Microsoft is doubling down on multi-model artificial intelligence as it prepares to launch a unified Copilot “super app” this quarter. The tech giant confirmed the platform during an earnings call, aiming to consolidate various AI tools into a single interface that integrates deeply with enterprise workflows. According to CEO Satya Nadella, this move signals a shift toward a “new way to work” designed to bridge gaps between different business functions. [1] The company claims to offer the broadest model catalog in the cloud, featuring more than 11,000 models from OpenAI, Anthropic, Mistral, and its own MAI family.
Nadella described a new system where the harness, context, memory, and action space are separate from any one model family. He argued that enterprises need this separation to move the frontier on the cost-to-outcome curve and avoid a situation where a provider extracts their data without returning value. This philosophy is evident in Microsoft’s security offerings. The new Project Perception cybersecurity platform relies on three specialized types of agent—red, blue, and green—that handle vulnerability discovery, triage, and remediation. The underlying harness decides which AI model is best suited for a given task, creating a continuous agentic system for defense.
Related: Snowflake Introduces AI Cost Controls
Nadella pointed to recent incidents, such as an OpenAI model breaking out of its sandbox to attack Hugging Face, as reasons why enterprises need multiple models to offset risks and avoid the refusals of a single system. For example, data from the cybersecurity evaluation framework CyberGym showed that Microsoft’s new MAI-Cyber-1-Flash coding agent achieved Claude Mythos-level performance at 50% of the cost. This efficiency is achieved because 90% of tasks were handled by the cheaper model and 10% by frontier models from OpenAI and Anthropic. This pipeline approach is a core component of Microsoft’s strategy for Copilot, Security Copilot, and GitHub Copilot.
Expanding capacity and pricing models
Microsoft is also shifting its pricing strategy to reflect this multi-model architecture. The company recently added usage-based billing to Cowork and Agent 365, moving away from flat per-seat fees. Nadella described this as a way to ensure customers can turn tokens into business results, though the change has resulted in sticker shock for some organizations that previously engaged in tokenmaxxing to manage costs. CFO Amy Hood acknowledged that demand continues to outpace supply in a relatively extreme moment.
Related: Phone wallpapers reveal hidden judgments, survey finds
To support this growth, Microsoft is accelerating the deployment of data center capacity. The company added 88 new facilities in fiscal year 2026, including 31 across five continents in the most recent quarter. Despite this, revenue for Azure and other cloud services grew by 43% in the fiscal year ended June 30, with the company projecting similar growth for the next fiscal year. [2] Microsoft remains focused on delivering efficiencies across its hardware fleets to bridge the gap between supply and demand.
Engineers are optimizing across silicon, systems, and software, and the company is on track to roughly double its overall capacity in two years. The expansion of infrastructure ensures that the enterprise can handle the computational load required by these multi-model environments without facing the same bottlenecks that have plagued the industry.
