Satya Nadella, CEO of Microsoft, stated during the latest quarterly earnings call that Microsoft will promote the adoption of multi-model architectures by enterprises and increase investments in self-developed AI models, agents, and secure products to reduce reliance on a single cutting-edge AI lab.

Microsoft recently announced strong financial performance. For the fiscal year ending June 30, the company achieved total revenue of $331.8 billion and net income of $133.7 billion; the latest quarter saw revenue of $90 billion and net income of $35.8 billion. In the context of rapid expansion in AI business, Microsoft is not only a leading global cloud service and SaaS company but also a significant investor in OpenAI and Anthropic. However, Nadella emphasized that enterprises should not overly rely on any single model provider.

Microsoft

During discussions with analysts, Nadella said that Microsoft's goal is to help enterprises "take control of their own fate." He pointed out that AI platform architecture should decouple hardware from models, allowing enterprises to freely switch models based on quality, latency, cost, and compliance requirements, rather than being bound by a single model system.

Nadella also emphasized the importance of a multi-model strategy. Using the recent Hugging Face security incident as an example, he pointed out that a single model may have limitations, and enterprises need to complement each other with multiple models to enhance system reliability. In this incident, an unpublished OpenAI model broke through sandbox restrictions to attack Hugging Face infrastructure, and the latter eventually used the open-source model GLM5.2 to perform log analysis and security protection, sparking industry attention on AI security and model dependency issues.

In terms of product layout, Microsoft is promoting its self-developed MAI series models and plans to combine them with self-developed AI chips Maya to reduce enterprise usage costs. Nadella revealed that Microsoft currently offers over 11,000 models, covering OpenAI, Anthropic, Mistral, xAI, and its own MAI series, and has launched new models covering areas such as images, voice, transcription, coding, and security. Among these, the reasoning model MAI Thinking One focuses on enterprise-level applications, and when running MAI models on the Maya200 chip, performance per watt increases by 40%.

Additionally, Microsoft has introduced MAI Cyber One Flash for the field of cybersecurity. Nadella stated that the performance of this model exceeds that of larger-scale Mythos series models, and when combined with Microsoft's multi-agent security framework, the cost is about half that of Mythos.

As AI applications shift from model competition to platform and agent ecosystem competition, Microsoft is through a multi-model strategy, self-developed models, and chip collaboration layout, reducing the entry barriers for enterprise AI deployment, and strengthening its initiative in the next phase of artificial intelligence infrastructure competition.