Microsoft is competing more openly with OpenAI and Anthropy than ever before

Microsoft is in a unique position where AI is outpacing the technology industry. It is one of the world’s largest cloud providers and software-as-a-service companies, while also owning valuable stakes in the two largest AI labs, OpenAI and Anthropic.

These incentives began to collide as Microsoft posted impressive financial results. The company just reported a very profitable quarter with revenue of $90 billion and net income of $35.8 billion. For the fiscal year ending June 30, Microsoft reported revenue of $331.8 billion with net income of $133.7 billion for the year.

And CEO Satya Nadella isn’t about to let the trajectory of Anthropic and OpenAI — which are expanding into applications and proxy infrastructure that could eventually allow them to have relationships with customers — hobble that kind of money.

Nadella has been urging organizations to use multiple models and stop relying on frontier AI labs for the harness/agent application layer.

He said doing so was dangerous, because it would require companies to share many of their internal secrets with model makers of questionable trustworthiness. He knows his clients. Enterprise IT fears data leakage and is tied to the vendor.

Now he told Wall Street analysts publicly during the company’s quarterly conference call on Wednesday that this is an opportunity for Microsoft to sell customers its own on-premises models, along with proxies, AI security and more, with the promise of lower costs.

In other words, it’s promoting Microsoft as an alternative to many of the cutting-edge services that OpenAI and Anthropic are developing for their own growth.

When UBS analyst Karl Keirstead asked Nadella specifically for his opinion on the open versus closed source debate roiling the AI ​​industry, and how Microsoft would benefit from it, Nadella came out swinging.

“The goal is to put the company in control of its own destiny,” the CEO said of companies. “We’re very clear about the kind of design architecture for the platform, which is that you have to keep your belt separate from the model…that means that any model at any given time is interchangeable.”

Microsoft, of course, sells a list of tools (aka AI agents) as well, under the Copilot name, including the GitHub Copilot coding agent. Programming agents are where a lot of AI dollars are spent today.

He used last week’s high-profile incident as evidence of his warnings.

“If you even look at the Hugging Face incident, the most important thing to take away from that is that you can’t rely on any one model,” Nadella said. “You may need multiple models to even address some of the challenges caused by one model. Like that’s the way to think about it, right? Which is that you can’t be subject to the rejection of one model.”

The incident involved an unreleased model from OpenAI that came out of its sandbox and successfully performed a large-scale hack on Hugging Face, all in pursuit of the best standard. In an attempt to understand what happened, Hugging Face initially tried to use the (unnamed) private Border Form, who refused to help him. So it turned to the open source Chinese model Z.ai GLM 5.2 to analyze the logs and defend its infrastructure. This incident has shocked the industry so much that Sam Altman now says that perhaps AI development should slow down a bit.

Nadella also explained that Microsoft happily sells its home-grown models, the MAI family, on its own home-grown AI chips, Maya, offering them as cheaper alternatives.

“Every customer wants the right model for every task based on quality, latency, cost and compliance. We offer the broadest catalog of models in the cloud with more than 11,000 models, including leads from OpenAI, Anthropic, Mistral and xAI, as well as our own MAI family,” he said.

He added: “We’re also accelerating our own model development. We’ve announced more than a dozen new models across image, audio, transcription, encryption and security, including our first inference model, the MAI reasoning model, all with cost-effective inference at the core of enterprise use cases. We’re co-designing these models with our own silicon, and are seeing 40% better performance per watt when running MAI models on Maya 200.”

As for the mythos? Nadella pointed to Microsoft’s new Mythos competitor announced earlier this week, the MAI Cyber ​​One Flash. “It delivers better performance than the much larger Mythos model, but at half the cost when combined with our multi-agent harness,” he said.

Certainly, Microsoft’s CEO says companies should use the leading models offered by OpenAI and Anthropic in their mix. But his bigger message is: Don’t trust them enough to depend on them.