What Surprised Us Most In 2025 - Y Combinator Podcast Recap

Podcast: Y Combinator Podcast

Published: 2025-12-22

Duration: 30 minutes

Summary

In 2025, the AI industry stabilized across model, application, and infrastructure layers. Companies are now focusing on domain-specific models and innovative infrastructure solutions like space-based data centers.

What Happened

The year 2025 marked a significant shift in the AI landscape, where the chaotic growth of previous years gave way to a more structured and buildable environment. This stabilization is evident as companies across the model, application, and infrastructure layers are positioned for profitability. Anthropic has emerged as a preferred choice among Y Combinator founders, with its adoption rate soaring from 20-25% to over 52%, surpassing OpenAI.

Gemini also saw increased use, with its user base among YC founders growing to 23% by the winter 2026 cycle. Known for its superior reasoning capabilities, some users have replaced traditional Google searches with Gemini. This trend indicates a shift in how people interact with AI technologies, preferring models that enhance cognitive tasks.

Startups are increasingly developing orchestration layers, allowing them to switch AI models based on their performance for specific tasks. This flexibility highlights a movement towards personalized AI solutions that can adapt to various needs. The market is compared to the telecom bubble, suggesting that while there might be overinvestment, it can lead to future opportunities, particularly in the application layer.

Infrastructure innovations, such as space-based data centers, are becoming more viable due to terrestrial constraints. Companies like Zephyr Fusion aim to generate gigawatts of energy in space, highlighting the push for sustainable and scalable solutions to meet the growing demand for computing power.

The interest in niche AI model companies is increasing, with more startups focusing on developing smaller, domain-specific models. These models, often fine-tuned on open-source platforms with reinforcement learning, demonstrate superior performance in targeted areas compared to their larger counterparts.

Despite skepticism about a potential AI bubble, the stabilization of the AI economy has created a clear playbook for building AI-native companies. The deployment phase of AI is expected to lead to an explosion of applications, offering abundant opportunities for startups to innovate and capture new markets.

A notable trend is companies achieving significant revenue milestones with lean teams. For instance, a company reached $100 million in annual recurring revenue with only 50 employees, showcasing the potential for high efficiency and profitability in the current AI landscape.

Key Insights