SPVs, Credit, and AI Datacenters (paulkedrosky.com)

Meta is reportedly lining up roughly $29 billion to accelerate AI datacenter buildout — about $26B in debt and $3B in equity — by leveraging private investment groups and special-purpose vehicles (SPVs)/joint ventures rather than putting the debt directly on its balance sheet. This mirrors a wider surge in GPU buys (more in Q1 this year than any full year before 2021) as hyperscalers race to secure Nvidia/Google accelerators. Private credit providers, often funded in part by insurance-company capital, are willing to lend at 200–300 basis points above investment-grade rates in these bespoke deals because the borrower is a mega-cap and the structure appears lower risk on paper. For the AI/ML community this is a double-edged sword: it accelerates access to massive, cheaper compute capacity now, but it also creates material credit and systemic tail-risks. Key technical concerns are "control without consolidation" (SPVs that serve a company but keep debt off its books), thin equity cushions (~10% reported), opaque counterparty exposure via insurer-backed credit, and incentives to overbuild capital-intensive datacenters on speculative workload growth. If demand stalls or margins compress, SPV losses could cascade to insurers and private-credit markets, prompting a sudden pullback in capital and compute availability. In short: the infrastructure boom buys speed today but concentrates leveraged risk in opaque vehicles that could disrupt AI compute supply and costs if the financing model breaks.