Clouded Judgment: Per-Token Pricing Is Becoming the Default SaaS Business Model for AI

Clouded Judgment (Jamin Ball)·medium signal

Jamin Ball's March 26 analysis argues AI companies must shift from seat-based SaaS to token-based pricing as inference becomes the dominant cost driver, with H100 rentals at $2-4/GPU-hour generating token-based revenue of $600-800/hour. Commodity output tokens run $0.15-0.20 per million while mid-tier models command $8-15/M, and Ball argues credit-based pricing abstractions will follow AWS's evolution from hourly VMs to granular, value-aligned pricing. The thesis: 'pricing model is your business model' — and companies that get token economics wrong will be structurally uncompetitive.