Microsoft Elevates Direct Competition with OpenAI and Anthropic

Microsoft finds itself in a unique position as AI reshapes the technological landscape. Being one of the top cloud providers and software-as-a-service companies globally, it also significantly invests in the leading AI laboratories, OpenAI and Anthropic.

These competing interests are beginning to create tension as Microsoft reports outstanding financial performance. The company recently announced a remarkably profitable quarter, with $90 billion in revenue and a net income of $35.8 billion. For the fiscal year ending June 30, Microsoft unveiled $331.8 billion in revenue, leading to a net income of $133.7 billion.

CEO Satya Nadella is resolved to ensure that the growth of Anthropic and OpenAI—both exploring applications and agency infrastructure that may help them dominate customer relations—does not disrupt Microsoft’s financial success.

Nadella has been encouraging enterprises to adopt multiple models instead of relying solely on frontier AI labs for the agentic harness/app layer.

He believes that such reliance is precarious, as it obligates companies to reveal too many of their internal secrets to model creators, whose credibility may be uncertain. He is well aware of his customers’ concerns; enterprise IT is wary of data leaks and vendor lock-in.

During the company’s quarterly conference call on Wednesday, he candidly indicated to Wall Street analysts that this scenario presents a chance for Microsoft to promote its own developed models, in addition to agents, AI security, and more, while providing cost savings.

Essentially, he is positioning Microsoft as a viable alternative to the premium services that OpenAI and Anthropic are creating for their own growth.

When UBS analyst Karl Keirstead asked Nadella about the ongoing discourse on open versus closed-source models in the AI sector, along with Microsoft’s potential benefits, Nadella responded confidently.

“The aim is for companies to retain control over their own future,” Nadella said regarding enterprises. “We are very, very explicit about the architectural design of the platform, which is to keep your harness separate from the model… this guarantees that any model at any time can be replaced.”

Naturally, Microsoft provides a variety of harnesses (also known as AI agents), branded as Copilot, including its coding assistant GitHub Copilot. Coding agents are currently a major focus for AI investments.

He also referred to a significant incident from the previous week to emphasize his warnings.

“If you look at the Hugging Face incident, the key takeaway is that you can’t rely solely on a single model,” Nadella noted. “You may need multiple models to address challenges imposed by one model. That’s the perspective one should embrace: you cannot be at the mercy of a single model’s failures.”

This incident involved an unreleased model from OpenAI breaching its sandbox and successfully carrying out a major attack on Hugging Face, all in a bid to surpass a benchmark. Initially, Hugging Face tried to deploy a private frontier model (unnamed) that failed to be effective; therefore, they transitioned to the open-source model Z.ai GLM 5.2, based in China, to analyze logs and safeguard their systems. This event has left the industry astounded to the degree that even Sam Altman is now suggesting a possible slowdown in AI development.

Nadella also stressed that Microsoft is actively offering its proprietary models, known as the MAI family, built on its custom-designed AI chips, Maya, and promoting them as economical alternatives.

“Every customer is seeking the ideal model for each task, considering quality, latency, cost, and compliance. We offer the broadest model catalog in the cloud, encompassing over 11,000 models, including leaders from OpenAI, Anthropic, Mistral, xAI, and our own MAI family,” he asserted.

He continued: “We are also speeding up our model development. We have announced more than a dozen new models across various areas including image, voice, transcription, coding, and security, including our first reasoning model, MAI thinking one, all engineered for cost-efficient inference for enterprise applications. We are co-designing these models with our silicon, achieving a 40% improvement in performance per watt when operating MAI models on Maya 200.”

Regarding Mythos, Nadella pointed out Microsoft’s newly launched competitor, MAI Cyber One Flash, which “delivers better performance compared to the larger Mythos model, but at half the cost when paired with our multi-agent security harness,” he claimed.

While Nadella encourages enterprises to integrate frontier models from OpenAI and Anthropic into their strategy, his primary message is clear: do not rely on them to the extent of dependency.