Loading...
LG Doucet and Martin examine the capital-intensive economics behind NeoClouds, asking why markets punish hyperscalers for rising capex while giving some AI infrastructure providers more latitude. They distinguish rapidly growing revenue and backlog from profit and free cash flow, focusing on the debt, equity, rent, electricity, and repayment costs required to finance GPU capacity.
A model for one megawatt of AI compute shows why residual GPU value is the critical assumption. If older chips can still command meaningful rental rates after initial contracts expire, returns could be attractive; if performance gains and expanding chip supply drive those rates much lower, the investment case weakens sharply.
The discussion also offers a framework for managing volatile markets: separate broad market selling from company-specific deterioration, build valuation scenarios, and understand what must go right before owning a stock. Martin remains invested in CoreWeave and Nebius while developing models to determine whether their potential upside adequately compensates for their capital needs and downside risks.
VIDEOS
