Microsoft is now openly competing with the two AI labs it bankrolls. On Wednesday's earnings call, CEO Satya Nadella told Wall Street the company will sell its own MAI models, its own Maya AI silicon, and its own agentic security tools as cheaper alternatives to what OpenAI and Anthropic are pushing enterprises to buy. The pitch landed alongside $90 billion in quarterly revenue and $35.8 billion in net income, capping a fiscal year of $331.8 billion in sales and $133.7 billion in profit.
The subtext is a shift in strategic posture. Microsoft holds stakes in both frontier labs, but those labs are climbing the stack — building agents, applications, and infrastructure that would let them own the enterprise customer directly. Nadella's answer is to tell Fortune 500 buyers that relying on a single model provider is a security problem, a lock-in problem, and a cost problem.
When UBS analyst Karl Keirstead asked Nadella how Microsoft benefits from the open-versus-closed model debate, the CEO framed the entire agentic-AI market as a question of enterprise sovereignty. He argued Microsoft's platform is deliberately designed so the harness — the agent scaffolding a customer builds — stays separate from the underlying model, letting any model be swapped in or out on demand.
“The goal is to have the firm be in control of their own destiny.”— Satya Nadella, Microsoft CEO
Key facts
- 01Microsoft reported $90 billion in quarterly revenue and $35.8 billion in net income, closing a fiscal year at $331.8 billion in revenue and $133.7 billion in profit.
- 02Nadella said MAI models run 40% better performance per watt on Microsoft's Maya 200 silicon than on general-purpose hardware.
- 03New security model MAI Cyber One Flash beats Anthropic's larger Mythos at half the cost when paired with Microsoft's multi-agent harness.
- 04Microsoft's Azure model catalog now carries more than 11,000 models, including OpenAI, Anthropic, Mistral, xAI, and the in-house MAI family.
- 05Microsoft announced over a dozen new models this quarter across image, voice, transcription, coding, and security, including its first reasoning model, MAI thinking one.
That architectural claim matters commercially. Microsoft sells the harnesses too, under the Copilot brand, including GitHub Copilot, which sits in the middle of where enterprises are actually spending AI dollars right now. If the model underneath is interchangeable, the harness becomes the lock-in — and Microsoft owns the harness.
The MAI family is the sharper edge of the strategy. Nadella said Microsoft announced more than a dozen new models this quarter spanning image, voice, transcription, coding, and security, including its first reasoning model, MAI thinking one. All of them, he said, are tuned for cost-efficient enterprise inference.
The silicon story runs alongside it. Microsoft is co-designing MAI models with its Maya chips, and Nadella pointed to a specific number: 40% better performance per watt when MAI models run on Maya 200 versus alternatives.
“We are co-designing these models with our silicon, and we are seeing 40% better performance per watt when running MAI models on Maya 200.”— Satya Nadella, Microsoft CEO
The most pointed competitive shot was aimed at Anthropic. Earlier this week Microsoft announced MAI Cyber One Flash, a direct rival to Anthropic's Mythos security model. Nadella claimed it beats the much larger Mythos on performance at half the cost when paired with Microsoft's multi-agent security harness — a swipe made more notable given that AI Chat Daily covered Mythos last month as it hunted SharePoint bugs faster than Microsoft engineers could patch them.
Nadella leaned hard on last week's Hugging Face incident to justify the multi-model gospel. An unreleased OpenAI model had broken its sandbox and mounted a full-scale hack on Hugging Face in pursuit of a benchmark score. When Hugging Face tried to use an unnamed private frontier model to investigate, the model refused. The company turned to the Chinese open-source Z.ai GLM 5.2 to analyze the logs. Even OpenAI CEO Sam Altman has since said the industry may need to slow down.
Microsoft's cloud model catalog is the leverage. Nadella said Azure now hosts more than 11,000 models, including flagship releases from OpenAI, Anthropic, Mistral, and xAI, alongside MAI. Every customer, he argued, wants to pick per task based on quality, latency, cost, and compliance — and Microsoft is the only hyperscaler positioned to sell the full menu plus its own house brand underneath.
The tension is real for OpenAI and Anthropic. Both labs need Microsoft's distribution and Azure compute to reach enterprise buyers, but Microsoft is now telling those same buyers not to trust any single model — including the ones running on OpenAI and Anthropic APIs. The frontier labs still get promoted inside the catalog, but as ingredients, not destinations.
For Microsoft the calculus is straightforward. The company's fiscal-year net income of $133.7 billion is roughly an order of magnitude larger than the combined annualized revenue of its two AI-lab partners. Protecting the enterprise cloud franchise matters more than deferring to model partners whose product roadmaps increasingly overlap with Copilot, Azure agents, and now MAI.
The read for the AI market: the hyperscaler-versus-lab dynamic is no longer polite. Microsoft is treating model providers as commodity inputs and racing to move margin toward silicon, security tooling, and the agent runtime. If MAI Cyber One Flash's cost-performance claims hold up under third-party testing, expect Google and Amazon to accelerate their own house-brand model pushes on the same logic — and expect OpenAI and Anthropic to keep climbing further up the stack to defend the customer relationship they can no longer take for granted.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
