The Costs of Building AI

Stephen Fallon, CFA, CAIA

AI infrastructure spending has sparked a debt surge. Could that be a warning sign or just the cost of doing business?

Artificial intelligence's capabilities have been eye-popping – and so has the debt being issued to fund it. Already in 2026, total debt has increased dramatically for hyperscalers, with issuance exceeding $200bn. This surge, along with some creative issuance methods, has brought back memories of prior debt-filled overbuilds, which ended with significant losses for investors.

Given this, we believe investors are right to take a skeptical eye to developments in this space – and to ask the question: could boom turn to bust?

First the positives. For one thing, although current levels of debt issuance are large in absolute terms, they’re not so large relative to the size of the enormous companies involved. Demand for their services is strong and growing and for the most part the leading corporates are seeing rapidly accelerating growth in revenues; earnings before interest, taxes, depreciation, and amortization (EBITDA); and operating cash flow. It’s worth noting, also, that it’s not just debt that’s been issued to fund the buildout: equity has also played a part.

Equally positive is how the scale of the AI buildout has been a strong contributor to economic growth in the US lately. This is likely to beget further growth as productivity-enhancing tools gain in acceptance and use. Corporate earnings have been healthy, which is a good sign of strength going forward. Trading in hyperscaler bonds may have had a little difficulty due to excess supply, but credit risk does not appear to be under stress.

We can point to other positives too. One difference relative to prior episodes of debt binges is the strength of the underlying companies. The hyperscalers remain some of the most innovative and, importantly, cash-generative companies we’ve ever known. The leaders of the dot-com era had shakier fundamentals than are seen among the hyperscalers today. And we’re seeing a similar trajectory for each as regards the growth of debt relative to operating cash flow.

How is the market pricing downside risk?

So much for the positives. What of downside risk?

We believe there are two areas in particular that are worth monitoring here: how the bonds of hyperscalers are trading – including how new issuance is received – and also how credit default swap (CDS) markets are pricing the risk of default.

Historically, the hyperscalers have been investment-grade stalwarts, characterized by tight credit spreads and CDS levels that consistently trade well inside broader investment-grade indices. But perhaps this could change, particularly if free cash flow expectations diminish as a result of current capital expenditures? Already, through parts of 2025 and throughout 2026 we have seen widening spreads within the AI ecosystem albeit from low levels. We believe investors would be well advised to keep a tight eye on developments here.

Circular financing

Another area of potential concern is how the hyperscalers and other AI firms have aggressively financed their own customers and suppliers – behavior that echoes prior crashes.

Even if some such arrangements are more defensible than others, they can still obscure organic demand, hide leverage off balance sheet, and create single points of failure within the ecosystem. Any unforeseen change in the landscape or surprise reduction in demand could cause future credit stress. (We’d also note, however, that though this behavior does merit concern, there can be benefits too, such as lower financing costs and better access to capital.)

It appears that the funding cycle has followed a trajectory: first free cashflow, then unsecured debt and eventually on to vendor financing. Perhaps the resistance to circular financing will prompt companies to return to the credit markets for funding – and, again, this is an area we believe is worth monitoring.

Other areas to watch

Competition from Asia is a key consideration too. China will continue to be a competitor in large language model research for the frontier labs but also in the production of datacenters, likely accelerated through state funding. Political backlash against datacenters in the United States could provide an opportunity for China to enhance its standing in the global artificial intelligence landscape to the detriment of other incumbents.

Other risk factors we’d be watching here are an existential change in the AI landscape (unlikely, we believe, as the market appears supply constrained for several years) and a reduction in the quality of collateral or balance sheet strength of the sponsors. This would likely happen through a significant change in the competitive environment.

Leverage enhances risk but doesn’t always create a bubble

Two further caveats to conclude: First, while the levels of debt required for the AI buildout are extraordinary, they are also well known. Investors are not dealing with a completely unfamiliar landscape and, so far, have tailored their expectations accordingly.

Second, we can point to the guiding hand of the market as a moderating force. Many times in the past we’ve seen the market push back on companies using excessive leverage. Although there are no guarantees, our hope is that this will keep investors from getting ahead of themselves while simultaneously reminding the hyperscalers of the need for a degree of restraint.

READ MORE FROM FEDERATED HERMES

Disclosures

Information provided at FederatedHermes.com is intended only for persons who are eligible to purchase products registered or services available in their jurisdiction.

Children under the age of 18: Users of the Federated Hermes website must be at least 18 years old. If you are under 18, do not use or provide any information on our website or through any of its features. Please contact us if you believe we might have any information from or about a child under 18. 

The value of investments, and income from them, may go down as well as up, and you may not get back the original amount invested.