The AI Bill Is Too Damn High: Tokenomics 101 — Eugene Cheah, Featherless AI
Featherless AI runs 2T tokens/day at $0.30/million, far below public cloud rates
“they deliver Opus 4.8-class performance. That is, this is a huge number of parallel requests and a large volume of tokens.”
Eugene Cheah of Featherless AI presented a tokenomics workshop showing that large-scale private inference deployments can achieve frontier-model performance at $0.30/million tokens—a fraction of public cloud pricing—using commodity AMD hardware. The talk targets organizations spending $10k+/month on AI and argues that below that threshold, finding compelling use cases matters more than token optimization. This signals a growing gap between enterprise-scale private inference economics and standard API pricing.