0
Applied AI·October 14, 2026·1 min read

CoreWeave, Neocloud Leverage a Concern for Gil Luria

Share

If AI infra providers are borrowing at ~9% and only earning ~5% returns, the leverage model breaks long before demand does. For heavy GPU consumers, this argues for shorter-term contracts and optionality—assume pricing and counterparties could re-rate quickly as capital markets tighten.

Applied AI

Data centers use as little as 60% of power consumed for actual AI work, Nvidia claims — failures and restarts 'consume' a whopping 10% of every watt used

If only ~60% of your power budget reaches useful AI compute and 10% is lost to failures and restarts, optimization at the facility and orchestration layer is now a direct margin lever. Infra leaders should be instrumenting failure-induced waste and pushing vendors like Vertiv and Nvidia for concrete efficiency roadmaps, not just more megawatts.