The figures released by Y Combinator for its Summer 2026 batch will surprise no one who has observed the market with clear eyes, yet they mark a threshold that demands acknowledgement. Of the two hundred and fourteen companies admitted to the cohort, one hundred and twenty-eight — precisely fifty-nine point eight per cent — were built, from their architectural foundations, upon large language model or multimodal AI infrastructure.

The proportion is not merely a function of investor enthusiasm. It reflects a structural reality: the cost of incorporating foundational model capabilities into a software product has fallen by more than ninety per cent over thirty-six months. The product categories that were economically impractical eighteen months ago — personalised legal guidance, sophisticated medical documentation, contextual financial analysis — are now viable at Series Seed valuations.

Three thematic clusters dominate the AI-native contingent. The largest comprises vertical workflow automation tools targeting professional services: legal document generation, medical coding, financial reconciliation. The second encompasses developer-facing infrastructure — observability, evaluation, and deployment tooling for production AI systems. The third, and perhaps most revealing of longer-term trends, consists of companies applying AI to physical-world operations: construction project management, agricultural yield optimisation, supply chain exception handling.