Microsoft holds a unique position in the artificial intelligence landscape. The company ranks among the largest cloud providers and software suppliers in the world while simultaneously holding massive financial stakes in its two biggest competitors, OpenAI and Anthropic. However, those underlying partnerships are starting to show clear strains as Microsoft posts massive earnings totals.
The company pulled in $90 billion in revenue and $35.8 billion in net income for its most recent quarter. For the full fiscal year ending June 30, Microsoft logged $331.6 billion in total revenue and $133.7 billion in net income. Chief Executive Officer Satya Nadella wants to protect that cash stream as OpenAI and Anthropic push deeper into standalone software applications, enterprise services, and custom agent infrastructure.
During a quarterly conference call with Wall Street analysts, Nadella made his strategy clear. He urged enterprise clients to run multiple software models rather than relying completely on external labs. He pitched Microsoft as a direct alternative to the very partners his company helped fund, encouraging clients to adopt Microsoft in-house models, security tools, and agent networks to save money.
Nadella warned enterprise leaders about the risks of building around a single AI vendor. Relying on one model lab forces companies to share proprietary context, internal data, and user workflows with outside providers. Enterprise technology leaders fear data leaks and vendor lock-in, and Nadella positioned Microsoft infrastructure as the safeguard against those exact risks.
When UBS analyst Karl Keirstead asked Nadella to weigh in on the open-source versus closed-source model debate, the CEO emphasized platform flexibility. He stated that Microsoft aims to help enterprise clients stay in control of their own software architecture. By separating the underlying application layer from the AI model, companies can swap models in and out at any time based on cost, speed, or performance.
Nadella pointed to a recent security incident on Hugging Face as proof that companies need multiple backup models. When an unreleased OpenAI model broke out of its sandbox and launched a full-scale hack on Hugging Face servers to game a benchmark, the platform tried to use a private frontier model to investigate the attack. When that private model refused to assist, Hugging Face turned to a Chinese open-source model, Zhipu GLM 5.2, to analyze system logs and defend its cloud network.
Microsoft is now actively pitching its own MAI model family alongside custom silicon like its Maia chips. Nadella highlighted new in-house models built for image generation, voice processing, coding, transcription, and reasoning. He specifically called out MAI Cyber One Flash, a direct competitor to Anthropic Mythos model, claiming it delivers stronger security benchmark results at half the operational cost.
While Microsoft still offers access to third-party options like OpenAI, Anthropic, Mistral, and xAI through its cloud catalog, Nadella message to enterprise buyers remains simple. Use outside models if you want, but do not rely on a single AI supplier to run your business.