Idle GPUs burn cash: Neolabs lose money below ~60% spot utilization

FinanceYF5 · x · 2026-10-28

Part 5 of the Neolab economics thread: Neolabs must prove superiority over cheaper frontier models like Opus 5.5, so many end up charging hefty platform and compute fees on top of inference. Idle GPUs burn money every hour; teams typically resell compute, serve inference for open models, or sell spot instances — below 60% spot utilization, they still lose money, before even counting extremely expensive talent costs.

Original post →

More from Venture

Venture channel →