The costs of building AI

AI infrastructure spending is creating meaningful credit risks, but the current buildout is not necessarily a bubble because hyperscalers remain unusually strong and cash-generative.

• Hyperscaler debt issuance has exceeded $200bn in 2026, reviving memories of earlier debt-funded overbuilds that ended in significant investor losses.
• The risks are partly offset by strong demand, accelerating revenue, EBITDA and operating cash flow, as well as funding from equity rather than debt alone.
• Hyperscaler bonds and CDS markets are key indicators: spreads have widened across parts of the AI ecosystem during 2025 and 2026, although from low levels and without broader credit stress.
• Circular financing, including hyperscalers financing their own customers and suppliers, may obscure organic demand, conceal leverage off balance sheet and create single points of failure.
• China’s state-supported data-center and large-language-model development, along with possible US political opposition to data centers, could intensify competition and weaken incumbent sponsors.

Takeaway: Investors should monitor bond issuance, spreads, CDS pricing, free cash flow and vendor financing closely, while recognizing that the scale and cash generation of the leading hyperscalers make a repeat of the dot-com collapse less likely.