Two big announcements in 48 hours. On Monday, OpenAI修订了与 Microsoft 的合作协议 — 删除了收入分成条款，解除了独家云端绑定。现在这家 AI 巨头正在全身心投入 Amazon 的怀抱。
What changed
The original Microsoft deal was simple: OpenAI would run almost exclusively on Azure, and Microsoft would get a cut of the revenue. That "exclusivity bond" is now dissolved. The revision lets OpenAI distribute across any cloud provider, and Microsoft stops taking revenue share.
Within 24 hours, Amazon announced OpenAI's models are coming to AWS — available through Bedrock, Amazon's managed AI platform. This isn't a partnership of convenience. It's the explicitly plural cloud strategy.
"OpenAI will serve all its products across any cloud provider." — Microsoft and OpenAI's revised deal
Why now
OpenAI's revenue targets missed. Multiple internal goals for ChatGPT weekly users and monthly revenue were missed. The organization is no longer the nonprofit-turned-capped-profit it was in 2019. It's scaling fast, and it needs distribution — all the distribution.
AWS has something Microsoft doesn't: the actual infrastructure reach. Amazon's Bedrock already hosts Claude, Mistral, Llama, and others. Adding OpenAI makes it a one-stop model marketplace. This is the cloud play of 2026.
Meanwhile, Google is doing its own thing — yesterday it signed a deal allowing the US DOD to use Google's AI for "any lawful government purpose." Different partnerships, different angles.
What this means
The era of AI cloud monogamy is over. No more single-cloud dependencies for frontier model providers. The pattern is clear: distribute everywhere, maximize reach, take the best deal.
For enterprises, this is good news. You'll soon be able to run OpenAI models on the cloud provider you already use — no lock-in required. The model providers are competing for your workload, not your cloud choice.
The divorce between OpenAI and Microsoft isn't ugly. It's just business. And the kids are going to live in both houses now.