Three years after Microsoft committed roughly $13 billion to OpenAI - establishing itself as OpenAI's exclusive cloud partner and gaining rights to resell OpenAI models through Azure - that arrangement is changing. According to Bloomberg, Microsoft plans to stop sharing revenue with OpenAI.
The precise mechanics aren't fully public. What's clear is that Microsoft is moving away from the financial terms that defined the partnership's early years. Microsoft has spent the last two years building alternatives: its own Phi model series (small, efficient models built for high-volume, low-cost deployment) plus access to Meta's Llama, Mistral, and Cohere models through Azure. The dependency that justified the original deal no longer exists the way it did.
For OpenAI, this removes a meaningful revenue channel. ChatGPT subscriptions and API access fees are OpenAI's primary income, and the Microsoft reseller arrangement represented significant distribution. OpenAI reported $3.4 billion in annualized revenue as of early 2025, but the company spends heavily on compute costs and is still working toward profitability. Losing guaranteed Microsoft revenue adds pressure.
The larger pattern: the hyperscaler-to-AI-lab funding model is unwinding. Microsoft, Google, and Amazon each made early bets on AI labs when they needed frontier model access. Now they're building their own model lineups. The terms that made sense in 2019 don't fit 2026.