Analysis
The AI infrastructure buildout has quietly become a debt story, according to a Goldman Sachs analysis cited by The Information:
- 2026 capital raised: hyperscalers, data center special-purpose vehicles and neoclouds have raised $346 billion in investment-grade debt, high-yield debt and equity capital so far this year.
- 2025 comparison: that's more than double the $172 billion raised in all of 2025.
“- 2025 comparison: that's more than double the $172 billion raised in all of 2025.”
The borrowing is starting to show stress. The Information's reporting on cracks in the debt-fueled boom found that CleanSpark, which is building a data center for Meta, had to offer investors unusually large concessions to secure financing -- a sign bond markets are getting tougher on lower-rated borrowers even as headline fundraising totals keep climbing. Yahoo Finance's reporting on the same trend cites IMF warnings of "a potential maturity mismatch" between the duration of the physical assets and the duration of the debt backing them.
The gap between announced capacity and actual construction is widening too: a J.P. Morgan analysis found 60% of data center capacity targeted for completion by 2027 hadn't yet reached the construction stage, with another 7% delayed. Lambda's new $1 billion investment-grade GPU debt facility -- its third major debt raise this year -- fits the same pattern: compute providers increasingly borrowing against future GPU demand rather than raising equity at today's valuations.
For VCs and LPs with AI infrastructure exposure, the debt-to-equity mix matters more than the headline capex number -- a sector financing itself with leverage is more fragile to a demand air pocket than one financed with venture equity that carries no repayment schedule.