A detailed analysis drawing a direct parallel between the 2008 subprime mortgage crisis and the current AI data center financing boom. The piece argues that AI data centers are being funded through Special Purpose Vehicles (SPVs) that structurally resemble CDOs, allowing hyperscalers like Meta and Google to hide hundreds of billions in debt off their balance sheets via Variable Interest Entities. The author contends that real AI compute demand is a fraction of what is being built — estimating over 15x more capacity in planning than actual demand — with the majority of that demand concentrated in two unprofitable companies (OpenAI and Anthropic). NVIDIA is accused of circular self-dealing by investing in and backstopping neoclouds that exist primarily to buy its GPUs. The systemic risk is amplified because pension funds, insurance companies, and retail investors are exposed through private credit. The author warns that when these SPVs fail to generate customer revenue, 80% or more of invested capital could be lost, triggering a financial crisis comparable to 2008.
