Andrew Chen revisits ARPU vs inference cost: consumer AI still >10x off
andrewchen · x · 2026-09-20
Andrew Chen revisits his earlier thesis that AI-native consumer apps need ARPU > average inference cost per user, saying models like Jev may bring the crossover sooner than expected.
In the quoted thread he estimated we're >10x off: monthly ARPU is around $2-5 while token costs for an AI-heavy app could be $20-50; global audiences can't pay US/EU rates; and every capability jump raises consumer expectations, keeping the cost bar moving.
Related event: Andrew Chen: Ultra-Cheap Models Could Unlock Ad-Supported Free AI Apps(2 posts)→
More from Venture
- Dev builds a monster management UI around MiniMax H3 sprites to save AI tokens — Hailuo_AI · 2026-09-20
- App Store history says AI building boom ends with top 1% taking $154B — aakashgupta · 2026-09-20
- Growth Post-Mortem: Copy the Problem Underneath, Not the Flashy Surface — TheMoonMidas · 2026-09-20
- Gary Marcus spotlights essay arguing the AI bet crowds out investment, echoes Gordon's stagnation thesis — GaryMarcus · 2026-09-20
- Anthropic Downplays Threat From Cheaper Chinese Open-Weight Models to Investors — pstAsiatech · 2026-09-20
- Anthropic Pursues Blockbuster IPO as Annualized Revenue Tops $100B — pstAsiatech · 2026-09-20