Anthropic and OpenAI file for IPOs at $965 billion and $852 billion valuations, but analyst argues frontier AI labs are chasing the wrong markets for sustainable revenue.

Anthropic confidentially filed for an IPO with a $965 billion valuation seeking to raise $60 billion; OpenAI is preparing its own draft with an $852 billion valuation also targeting $60 billion. SpaceX is pursuing a $1.75 trillion listing.

The article argues frontier AI labs are optimized for the top 15% of the global AI market—enterprises with fast networks and deep compute budgets—where demos are impressive but revenue potential is limited. Open-source alternatives deliver comparable results at lower cost, and corporate buyers report struggling to find ROI.

The author identifies untapped markets: highly digitized economies like the U.S. and Europe need AI for COBOL modernization (43% of core banking systems and 95% of ATM transactions still run on COBOL); 51 'Break Out' economies like India, Brazil, and Indonesia need AI credit scoring for populations with mobile transaction histories but no formal credit access; Sub-Saharan Africa and South Asia can unlock $6.1 billion through AI crop-disease detection across seven African countries for 14 million smallholder farmers.

Bain & Company warns AI will need $2 trillion in annual revenue by 2030 to justify compute spending, representing an $800 billion shortfall. Open-weight models are compressing inference prices an estimated 30% to 50% a year, capping margins any model layer can defend. Oracle disclosed $248 billion in data-center leases running 15 to 19 years against customer contracts often running five years.