Nadella's public showdown: Don't just believe in OpenAI and Anthropic, Microsoft wants to be the cheapest. Microsoft is in an unprecedented and delicate position - it is one of the world's largest cloud service providers, and it also holds huge shares in two leading AI laboratories, OpenAI and Anthropic.

Nadella's public showdown: Don't just believe in OpenAI and Anthropic, Microsoft wants to be the cheapest. Microsoft is in an unprecedented and delicate position - it is one of the world's largest cloud service providers, and it also holds huge shares in two leading AI laboratories, OpenAI and Anthropic. But as the two companies continue to expand into application layers and agent infrastructure in an attempt to directly control customer relationships, CEO Satya Nadella clearly does not intend to let this trend threaten Microsoft's cash cow. Microsoft just handed over an explosive financial report: fourth-quarter revenue of $90 billion and net profit of $35.8 billion. For the entire fiscal year 2026, revenue was $331.8 billion and net profit was $133.7 billion. Nadella's strategy is clear - he has recently repeatedly shouted to enterprise customers not to rely solely on models from cutting-edge AI laboratories. Binding the agent framework to the underlying model is dangerous because it means handing over a lot of internal secrets to model vendors with questionable credibility. From the Hugging Face incident to self-developed MAI, Microsoft showed a full set of cards. During Wednesday’s earnings call, Nadella directly showed off to Wall Street analysts: This is Microsoft’s opportunity. He proposed that the core architectural principle is to "separate the agent framework from the model", and any model can be replaced at any time - and Microsoft happens to sell a series of agent frameworks under the Copilot brand, including GitHub Copilot. He even cited last week's Hugging Face intrusion as evidence: After an unreleased model broke through the sandbox and invaded Hugging Face, the latter initially tried to use a private frontier model to analyze the attack. As a result, the model refused to assist, and ultimately had to turn to the Chinese open source model Z.ai GLM 5.2 to complete the defense. Nadella also showed off Microsoft's self-developed arsenal: the MAI series models are equipped with self-developed Maya chips, which increase performance per watt by 40% when running on Maya 200. More than 11,000 models are available on Microsoft Cloud, covering OpenAI, Anthropic, Mistral, xAI and the self-developed MAI family. The newly released MAI Cyber ​​One Flash is claimed to surpass the Anthropic larger Mythos model at half the cost. Nadella's subtext could not be clearer: Cutting-edge models can be used, but don't trust them to the point of dependence - and Microsoft should make it an option that you can change at any time and afford. via AI News (author: AI