Reasoning
Major AI providers (OpenAI's backers, Google, Microsoft, Meta, Anthropic) are experiencing intense competitive pressure in LLM inference, with pricing per token declining 85-95% since 2023 (e.g., Claude 3 Opus to Haiku pricing compression, GPT-4 to GPT-4o mini). Given 2026 earnings calls occur in early 2027, providers will have ~18 months to face margin pressures from: (1) massive inference scale at commoditized prices, (2) capital intensity of training/inference infrastructure, and (3) competitive race-to-bottom dynamics. Historical precedent: semiconductor and cloud infrastructure companies regularly discuss margin compression when facing commoditization (AMD/Intel in CPU markets, AWS pricing pressure discussions). The explicit acknowledgment specifically in earnings calls is likely given investor demands for transparency on unit economics in AI—CFOs have already begun discussing "margin pressure" obliquely in 2024-2025 calls.Key uncertainty
Whether competitive pricing stabilizes at a profitable level before 2026 earnings (H1 2027), or if providers achieve sufficient inference efficiency gains through optimization to maintain margins despite volume growth, which could make pricing compression less explicit as a margin driver versus efficiency gains.