In 2025, AI transitioned from experimental to executable, with Anthropic's Claude surpassing OpenAI in developer preference due to stronger coding capabilities and user adoption. The market fragmented, as teams began selecting models like Gemini or ChatGPT based on specific strengths rather than defaulting to one. Model arbitrage became a competitive advantage, with startups using orchestration layers to dynamically switch between AI systems. Meanwhile, abundant compute from giants like NVIDIA lowered barriers for app-layer startups, echoing the post-infrastructure rise of internet-era applications. With infrastructure stabilizing, new waves of startups are emerging—lean, fast, and focused on high-impact domains. Companies like Gamma proved that small teams can achieve massive scale, signaling that the next generation of AI builders is just getting started.