate the next phase of the digital economy.

But the most important question about the AI boom is no longer simply how much technology is being built.

It is becoming:

Who is financing it?

That question matters because the economics of artificial intelligence are changing.

The first phase of the AI boom was largely financed by the extraordinary cash generation and balance-sheet strength of the largest technology companies. The next phase requires an entirely different scale of capital.

Data centres, power infrastructure, advanced semiconductors, networking equipment and AI factories require enormous upfront investment. The five largest hyperscalers are expected to spend more than $1 trillion on AI-related capital expenditure across 2025 and 2026, according to the Bank for International Settlements. Their investment is increasingly exceeding what can comfortably be financed from operating cash flow alone. (Bank for International Settlements)

The consequence is already visible.

The AI boom is moving from cash-flow financing toward debt financing.

And that changes the story completely.

FROM CASH TO CAPITAL MARKETS

The extraordinary profitability of companies such as Microsoft, Alphabet, Amazon, Meta and Oracle initially made the financing question appear almost irrelevant.

These companies had something most industries could only dream of: enormous operating cash flows, high margins, investment-grade credit ratings and access to some of the deepest capital markets in the world.

But AI infrastructure is different from traditional software investment.

A software company can develop a product with relatively limited physical capital.

An AI infrastructure company—or a hyperscaler building AI capacity—needs land, buildings, electricity, cooling systems, networking infrastructure, chips and specialised hardware.

The capital intensity is therefore moving closer to that of traditional infrastructure industries.

And infrastructure has to be financed.

The BIS has documented a significant increase in debt financing by hyperscalers. Their gross corporate bond issuance exceeded $100 billion in 2025, with much of the debt issued at long maturities. (Bank for International Settlements)

The Bank of England has gone further.

Its July 2026 Financial Stability Report noted that the five major AI hyperscalers’ year-to-date investment-grade debt issuance was broadly comparable in scale with UK gilt issuance over the same period. By early May, those five companies accounted for more than 15% of year-to-date US investment-grade issuance, despite representing only around 3% of outstanding US investment-grade debt at the end of 2025. (Bank of England)

This is not a marginal development.

It means that the AI investment cycle is becoming a meaningful event in the global credit market.

THE BOND MARKET IS BECOMING PART OF THE AI INFRASTRUCTURE

This creates a fundamental shift in the relationship between technology and finance.

For decades, financial markets were primarily regarded as a mechanism supporting technological companies.

Now the relationship is becoming much more direct:

Capital markets are becoming part of the infrastructure of AI itself.

Every new data centre requires capital.

Every new generation of computing infrastructure requires capital.

Every additional gigawatt of electricity capacity requires capital.

And increasingly, a significant portion of that capital is being obtained through debt.

This creates a feedback loop.

AI expectations encourage investment.

Investment requires financing.

Financing creates debt.

Debt requires future cash flows.

Future cash flows depend on AI adoption, pricing power and productivity gains.

Therefore:

the financial sustainability of the AI boom increasingly depends on the economic return generated by the technology.

That is the point at which the technology story becomes a financial story.

THE REAL QUESTION IS NOT WHETHER AI WILL GROW

There is a temptation to frame this debate as a simple choice.

Either AI is a technological revolution or it is a bubble.

That is too simplistic.

AI can simultaneously be one of the most important technological transformations in modern economic history and contain significant financial vulnerabilities.

The internet revolution was real.

The railway revolution was real.

The automobile revolution was real.

Yet individual investments made during those revolutions did not all generate adequate returns.

The same distinction must be applied to AI.

The existence of transformational technology does not guarantee that every dollar invested in it will earn an attractive return.

That distinction becomes increasingly important as financing moves from equity toward debt.

Equity investors can theoretically absorb years of disappointing returns.

Credit investors cannot operate under the same assumptions.

Debt has contractual obligations.

Interest must be paid.

Principal must eventually be refinanced or repaid.

This creates a different economic discipline.

AI may be revolutionary. The financing of AI still has to obey the mathematics of credit.

THE HIDDEN LAYER: PRIVATE CREDIT AND “SHADOW BORROWING”

The most interesting development may not even be visible in conventional corporate bond statistics.

The BIS has identified a growing use of off-balance-sheet structures to finance AI infrastructure.

Instead of a hyperscaler directly financing every data centre, a separate vehicle can acquire or develop the facility.

That vehicle raises debt from private credit investors and other institutional investors.

The hyperscaler may then commit to a long-term lease or capacity agreement and sometimes provide guarantees.

Economically, this can resemble borrowing even though much of the debt does not appear directly on the hyperscaler’s balance sheet.

The BIS describes these arrangements as “shadow borrowing.” (Bank for International Settlements)

This is important because the headline debt figures can therefore underestimate the economic amount of capital committed to the AI infrastructure build-out.

The risk does not disappear.

It simply moves.

From the hyperscaler’s balance sheet, it can migrate into:

● private credit funds;

● special-purpose vehicles;

● infrastructure funds;

● insurers;

● banks providing financing lines;

● asset-backed securities;

● data-centre operators;

● institutional investors.

This creates a much wider financial ecosystem around AI.

And it creates connections.

A problem in AI investment could therefore eventually become a problem in credit markets even without the largest technology companies themselves experiencing immediate financial distress.

THE MOST IMPORTANT WORD MAY BE “CIRCULAR”

There is another development that deserves far more attention.

Circular financing.

The BIS has highlighted increasingly complex relationships between hyperscalers, chip manufacturers, AI laboratories and infrastructure providers.

In some cases, technology companies or chip manufacturers invest in AI companies that subsequently commit to purchasing computing capacity, chips or other services from companies connected to the original investors.

The result is a financing ecosystem in which capital, investment commitments and future revenues can circulate between participants.

The BIS has specifically warned that such arrangements can make the underlying economics more difficult to assess because the same ecosystem can simultaneously generate investment, financing commitments and expected future revenues. (Bank for International Settlements)

The recent Nvidia–MediaTek transaction illustrates why investors are increasingly scrutinising these relationships.

Nvidia announced on August 31 that it had invested $3.5 billion in convertible bonds issued by MediaTek, alongside an expanded partnership involving AI infrastructure and Nvidia’s technology ecosystem. (NVIDIA Investor Relations)

There is nothing inherently problematic about strategic investment.

The issue is analytical.

When the same companies become simultaneously:

investors + suppliers + customers + financing counterparties

traditional financial metrics become harder to interpret.

Revenue growth can still be genuine.

Demand can still be genuine.

Technology can still be revolutionary.

But the financial ecosystem surrounding that demand becomes increasingly interconnected.

And interconnected systems require investors to understand not only individual balance sheets, but also the network connecting them.

THE EUROPEAN BOND MARKET IS ALREADY FEELING THE EFFECT

One of the most revealing developments is happening far beyond Silicon Valley.

The European Central Bank recently highlighted the growing presence of US technology companies in the euro-denominated corporate bond market.

Amazon, Alphabet and Microsoft are increasingly issuing euro bonds to finance their expansion.

US technology companies now account for just under 10% of gross new euro-denominated corporate bond issuance by non-financial corporations, according to the ECB. Their share of euro “reverse Yankee” issuance has also risen sharply, while their outstanding euro-denominated bonds have reached around €40 billion. (European Central Bank)

This is an important development because it demonstrates that the AI investment cycle is not confined to the US capital market.

The financing requirements of American technology companies are increasingly competing for European savings.

And the ECB has raised an obvious question:

What happens when a small group of exceptionally large technology companies becomes a dominant borrower across multiple international bond markets?

The issue is not necessarily that investors cannot absorb the debt.

So far, they largely can.

The issue is what happens when the supply becomes structurally large enough to influence pricing across the market.

The ECB has warned that the growing presence of US technology companies could potentially crowd out other issuers and affect financing conditions more broadly. (European Central Bank)

That includes companies that have nothing to do with artificial intelligence.

And potentially, over time, sovereign borrowers as well.

AI IS NOW COMPETING WITH GOVERNMENTS FOR CAPITAL

This is where the story becomes macroeconomic.

Governments around the world are already carrying historically high levels of debt.

At the same time, they face rising demands for defence spending, infrastructure, ageing-related expenditure, energy security and industrial policy.

Now add an enormous private investment cycle in AI infrastructure.

The world therefore faces a new competition:

Governments need capital.

Technology companies need capital.

Infrastructure companies need capital.

The financial system has to allocate it.

Capital is not infinite.

Savings are not infinite.

Insurance-company balance sheets are not infinite.

Pension funds are not infinite.

Banks cannot expand credit without constraints.

Private credit cannot absorb unlimited risk without eventually demanding compensation.

The question is therefore not whether the world has enough money in absolute terms.

The question is:

At what price will capital be available to finance all these competing investment needs?

That is a very different question.

THE COST OF AI MAY NOT BE MEASURED IN DOLLARS PER GPU

Investors often focus on the cost of chips.

They should also focus on the cost of capital.

A data centre may be economically viable at one financing cost and unattractive at another.

A project generating a strong return on invested capital at a 4% cost of funding can look very different if its marginal cost of capital rises materially.

This becomes particularly important because AI infrastructure has a long economic life while parts of the technology inside it can become obsolete much faster.

The Bank of England has highlighted this mismatch as a potential vulnerability: long-term debt is increasingly being used to finance AI infrastructure, while some of the underlying servers, chips and technologies have significantly shorter economic lifecycles. (Bank of England)

That creates a classic financial question:

Are we financing short-lived technology with long-lived liabilities?

If the answer is yes, refinancing and asset-obsolescence risk become increasingly important.

THE GREAT TEST: PRODUCTIVITY

Ultimately, the debt accumulated to finance AI must be supported by economic returns.

That does not necessarily mean the hyperscalers themselves need to generate all the returns.

AI can generate value throughout the economy.

It can increase productivity.

It can reduce labour costs.

It can create new industries.

It can accelerate scientific research.

It can improve logistics and manufacturing.

It can transform financial services.

It can increase the efficiency of existing capital.

But there is a crucial difference between economic value creation and financial returns on investment.

A technology can create enormous social and economic value while producing disappointing returns for investors who paid too much for the infrastructure required to deploy it.

That distinction is one of the central financial questions of the AI era.

The BIS has already warned that the sustainability of the AI investment boom depends heavily on whether AI-related earnings eventually justify the enormous investment expectations embedded in markets. It has also noted that equity markets have moved considerably further ahead than debt markets in pricing AI optimism. (Bank for International Settlements)

This divergence deserves attention.

Equity investors are pricing extraordinary future growth.

Credit investors are increasingly asking whether the underlying cash flows will be sufficient to support the debt required to build that future.

THIS DOES NOT LOOK LIKE 2008 — AND THAT MATTERS

It would be a mistake to describe these developments as an immediate repeat of the global financial crisis.

The largest hyperscalers generally have strong balance sheets, high credit ratings and significant cash-generation capacity.

The Bank of England explicitly notes that the debt issued so far has been readily absorbed by investors, reflecting strong demand and the credit quality of the major issuers. (Bank of England)

So the argument is not:

“AI debt equals financial crisis.”

That would be intellectually weak.

The more interesting argument is:

“AI is changing the transmission mechanism through which technology investment can affect financial markets.”

The risk lies not necessarily in the first-order debt of Microsoft, Alphabet or Amazon.

It may lie further down the financing chain.

Private credit.

Data-centre financing vehicles.

Infrastructure funds.

Highly leveraged operators.

Equipment financing.

Structured products.

Banks providing liquidity.

Insurers holding long-duration assets.

And companies whose business models depend on the assumption that AI infrastructure will continue expanding indefinitely.

The more interconnected the system becomes, the more important second-order effects become.

THE NEXT AI RISK MAY COME FROM CREDIT, NOT EQUITIES

For years, investors have watched AI valuations.

That remains important.

But the next phase requires watching something else:

credit spreads.

If AI equities decline moderately, the economic consequences may remain limited.

If equity valuations fall sharply and investors simultaneously begin questioning the cash-flow assumptions behind AI infrastructure debt, the story changes.

Credit spreads can widen.

Refinancing costs can increase.

Private credit valuations can come under pressure.

Investment projects can be delayed.

Data-centre construction can slow.

Equipment orders can be cancelled.

Banks can become more selective.

And capital expenditure can fall precisely when the economy has become increasingly dependent on it.

The BIS has explicitly warned that a sharp repricing of AI-related equity risk could trigger a reassessment of corporate credit risk and tighter credit conditions, with potential consequences for aggregate investment. (Bank for International Settlements)

That is the transmission mechanism investors should understand.

THE PARADOX OF THE AI ECONOMY

There is a deeper paradox here.

AI is supposed to make the economy more productive.

But in the short term, AI may also make the economy more capital intensive.

That creates a fascinating economic tension.

The technology promises:

more output with fewer resources.

But building the technology initially requires:

more capital, more electricity, more infrastructure, more semiconductors and more financing.

The transition therefore has two phases.

Phase One

Capital expenditure → debt → infrastructure → capacity

Phase Two

Capacity → productivity → revenues → cash flow → return on capital

The financial system is effectively being asked to finance Phase One before investors know precisely how successful Phase Two will be.

That is not unusual.

Every major infrastructure revolution has involved some version of this problem.

But the scale is different.

And the speed is unprecedented.

THE CAPITAL MARKET IS BECOMING PART OF THE AI COMPETITION

There is another consequence that deserves attention.

AI competition is no longer simply a competition between algorithms.

It is increasingly a competition between balance sheets.

The companies capable of securing the cheapest and deepest sources of capital can build infrastructure faster.

Those with higher credit quality can borrow more cheaply.

Those with stronger equity valuations can raise capital more easily.

Those with strategic relationships can secure long-term supply.

Those with access to private credit can finance projects outside conventional corporate structures.

This means financial strength itself can become a competitive advantage in the AI race.

The winners may not simply be those with the best models.

They may be those with:

the best technology + the strongest balance sheet + the cheapest capital + the best infrastructure + the longest financing horizon.

That is a fundamentally different competitive landscape.

WHAT INVESTORS SHOULD WATCH NOW

The most important indicators are therefore changing.

Investors should watch at least seven things.

1. Hyperscaler free cash flow

If capex continues rising faster than operating cash flow, dependence on external financing increases.

2. Corporate bond issuance

The volume, maturity and currency of new debt matter—not just the headline amount.

3. Credit spreads

The equity market can remain optimistic while credit investors become increasingly cautious.

4. Private credit exposure

The opacity of private financing means that risk can accumulate outside conventional bond markets.

5. Data-centre economics

Occupancy, power costs, utilisation and long-term customer commitments will determine whether infrastructure generates sufficient returns.

6. AI revenue versus AI investment

Eventually, the growth in revenues and profits must justify the extraordinary level of capital expenditure.

7. Circular financing

Investors need to understand whether reported demand is genuinely independent or partly supported by interconnected financing and purchasing commitments.

These indicators will tell us considerably more about the sustainability of the AI boom than the daily movement of any single AI stock.

THE BIGGER ECONOMIC QUESTION

The AI revolution is now entering a new stage.

The first stage was about technology.

The second was about valuation.

The third is increasingly about capital.

And that transition matters.

The world is simultaneously facing enormous government borrowing requirements, rising infrastructure needs, geopolitical uncertainty and a historically large private investment cycle in artificial intelligence.

The competition for capital is therefore becoming one of the defining economic questions of the next decade.

The AI boom may succeed.

Indeed, it may transform global productivity more profoundly than most current forecasts anticipate.

But success does not eliminate the financing problem.

It makes understanding it more important.

Because every technological revolution ultimately faces the same fundamental question:

How much capital must be invested before the promised productivity arrives—and who carries the risk while we wait?

That is the question investors should be asking now.

Not whether AI is real.

It is.

Not whether AI will change the economy.

It will.

The question is whether the financial architecture being built around AI can generate returns commensurate with the enormous amount of capital now being committed to it.

The answer will determine not only the future of technology companies, but potentially the next phase of global credit markets.

The AI revolution is no longer being financed only by innovation.

Increasingly, it is being financed by debt.

And once technology becomes dependent on credit, the AI story becomes a story about finance.

That is where the next great risk—and potentially the next great opportunity—may emerge.

Roberto Balbino
CEO, Fintech SCB
Independent Analyst in Finance, Economics, Geopolitics and Global Markets