As AI overtakes the technology industry, Microsoft is in a unique position. It is one of the world’s largest cloud providers and software-as-a-service (SaaS) companies, and holds valuable stakes in two of the largest AI labs, OpenAI and Anthropic.
As Microsoft announces blockbuster financial results, these incentives are starting to collide. The company reported a very profitable quarter with $90 billion in revenue and $35.8 billion in net income. For the fiscal year that ended June 30, Microsoft reported revenue of $331.8 billion and net profit of $133.7 billion.
And CEO Satya Nadella isn’t about to let that kind of cash derail the trajectory of Anthropic and OpenAI, which are expanding into applications and agent infrastructure that will ultimately enable them to own their customer relationships.
Nadella has been preaching to companies to use multiple models and stop relying on cutting-edge AI labs for their agent harness/app layer.
Doing so is dangerous, he said. Because the company has to share too many internal secrets with model makers whose credibility is questionable. He knows his customers. Enterprise IT fears data breaches and vendor lock-in.
Now he’s publicly told Wall Street analysts on a quarterly conference call on Wednesday that this is an opportunity for Microsoft to sell customers its own development model with agents, AI security, and more, while promising lower costs.
In other words, he’s presenting Microsoft as an alternative to OpenAI and many of the advanced services Anthropic is developing to fuel its own growth.
When UBS analyst Karl Keirstead asked Nadella specifically about the open-source vs. closed-source debate rocking the AI industry and how Microsoft stands to benefit from it, Nadella stepped up.
“The goal is to put the company in control of its own destiny,” the CEO said of the company. “We are very clear about the structural design of the platform, which means the harness must be separate from the model, which means any model can be replaced at any time.”
Of course, Microsoft also sells a menu of harnesses (aka AI agents) under the Copilot name, including GitHub Copilot, a coding agent. The majority of AI spending today is spent on coding agents.
And he used a high-profile incident last week as evidence of his warning.
“The first thing we need to know from the Hugging Face incident is that we can’t rely on just one model,” Nadella said. “You may need multiple models to solve some of the problems caused by one model. That’s the way to think about it, right? You can’t reject one model.”
The incident involved an effort by OpenAI’s undisclosed model to break out of the sandbox and successfully mount a full-scale hack against Hugging Face to achieve a benchmark. Trying to understand what happened, Hugging Face attempted to use a private Frontier model (unnamed), which initially refused to help. So we chose Z.ai GLM 5.2, a Chinese open source model, to analyze logs and protect our infrastructure. This incident shocked the industry so much that even Sam Altman is saying that AI development needs to slow down a bit.
Nadella also made it clear that Microsoft is willing to sell its MAI family, a homegrown model of its own AI chip, Maya, and present it as a cheaper alternative.
“Every customer wants the right model for each task based on quality, latency, cost and compliance. We offer the most extensive model catalog in the cloud, with over 11,000 models, including leads from OpenAI, Anthropic, Mistral, xAI and our own MAI suite,” he said.
“We are also accelerating the development of our own models. We have announced more than a dozen new models across image, voice, transcription, coding, and security, including our first inference model, the MAI Incident Model, all with cost-effective inference at the core of enterprise use cases. We are co-designing these models with our silicon and are seeing 40% better performance per watt when running MAI models in Maya 200.”
What about mythology? Nadella pointed to Microsoft’s new Mythos competitor, MAI Cyber One Flash, announced earlier this week. “It achieves better performance than the much larger Mythos model, but when combined with a multi-agent security device, the cost is halved,” he said.
Of course, the Microsoft CEO says companies should use a mix of OpenAI and the frontier models provided by Anthropic. But his bigger message is: Don’t trust them enough to depend on them.
If you purchase through links in our articles, we may receive a small commission. This does not affect our editorial independence.
