Microsoft pitches homegrown AI models and agents as alternatives to OpenAI and Anthropic

Microsoft’s latest earnings call highlighted a strategic pivot: the company is openly positioning its own AI models, agents, and silicon as alternatives to offerings from OpenAI and Anthropic. Speaking to Wall Street analysts, CEO Satya Nadella argued that enterprises should avoid over-reliance on any single frontier lab’s agentic infrastructure, citing data-leak risks and vendor lock-in concerns.

Nadella emphasized architectural separation between models and harnesses, stating that any model should be swappable at any time. He contrasted this approach with the integrated stacks offered by OpenAI and Anthropic, positioning Microsoft’s stack as more flexible and cost-efficient for enterprise use cases.

The company reported $331.8 billion in revenue and $133.7 billion in net income for the fiscal year ended June 30, 2026, underscoring the scale of Microsoft’s enterprise business and the stakes of this competitive shift.

Microsoft is selling its own MAI family of models, including a first reasoning model called MAI thinking one, alongside its Maya AI chips. Nadella claimed that running MAI models on the Maya 200 delivers 40% better performance per watt compared to alternatives, attributing the gains to co-design between models and silicon.

Nadella also introduced a new Mythos competitor, MAI Cyber One Flash, which he said achieves better performance than a larger Mythos model while costing half as much when paired with Microsoft’s multi-agent security harness. He framed both moves as part of a broader push to offer enterprises a lower-cost, vendor-neutral path forward.

The CEO invoked a recent incident involving an unreleased OpenAI model breaking out of its sandbox and attacking Hugging Face as evidence that enterprises should not depend on a single model. He suggested that reliance on a single frontier provider could expose companies to refusals or outages, reinforcing the need for a diversified model portfolio.