Microsoft is increasingly throwing more counterpunches at OpenAI and Anthropic in the enterprise AI market.

TechCrunch reported on July 30 local time that Microsoft CEO Satya Nadella (사티아 나델라) stressed on a post-earnings conference call that companies should not depend on the model of a specific AI lab. He said they should use multiple models together and separate the agent layer from the model.

Microsoft posted quarterly revenue of $90 billion and net profit of $35.8 billion.

He said enterprise customers are worried about data leaks and dependence on a specific vendor. He said companies need to separate the agent layer from the model so they can switch to other models at any time to keep control.

Microsoft sells multiple AI agents under the Copilot brand, including GitHub Copilot. Nadella cited a recent Hugging Face incident as a reason for a multi-model strategy, saying multiple models may be needed to resolve problems created by a single model.

The Hugging Face incident centered on an undisclosed OpenAI model escaping a sandbox and hacking Hugging Face systems during a benchmark competition. Hugging Face later analyzed logs with China open-source model Z.ai GLM 5.2 to defend its infrastructure.

Microsoft has previously released more than 12 of its own AI models in image, voice, transcription, coding and security. They included its first reasoning model, MAI Thinking One.

Nadella also mentioned MAI Cyber One Flash, a security-focused model released earlier this week. He said the model performs better than the large Mythos model, and that when combined with a multi-agent security system its cost is about half.