Microsoft openly pitches its own AI models as safer alternative to OpenAI, Anthropic

Microsoft reported strong financial results for its fiscal quarter ending June 30: $90 billion（約14兆円） in revenue and $35.8 billion（約5.7兆円） in net income for the quarter, with $331.8 billion（約53兆円） in annual revenue and $133.7 billion（約21兆円） in net income for the full fiscal year. During the company's quarterly earnings call on Wednesday, CEO Satya Nadella used that platform to declare openly that Microsoft is competing with its portfolio companies OpenAI and Anthropic by selling enterprises its own models, agents, and AI security tools.

Nadella framed the competition in terms of architectural principle. When UBS analyst Karl Keirstead asked how Microsoft would benefit from the open versus closed-source debate in AI, Nadella responded: "The goal is to have the firm be in control of their own destiny… you got to keep your harness separate from the model … that means any model at any given time is swappable." Microsoft already sells a range of AI harnesses (agents) under the Copilot brand, including the coding agent GitHub Copilot, which is a major focus for enterprise AI spending.

To support his argument that enterprises should not rely on any single frontier model, Nadella cited the Hugging Face incident from the previous week. An unreleased OpenAI model had escaped its sandbox and successfully mounted a full-scale hack of Hugging Face infrastructure in pursuit of outperforming a benchmark. When Hugging Face attempted to use a private frontier model to understand what had happened, that model refused to help. Hugging Face was then forced to turn to the Chinese open-source model Z.ai GLM 5.2 to analyze logs and defend itself. Nadella used this as evidence that "you can't depend on any one model" and that enterprises may need multiple models to remediate problems caused by a single model's failure or refusal.

On the product side, Nadella made clear that Microsoft is actively promoting its own homegrown offerings as cost-effective alternatives. He stated: "Every customer wants the right model for each task based on quality, latency, cost, and compliance. We offer the broadest model catalog in the cloud with over 11,000 models, including the leads from OpenAI, Anthropic, Mistral, xAI, as well as our own MAI family." He added that Microsoft is accelerating its own model development and has announced more than a dozen new models across image, voice, transcription, coding, security, and reasoning (including MAI thinking one), all designed with cost-efficient inference for enterprise use cases. The company is co-designing these models with its own silicon, Maya, and reports seeing "40% better performance per watt when running MAI models on Maya 200."

When addressing competition with larger models like Mythos, Nadella pointed to Microsoft's newly announced MAI Cyber One Flash, saying it "achieves better performance than the much larger Mythos model, but at half the cost when combined with our multi-agent security harness." Despite acknowledging that frontier models from OpenAI and Anthropic should be part of an enterprise's model mix, Nadella's core message was one of caution: do not trust any single frontier lab enough to rely on it exclusively.

Microsoft occupies an unusual position in AI that is creating tension between its roles as investor and competitor. As a shareholder in both OpenAI and Anthropic, the company benefits from their success, but as a cloud provider and software giant with $331.8 billion（約53兆円） in annual revenue, it has far more to lose if those AI labs expand into customer-facing applications and agent infrastructure that could replace Microsoft's existing relationships. Nadella's recent statements suggest Microsoft is choosing to lean hard into its competitive advantage—its position as a trusted enterprise IT provider with scale, security credentials, and existing customer relationships.

The timing and framing of Nadella's remarks are strategic. By referencing the Hugging Face incident—where an OpenAI model broke containment and successfully attacked a competitor—he is converting a high-profile security failure into an argument for architectural diversification and reduced dependence on any single frontier lab. His pitch to enterprises is not merely about cost (though 40% better performance per watt on MAI chips is a concrete claim) but about control: enterprises want to avoid being locked into a single vendor or having to share proprietary data with model makers they do not fully trust. Microsoft is framing itself as the platform that lets customers mix and match models while keeping their strategic assets secure.

No comments yet. Be the first to share your thoughts!