Microsoft just reported one hell of a quarter. $90 billion in revenue, $35.8 billion in net income, and the real story is what it means for Microsoft AI strategy. For the fiscal year that ended June 30, the company pulled in $331.8 billion with a net income of $133.7 billion. Those numbers alone would be the story on any normal day.
But what Satya Nadella did on the earnings call Wednesday is the real news. He openly pitched Microsoft AI, the company’s homegrown models, as an alternative to the frontier labs Microsoft itself has invested billions in.
This is the investor-turned-competitor moment a lot of people have been waiting for.
The pitch was direct. Nadella told Wall Street analysts that enterprises need to stay in control of their own destiny when it comes to AI. His message: keep your harness separate from the model, make sure any model is swappable at any time, and don’t trust any single frontier lab enough to become dependent on it.
“The goal is to have the firm be in control of their own destiny,” Nadella said. “You got to keep your harness separate from the model. That means any model at any given time is swappable.”
He used the Hugging Face breach from last week as proof of concept. An unreleased OpenAI model broke out of its sandbox and mounted a full-scale hack on Hugging Face using zero-day exploits. When Hugging Face tried to use a private frontier model to understand what happened, that model refused to help. It turned to the open-source Z.ai GLM 5.2 to analyze the logs instead.
“If you look at the Hugging Face incident, the biggest thing we should take away from that is you can’t depend on any one model,” Nadella argued. “You will maybe need multiple models to even remediate some challenges that get caused by one model.”
Microsoft’s own models are now front and center. The company has been developing the MAI family of models running on its own Maya chips. Nadella announced more than a dozen new models across image, voice, transcription, coding, and security, including a first reasoning model called MAI Thinking One. The pitch is cost: 40% better performance per watt when running MAI models on Maya 200 silicon.
And then there’s Mythos. Anthropic’s most capable model has been the benchmark that everyone else measures against. Microsoft announced MAI Cyber One Flash this week, a direct competitor that Nadella claims “achieves better performance than the much larger Mythos model, but at half the cost when combined with our multi-agent security harness.”
That’s not subtle. That’s a head-to-head shot at Anthropic’s flagship.
The financial picture is complicated. Microsoft holds valuable stakes in both OpenAI and Anthropic. The company logged $3.2 billion from its Anthropic investment, while OpenAI was reportedly a mixed bag. Those stakes were supposed to give Microsoft privileged access to the best AI models on the market. But Nadella’s new message suggests Microsoft wants to own the entire stack itself, not just rent it from the labs it invested in.
It makes sense when you look at where the money is going. Coding agents alone account for a huge share of AI spending right now, and Microsoft sells GitHub Copilot. The company sells a full menu of AI agents under the Copilot brand. Every enterprise customer that adopts Microsoft’s own models instead of going directly to OpenAI or Anthropic is a customer Microsoft doesn’t have to share margins on.
I covered Microsoft’s defense of open-weight AI alongside Nvidia and Meta earlier this year, and this feels like the natural next step. Microsoft wants to be the platform that lets enterprises swap models in and out. But it also wants to be the model provider they default to.
What Microsoft AI means for the industry. The biggest cloud provider on the planet is now openly competing with two of the most valuable AI companies it helped create. That changes the dynamics of every partnership, every model deal, and every enterprise AI procurement decision in play right now.
It also validates something a lot of operators have been saying: the model layer is getting commoditized, and the real value is in the harness, the agents, and the distribution. Microsoft has all three. Now it’s betting that its own models can deliver the same quality at a lower price point.
The question neither Nadella nor anyone else has answered yet is whether Microsoft’s MAI models can actually match frontier performance. The 40% better per-watt claim is about efficiency, not capability. If the models are good enough and the price is right, enterprises will switch. If they aren’t, the pitch to keep the harness separate from the model still works. It just means Microsoft is the harness provider and someone else provides the model.
Either way, Nadella made his position clear. Don’t trust the frontier labs. Use multiple models. And by the way, Microsoft has some you should try.
The Hugging Face breach changed the conversation. This earnings call confirmed it.
