- The AI market is bifurcating into expensive frontier models and cheap, 'good enough' models for many agentic tasks, with aggressive pricing from competitors like Grok 4.5 and GLM5.2.
- Winners include semiconductor companies and the hardware supply chain, as value shifts from software to hardware due to supply constraints and increased demand from cheaper models.
- Coding agents (e.g., Cursor) benefit from cheap models, enabling profitable operations and access to valuable user data that can train future models.
- End users and consumers are major winners, gaining high-quality AI intelligence at low prices compared to earlier models like GPT4.
- Frontier AI labs may lose revenue if users switch to cheaper models, but they could respond by restricting access to top models via managed platforms to avoid commoditization and protect against distillation.
- The B2C market remains a wildcard, with potential for LLM-adjacent advertising to monetize large user bases like ChatGPT's 1 billion MAUs, shifting focus back from B2B.
- Overall, margins in pure model inference are heading to zero, with value captured by hardware supply chains and users, while frontier labs must maintain innovation leads to avoid commoditization.