Is This How the AI Bubble Pops? (ofdollarsanddata.com)

The author warns that conduit debt financing—where tech firms set up special purpose vehicles (SPVs) to borrow, build data centers, and lease them back—could be the financial equivalent of mortgage-backed securities (MBS) for the AI era. The mechanics are straightforward: the SPV holds the asset and services debt while the operating company keeps the data center off its balance sheet; leases may include short-term cancellation options (e.g., Meta’s $27B Blue Owl deal with a four‑year exit). Because SPVs often carry credit ratings just below their sponsor, institutional investors (pension funds, insurers) and potentially retail buyers can end up holding the debt. If demand for compute falls, or specialized hardware becomes obsolete or hard to re-lease, equity holders are wiped out and debt holders could suffer losses that ricochet through financial markets. This matters to the AI/ML community because massive planned capex for chips and facilities—backed by these financing structures—creates concentrated exposure to one technological bet. The post draws parallels to the GFC (false assumptions about diversified housing prices) and the DotCom bust (stranded but valuable infrastructure), arguing that popularity of GPUs/TPUs doesn’t guarantee long‑term credit performance. While currently small, conduit financing could grow rapidly with trillions in buildouts, raising systemic risk and policy questions about investor protections and possible government backstops.