Y Combinator President Garry Tan States There Are Only Two Winning Positions for Foundation Models: Absolutely Best or Best Among the Cheapest
Y Combinator President Garry Tan stated that there are only two winning positions for foundation models: absolutely the best, or the best among the cheapest.
He pointed out that the middle ground for capital will lead to death. Some responses suggested that speed, total success cost, or existing contracts of companies constitute additional dimensions, but the core view focuses on the extremes of cutting-edge performance and cost efficiency.
Tan previously predicted that OpenAI and Anthropic would launch plans priced at $1,000 and $10,000 per month, reserving the absolutely best cutting-edge models for metered access. He also emphasized that improvements in larger models will enhance the performance of smaller models, making intelligence more accessible.
YC observations show that in some batches, 95% of company codebases are generated by AI, with small teams achieving rapid revenue growth through cutting-edge models. Tan advocates for founders to establish AI-native organizations, encoding processes as skill documents, leveraging work connections rather than just weights.
He supports U.S. open-source labs obtaining cutting-edge capabilities through legal distillation to maintain competition and avoid monopoly by a single company. Some responses pointed out that integrating proprietary data into weights has independent value.
Capital is flowing towards the extremes of absolute cutting-edge and lowest-cost reasoning, driven by model performance and price curves. Cutting-edge labs and low-cost providers benefit, while mid-tier models and undifferentiated packaging face pressure.
Source: Public Information
ABAB AI Insight
Garry Tan has long promoted the transformation of YC batches towards AI-native models, observing small teams achieving nine-figure revenues with cutting-edge models, and advocating for encoding processes as reusable skills while supporting legal distillation to maintain open-source competition.
In terms of capital flow, funds are concentrating from mid-tier models towards the extremes of absolute best and lowest cost, motivated by performance gaps and price declines that eliminate differentiation in the middle, reallocating resources from generic packaging to metered cutting-edge and efficient reasoning.
Similar cases include the early cloud computing shift from mid-tier hosting to ultra-scale and low-cost edge differentiation, as well as chips moving from general-purpose to high-performance and low-power extremes. Current foundation models are in the stage of differentiation from experimentation to commercial scale, with cutting-edge and low-cost open-source paths running in parallel.
Essentially, this represents a transfer of pricing power: model competition is shifting from equal capabilities to the Pareto frontier of performance and cost, with the mechanism being that declining marginal costs and converging capabilities cause middle products to lose their premium, forcing capital and users to concentrate towards the extremes.
ABAB News · Cognitive Laws
No space in the middle, capital moves to the extremes
Best or cheapest, others struggle to survive
When performance converges, cost determines the outcome