For years, the deal was simple: Microsoft pumps billions into OpenAI, and in return gets to put GPT models inside Copilot, Windows and Office. That arrangement made Microsoft the face of enterprise AI without having to build frontier models itself. But in an interview this week, Microsoft CEO Satya Nadella all but announced that the era of depending on someone else's AI is over at Redmond.
01.What Nadella actually said about Microsoft MAI models
Addressing a question on why a company commanding a massive $175 billion AI budget has not yet fielded a frontier model of its own, Nadella was blunt. "We are well on our way building our MAI models. Our goal is to basically hill climb from the bottom, by the way, not distilling anything," he explained. And then came the line that will be quoted for years: "My advice is more like use all but be independent of all."
He described his own test for whether Microsoft is truly independent: pull out a model and see whether you can still retain the eval, the evaluation score you care about. "If I can't, that means you really are dependent on something that may or may not be yours."
02."You're not a supplier to two model companies"
Here's the thing nobody's talking about: this isn't just about pride. Nadella argued that cloud giants cannot simply be passive compute suppliers to a couple of frontier labs. "If you're a hyperscaler, you're not a supplier to two model companies, that's not a business we need more," he said. He even reached back to his early career working on Windows interoperability with Unix — opening up made Microsoft more used, not less, because it deepened enterprise penetration. He's applying the same playbook to AI architecture now.
03.The switch has already been flipped
This is not a plan for 2028. In July, Nadella revealed in a long post on X that Microsoft had already started routing work inside its own products away from OpenAI and Anthropic, towards models it built itself. GitHub Copilot, Excel and Outlook were the first to show what he called promising early results, with Copilot Chat and PowerPoint next in line. If you noticed Copilot feeling a bit different lately, this is why.
We covered it when GPT-6 Astra landed on Azure AI Foundry — Microsoft happily sells everyone's models, including OpenAI's. But selling access and betting your own products on someone else's model are two very different decisions, and Microsoft has clearly separated them.
04.The real driver: every token has a price now
Nadella made a point that every Indian startup founder and engineering manager should tattoo somewhere visible: software now has a real marginal cost for the first time. When a spreadsheet formula or an email draft can burn tokens, the choice of model stops being a technical preference and becomes a business decision. His rule — frontier models for frontier problems, and everything else gets pushed to something cheaper that performs just as well. Microsoft claims its MAI models already beat general-purpose frontier models on many use cases while burning a fraction of the tokens.
This is exactly the cost discipline we keep hammering on about. If a company with a $175 billion budget is counting tokens, a Bengaluru startup with a ₹5 lakh monthly AI bill has no excuse. A solid cost-optimization chapter in our GPT-6 Astra — Zero to Hero Guide walks through caching, model routing and budget control for exactly this reason — the savings show up directly in your margins.
05.Does this hurt OpenAI?
Honestly, less than the headlines suggest — at least immediately. OpenAI's models still lead on the hardest problems, and the Azure partnership still prints money for both sides. But the strategic message is unmistakable: Microsoft refuses to be hostage to any single model vendor, including the one it funded. And OpenAI is now competing with its biggest investor for the same enterprise workloads, while GPT-6 Astra fights for enterprise share against Anthropic's Claude. The AI stack is fragmenting, and buyers — including every Indian enterprise — are the winners, because everyone is suddenly competing on price per useful output.
Two more from the blog that connect to this: Copilot Lawsuit: US Court Sides With OpenAI and Microsoft in DMCA Appeal and OpenAI's GPT-5.6 Sol Left Secret Notes to Future Models.
Key Takeaways
- Microsoft is building MAI models from the ground up — "hill climb from the bottom, not distilling anything"
- Nadella's philosophy: "use all but be independent of all"
- GitHub Copilot, Excel and Outlook already route work to in-house MAI models; Copilot Chat and PowerPoint are next
- Software has a real marginal cost now — model choice is a business decision, not a technical one
- Hyperscalers won't accept being mere suppliers to two model companies
Frequently Asked Questions
What is Microsoft building?
Satya Nadella says Microsoft is building its own MAI models "from the bottom" rather than distilling anyone else's work.
What is Nadella's guiding philosophy?
"Use all but be independent of all" — no more defaulting to a single AI partner.
Which products are affected?
Microsoft has already started routing work in GitHub Copilot, Excel and Outlook away from OpenAI and Anthropic models, with Copilot Chat and PowerPoint next. The driver is economics — software now has a real marginal cost, so every token counts.
What This Means for You
If you're a developer or an IT decision-maker in India, stop building your product around one AI vendor. Nadella — the man who wrote the biggest AI cheques in history — is telling you the winning strategy: keep the option to swap models without your quality collapsing. That means clean abstraction layers, proper evals for your own use cases, and ruthless cost tracking. And if you're early in your AI journey, the GPT-6 Astra — Zero to Hero Guide covers model selection, cost optimization and agent-building from a beginner's level — the same skills Microsoft is betting its entire product line on.