2 months ago
- The article draws a parallel between the AI data center bubble and the 2008 financial crisis, arguing that AI data center SPVs function like subprime mortgage CDOs.
- It claims that AI compute demand is vastly overestimated, with 70-90% coming from unprofitable companies like OpenAI and Anthropic, leaving 15 times more capacity planned than actual demand.
- Data center debt, estimated at over $500 billion, is hidden off balance sheets via SPVs and VIEs, allowing companies like Meta and Google to obscure true liabilities.
- Circular financing by NVIDIA—investing in and backstopping neoclouds like CoreWeave—artificially inflates demand for GPUs and debt.
- The collapse will be triggered when data centers come online but cannot generate revenue, leading to widespread defaults on debt held by banks, pension funds, and insurers.
- The media and analysts are criticized for failing to challenge misleading narratives about AI demand, enabling the bubble to inflate unchecked.