The core lie behind AI economic fraud is that “compute capacity is scarce”

With interest costs and aggressive 6 year depreciation schedule of GPUs (datacenters used to depreciate over 3 years), Coreweave lost 24c for every $ in revenue it has. If compute were actually scarce, they could charge enough to actually break even.

Meanwhile, they have a $36B capex budget for the year to reduce scarcity even more, with only 250mw more of capacity expected to be complete this year. They have an 11% interest rate on their debt. If utility partners have delays in connecting their future datacenters, they face massive depreciation/interest costs, and cannot sue for damages related to those costs.