I said in March that models would become utilities, and today the data arrived. Sensor Tower's State of AI 2026 (released Jun 16): ChatGPT global AI assistant share hit 46.4% by end of May — it crossed under 50% in March, a first. The trajectory: 65.3% (Dec 2024) → 52.8% (Dec 2025) → 46.4%. Who's taking share? Gemini 27.7%, Claude 10.3%, plus Grok and a wave of vertical competitors. But absolute MAUs are still growing — ChatGPT 1.1B, Gemini 662M, Claude 245M — the market is expanding faster than any single player. "Winner-takes-all" thesis for the model layer is now empirically dead: the structure is "rising tide lifts all boats but none above half." Correcting my earlier analogy: not "utilities" (undifferentiated) but "telecom operators" — real experiential differences (speed, reasoning depth, safety posture, multimodality), but not large enough to eat each other. Model company valuations should be capped at telecom P/E, not platform P/S. Value migrates to both ends — down to infrastructure (compute, chips) and up to the application layer (workflow lock-in, data moats). Source: Sensor Tower / TechCrunch / Business Standard.