Data Center Fragility

Описание: ‪@Benny_Squirrel‬ Episode 57: *Ameesh Agarwal (CIO, University Place Asset Management)* joined Benny and the 🐿️ to dissect the economics of AI data center investing. His firm underwrites the *base infrastructure layer* — shell, power, and location — avoiding GPU exposure due to uncertain 4–7 year depreciation cycles and negative unit economics.

The core argument: **CoreWeave is structurally insolvent**. Buying GPUs on leverage, leasing them to frontier labs, and hoping to recover CapEx before the chips depreciate is a losing trade. Investment banks mask this by applying EBITDA multiples to a non-perpetual cash flow stream — a fundamental valuation error. The parallel to **WeWork is explicit**: buzzword-dressed arbitrage with no real moat.

Not all neoclouds are equal — former Bitcoin miners (Applied Digital, TeraWolf) with owned power rights are far better positioned than CoreWeave. The entire ecosystem depends on capital markets staying permanently open, drawing uncomfortable comparisons to **CDO structures in 2006–2007**, with Magnetar — CoreWeave's largest shareholder — running the same playbook.