Debating OpenAI's Training Funding Logic
nptacek · x · 2026-07-19
This post attempts to steelman a core concern within OpenAI: if they cannot sell inference tokens at a high enough markup, they may fail to secure the funding needed for the next massive training round. The author agrees this premise is largely sound, adding that the continuous emergence of powerful open-source models will further erode OpenAI's ability to extract high gross margins from inference services. The controversy lies in the final logical leap: OpenAI assumes that if even they cannot fund their next training round, it means "no one can," because the business model is fundamentally unprofitable. The author and many others find this conclusion overly dogmatic.
More from Companies & People
- UK’s AISI may move into the Cabinet Office as an AI taskforce is planned — ShakeelHashim · 2026-07-21
- Eric Xing argues AI university leaders should be practical operators, not distant administrators — YiMaTweets · 2026-07-21
- OpenAI’s Build Week buildathon drew 40 people for 11 hours with Codex — paw_lean · 2026-07-21
- Etsy’s handmade identity erodes as AI knockoffs and mass-produced goods spread — Wired AI · 2026-07-21
- Two FDEs shipped a FedEx invoicing rebuild in 3.5 months with 122 PRs — alex_verem · 2026-07-21
- YURI says they will speak at the second WAIC 2026 event — hq4ai · 2026-07-21