AI’s Financing Boom Is Becoming a Balance-Sheet Story

By Robert Young · Draft prepared September 14, 2026 · 8:30 AM PT

Artificial intelligence has largely been treated as an investment story: how much companies will spend, which models will win, how large the productivity opportunity might be, and whether today’s technology valuations are justified by tomorrow’s earnings.

Finance leaders should add another question: Who is financing all of this—and what happens if the returns arrive later than the obligations?

That question became more urgent this week. In a September 10 speech, Bank for International Settlements General Manager Pablo Hernández de Cos said the largest technology companies’ capital spending is beginning to outpace cash flow, pushing the AI investment boom toward greater reliance on debt and private credit. He also highlighted opaque and interconnected financing structures, including arrangements in which chipmakers and hyperscalers invest in AI companies that then commit to buying their chips and computing capacity. BIS, September 10, 2026

Today, Reuters reported that the BIS sees signs of vulnerability in AI-driven market momentum as investors weigh rising technology-sector debt, opaque financing structures and the possibility that expected AI profits arrive more slowly than anticipated. Reuters, September 14, 2026

From an earnings question to a balance-sheet question

For much of the AI cycle, the downside case has centered on valuation. If AI does not generate the revenue or productivity investors expect, technology companies may simply deserve lower earnings multiples.

Debt changes the equation.

Once expected future AI cash flows are matched against fixed financial commitments, disappointment can affect more than equity value. It can affect liquidity, refinancing capacity, covenant headroom and the availability of capital itself.

The BIS has been building this case throughout 2026. Its Annual Economic Report says direct-lending funds have quadrupled lending to AI and information-technology sectors over five years, to roughly 15% of their portfolios. The BIS noted that these loans tend to be larger than loans in other sectors while terms remain broadly similar—raising questions about lending standards and risk pricing. BIS Annual Economic Report 2026

Separate BIS research found that outstanding private-credit loans to software-as-a-service companies rose from about $8 billion in 2015 to more than $500 billion by the end of 2025, equal to roughly 19% of total direct loans. BIS Quarterly Review, March 2026

And in July, the BIS estimated that business development companies had lent about $115 billion to software firms—roughly one-fifth of their lending and more than 80% of their technology portfolios—while spreads had narrowed and exposure had become concentrated among a relatively small group of large lenders. BIS Bulletin 128, July 14, 2026

The important conclusion is not that private credit is bad or that AI investment is necessarily a bubble. It is that the financing structure is becoming part of the investment thesis.

Good technology can still be a bad capital allocation

CFOs encounter this distinction constantly. An investment can be strategically useful and still fail financially.

A new system might increase productivity but take seven years to repay an investment justified on a three-year assumption. A data-center contract might eventually support enormous demand but create years of fixed expense before utilization catches up. An AI implementation can produce real efficiency while generating far less incremental revenue than its original business case assumed.

None of those outcomes means the underlying technology failed. They mean the capital-allocation assumptions were wrong.

That distinction matters because enormous amounts of capital are being committed before the industry’s ultimate economics are known. The BIS estimates that the five largest technology companies alone are on track to spend more than $1 trillion on AI-related capital expenditure across 2025 and 2026. BIS

The useful finance questions are not simply whether AI creates value, but how much value, how quickly, for whom—and against how much committed capital.

Debt eliminates some of the luxury of waiting

Equity capital can tolerate uncertainty for a surprisingly long time. Debt usually cannot.

Interest payments arrive whether utilization is 40% or 90%. Maturities arrive whether projected revenue has materialized or not. Refinancing markets can tighten precisely when borrowers need them most.

The AI infrastructure buildout also includes obligations that do not always appear as conventional corporate debt. BIS research on AI infrastructure financing describes special-purpose vehicles and joint ventures that own data-center assets, raise debt privately and sign long-term leases or capacity commitments with hyperscalers. Economically, these structures can resemble borrowing even when much of the debt sits outside the operating company’s balance sheet. BIS, “Financing the AI infrastructure boom”

That makes the timing of returns almost as important as the ultimate return.

The CFO test for AI capital

Boards evaluating major AI commitments should probably insist on something closer to a capital-investment framework than a technology-strategy presentation.

1. What cash flow is actually incremental?

Cost avoidance, employee productivity, revenue growth and strategic capability are all legitimate benefits. They are not interchangeable. Forecasts should distinguish hard-dollar savings from capacity creation and speculative future revenue.

2. What happens if adoption is slower?

AI business cases frequently assume rapid utilization. The more revealing model may be the downside case: what happens to return on invested capital if utilization takes two additional years to reach plan?

3. How much of the commitment is reversible?

A company purchasing computing capacity month to month has a very different risk profile from one making a multiyear infrastructure commitment. The distinction between variable and fixed cost can matter as much as the headline price.

4. How is the investment financed?

Cash-financed investment can destroy shareholder value. Debt-financed investment can create a liquidity problem. Off-balance-sheet commitments can create economic leverage without looking like conventional borrowing. Those risks should not be modeled as though they are identical.

5. What assumption would make management stop?

Every major capital program should contain an exit condition. If utilization, revenue, productivity or unit economics fail to reach defined thresholds, management should know in advance when additional investment stops.

Private credit is not necessarily the weak link

There is an important counterargument. Private-credit managers can monitor borrowers closely, negotiate stronger protections than dispersed public bondholders and restructure financing relatively quickly when circumstances change. Secured lending and comparatively low leverage in parts of the market may also limit spillovers.

So the growth of private lending may represent useful financial innovation rather than financial excess. And the largest technology companies remain highly profitable, with balance sheets capable of supporting substantial investment.

The risk is therefore not simply “too much debt.” It is the interaction of optimistic forecasts, concentrated exposure, leverage, fixed commitments and uncertain timing of returns.

Watch the financing, not just the technology

Technology shares weakened Monday as investors reacted to renewed debate over the pace of AI development. The daily market move will eventually become noise. The financing trend probably will not. Reuters, September 14, 2026

If the AI buildout continues at anything close to its anticipated scale, more capital will need to come from somewhere. As companies move beyond internally generated cash, creditors increasingly become participants in the AI thesis.

That creates a different feedback loop. When equity investors become less enthusiastic, prices fall. When creditors become less enthusiastic, capital itself becomes more expensive or unavailable.

AI does not have to fail for the financing to become problematic. The technology could transform the economy exactly as its strongest advocates expect while individual companies, lenders and investors still misprice the cost, timing or structure of the investment cycle.

Railroads transformed transportation. Electrification transformed industry. The internet transformed nearly everything. None of those truths prevented capital from being misallocated along the way.

The lesson for the AI era is similar:

Being right about the technology is not enough. You also have to be right about the price, the timing—and increasingly, the balance sheet.

Primary sources and further reading

Our reporting and correction standards are available on the Editorial Standards page.