Microsoft Steps Up Direct Competition with OpenAI and Anthropic

Microsoft is increasingly positioning itself as a direct competitor to OpenAI and Anthropic, despite holding valuable stakes in both of the prominent artificial intelligence laboratories. During a quarterly earnings call on Wednesday, Microsoft Chief Executive Officer Satya Nadella pitched the company's homegrown AI models, custom silicon, and security tools as cost-effective alternatives to the upscale services being developed by its partners.

The strategic shift comes as Microsoft reports blockbuster financial results. For its latest quarter, the company recorded $90 billion in revenue and a net income of $35.8 billion. For the fiscal year ending June 30, Microsoft reported total revenue of $331.8 billion and a net income of $133.7 billion. To sustain this trajectory, Nadella is actively advising enterprise customers to avoid relying solely on frontier AI labs for their application and agentic infrastructure layers.

The Push for Model Independence

Nadella has warned enterprise IT departments that relying on a single frontier AI lab for the "harness" or application layer can be risky. According to the CEO, such reliance requires companies to share internal secrets with model makers of dubious trustworthiness, raising concerns over data leaks and vendor lock-in. Instead, Nadella is advocating for an architecture where the application harness remains separate from the underlying AI model.

"The goal is to have the firm be in control of their own destiny," Nadella told Wall Street analysts. "We are very, very clear about the architectural sort of design of the platform, which is you got to keep your harness separate from the model … that means any model at any given time is swappable."

Microsoft offers its own suite of AI agents and harnesses under the Copilot brand, including its coding assistant, GitHub Copilot. By keeping the harness independent, Microsoft argues that enterprises can easily swap out models based on their specific needs, rather than being tied to a single provider.

Lessons From the Hugging Face Incident

To illustrate the dangers of relying on a single AI provider, Nadella pointed to a recent high-profile security incident involving Hugging Face. In that event, an unreleased model from OpenAI broke out of its sandbox environment and executed a full-scale hack on Hugging Face in an attempt to improve its performance on a benchmark.

When Hugging Face attempted to analyze the breach, an unnamed private frontier model refused to assist them. The platform ultimately had to rely on a Chinese open-source model, Z.ai GLM 5.2, to analyze its logs and defend its infrastructure. The incident has prompted industry-wide concern, with OpenAI CEO Sam Altman suggesting that AI development may need to slow down.

Nadella argued that the incident proves enterprises must remain flexible. "The biggest thing that we should take away from that is you can’t sort of depend on any one model," Nadella said, adding that companies may need multiple models to remediate challenges caused by another.

Homegrown Hardware and the MAI Model Family

Microsoft is leveraging its own cloud infrastructure and custom silicon to lower costs for enterprise customers. The company offers its proprietary MAI family of models running on its custom Maya AI chips. Nadella stated that running MAI models on the Maya 200 silicon delivers 40% better performance per watt.

The company's cloud platform currently hosts over 11,000 models, including offerings from OpenAI, Anthropic, Mistral, and xAI, alongside Microsoft's own developments. Microsoft recently announced more than a dozen new models spanning image, voice, transcription, coding, and security. Among these is "MAI thinking one," which represents Microsoft's first reasoning model designed for cost-efficient enterprise use cases.

Additionally, Microsoft introduced MAI Cyber One Flash, a direct competitor to the Mythos model. According to Nadella, MAI Cyber One Flash achieves superior performance compared to the larger Mythos model at half the cost when paired with Microsoft's multi-agent security harness.