AI’s Borrowing Boom Is Rewriting the Quality-Investing Test

Investors accustomed to judging technology companies by growth and margins now have another number to watch: the bill for building the business.

In a [September 22 report](https://www.reuters.com/legal/transactional/corporate-bond-buyers-get-picky-with-flood-ai-debt-2026-09-22/), Reuters described bond buyers becoming more selective about AI-linked debt as the volume and unpredictability of issuance increased. That is not, by itself, evidence of deteriorating creditworthiness. But it raises a question equity investors should take seriously: when a business needs much more capital to keep growing, how much of that growth will ultimately belong to shareholders?

Oracle’s latest results offer a useful case study. The opportunity is substantial. So is the need to separate commercial momentum from investment returns.

A bigger business can be a different business

In its [September 10 earnings release](https://investor.oracle.com/investor-news/news-details/2026/Oracle-Announces-Q1-Results-Driven-by-Triple-Digit-Growth-in-Cloud-Infrastructure-Revenues/default.aspx), Oracle reported fiscal first-quarter revenue of $19.3 billion, up 30%. Cloud infrastructure revenue rose 121% to $7.4 billion, while remaining performance obligations reached $664 billion. The company also reported roughly $23 billion of operating cash flow and negative free cash flow of about $5 billion as infrastructure investment continued.

Those figures describe rapid expansion. They do not settle the question of investment quality.

A software business can sell another subscription with relatively little additional physical capital. A computing service must also supply the machines, electricity and facilities behind each workload. Both can build durable competitive advantages. Their economics, however, must be examined differently.

The analytical mistake is to carry an old valuation framework into a changing business without revisiting its assumptions. Strong customer relationships may survive the transition while the cash required to serve those customers rises sharply. Revenue growth can accelerate at precisely the moment shareholders need to become more demanding about returns.

Backlog is a promise with a delivery cost

Contracted revenue gives investors more information than an optimistic forecast. But a backlog is neither cash in the bank nor profit waiting to be recognised. Its value depends on when customers pay, the costs of delivery and the risks embedded in the contract.

An investor should therefore read a major computing contract like an infrastructure commitment. Who pays for equipment? Who absorbs higher power costs? Can the customer change the volume or timing? What happens if a newer chip makes the installed fleet less competitive?

These are questions to investigate, not claims about undisclosed terms in Oracle’s agreements. They illustrate why two contracts with the same headline value can create very different outcomes for shareholders.

Prepayments can reduce financing pressure. They also create obligations to deliver future services. Customer-funded equipment may improve a provider’s capital efficiency, but investors still need to understand the associated pricing and operating commitments. The financing structure matters almost as much as the order announcement.

The same discipline applies to forecasts of demand. A customer wanting more computing power is encouraging. A customer able and obligated to pay enough for that capacity over its economic life is more valuable.

The return on the next dollar

For quality investors, the key measure is the return generated by new capital, rather than the historical profitability of assets already in place.

Consider a hypothetical project requiring $10 billion of investment. If it eventually produces $1.5 billion of sustainable annual after-tax operating profit, its return on that initial capital is 15%. At $700 million, the return is 7%. Those outcomes imply very different economics even if both projects produce impressive revenue growth. Neither figure is an estimate for Oracle.

Timing complicates the calculation. Capital is often committed before revenue arrives, so a temporarily lower reported return need not mean management has made a bad investment. Equally, describing expenditure as growth investment does not establish that it will earn an adequate return.

Depreciation deserves particular attention. Spreading equipment costs over a longer accounting life lifts near-term earnings relative to a shorter life, all else equal. It does not extend the hardware’s economic usefulness. Investors should test cash flows under different replacement cycles instead of treating an accounting assumption as an engineering guarantee.

Quality must be earned again

My view is that the right response to the AI investment cycle is a tougher underwriting standard, rather than blanket enthusiasm or blanket avoidance.

A capital-intensive expansion can strengthen a franchise if scarce capacity, customer integration and reliable execution support attractive returns. It can also turn a highly profitable incumbent into the financier of someone else’s competitive race.

The distinction will emerge through cash generation, capacity utilisation and returns across successive rounds of investment. A single quarter cannot resolve it.

Bond investors’ selectivity is a useful prompt for shareholders to revisit their own assumptions. The eventual winners need to do more than sell computing power. They must retain enough of the economics to reward the capital that made it possible.