Microsoft Is Building a Copilot Super App With Multi-Model AI Support

Microsoft is officially doubling down on its AI ambitions with the announcement of a new Copilot super app. During a recent earnings call, CEO Satya Nadella confirmed the company is building a unified platform set to launch this quarter. This new hub will consolidate various Copilot tools, including chat, Cowork, long-running Autopilot agents, and the always-on Microsoft Scout, which is powered by OpenClaw. The move signals a direct push to compete with OpenAI’s ChatGPT Work and Anthropic’s Claude Cowork, aiming to capture more of the enterprise workflow.

What’s Inside the Copilot Super App

The super app is designed to be deeply integrated into Microsoft’s broader ecosystem. It will be wired into governance platforms like Agent 365, IT Ops, SecOps, FinOps, and various business processes. According to Nadella, CRM and ERP systems will function as skills and plug-ins that feed into the core work environment. He described the initiative as “the coming together of a new way to work,” allowing enterprises to take enterprise-wide workflows and wire them directly into the platform.

Microsoft also highlighted the current traction of its Copilot ecosystem. Nadella noted that everyday Copilot “usage intensity” has reached levels comparable to Outlook or Teams. Furthermore, paid seats have now surpassed 30 million, providing a massive user base for the new platform to leverage. This integration strategy aims to keep users within the Microsoft ecosystem while offering the flexibility to move between different AI models based on the task at hand. For IT administrators, this means a centralized control plane for managing AI agents across the organization, reducing the friction of switching between disparate tools.

Why Microsoft is Betting on Multi-Model AI

A core pillar of this strategy is the acknowledgment that enterprises do not want to be locked into a single AI provider. Customers are increasingly demanding the ability to switch between open, closed, and frontier models depending on the specific requirements of a job. This trend is reflected in Microsoft’s own data: the company has tracked a fivefold increase in the number of customers building with models from multiple providers since the beginning of the year.

To support this, Microsoft claims to offer the broadest model catalog in the cloud, boasting more than 11,000 models from partners like OpenAI, Anthropic, Mistral, and its own MAI family. Nadella explained that the company is building a “new model system” where the harness, context, memory, and action space are separate from any one model family. This architecture allows enterprises to balance the advantages of frontier models with lower-cost options and open weights with closed weights.

For instance, data from the cybersecurity evaluation framework CyberGym showed that Microsoft’s new MAI-Cyber-1-Flash coding agent achieved performance levels comparable to Claude Mythos at 50% of the cost. This efficiency was achieved because 90% of tasks were completed by Cyber-1-Flash, while the remaining 10% were handled by frontier models from OpenAI, Anthropic, and others. This pipeline approach ensures that organizations can use the right model for the right task without overspending.

The multi-model approach is also evident in Microsoft’s Project Perception cybersecurity offering. This platform features red, blue, and green agents, with an underlying harness that decides which AI model is best suited for a given task. Red team agents discover vulnerabilities, blue team agents triage issues, and green team agents propose remediation plans. Nadella emphasized that in cybersecurity, having a multi-model approach is critical for creating a continuously operating agentic system for defense. He also pointed to recent industry incidents, such as an OpenAI model going rogue in a sandbox, to illustrate why enterprises should not be subject to the refusals of a single model.

Usage-Based Pricing and Infrastructure Push

Alongside the product announcements, Microsoft is shifting its financial model to accommodate the growing demand for AI. The company is moving from a strict per-seat model to a per-seat-plus-consumption pricing structure. Usage-based billing has already been added to Cowork and Agent 365, with plans to extend this trend across other products. While this shift has caused some sticker shock and “tokenmaxxing” at various companies, Nadella framed it as a necessary step to ensure every customer can turn tokens into business results.

The infrastructure backing these services is also expanding rapidly. Microsoft added 88 data centers in fiscal year 2026, including 31 across five continents in the past quarter alone. The company claims it is bringing capacity online “faster than ever,” reducing dock-to-live times for new GPUs in its largest regions by nearly 50% over the fiscal year. However, CFO Amy Hood acknowledged that demand currently exceeds available supply in a “relatively extreme moment.”

Reflecting this growth, revenue for Azure and other cloud services grew by 43% in Microsoft’s fiscal year ended June 30. The company expects similar revenue growth of 45% in fiscal year ’27. Hood noted that Microsoft is focused on delivering efficiencies, including in its CPU and GPU fleets, and is on track to roughly double its overall capacity in two years. By optimizing across silicon, systems, and software, Microsoft aims to meet the surging demand for AI capabilities while maintaining cost efficiency for its enterprise customers.

What This Means for You

For enterprise IT leaders and developers, Microsoft’s push toward a Copilot super app and multi-model architecture signals a significant shift in how AI will be consumed. The move away from single-provider lock-in means organizations can now optimize costs by routing simpler tasks to cheaper models while reserving expensive frontier models for complex problem-solving. This is particularly relevant for cybersecurity and development workflows, where the right model can drastically reduce operational expenses.

The transition to usage-based pricing will require closer monitoring of AI consumption. While it offers flexibility, it also introduces variable costs that can scale quickly. Organizations should evaluate their current AI workloads to understand potential spend changes. Additionally, the rapid expansion of data center capacity suggests that latency and availability issues may decrease, making AI tools more reliable for global teams. As the platform launches this quarter, users can expect a more unified interface for managing these diverse AI capabilities.

Source: Computerworld

Over to you: Will the shift to usage-based pricing change how your organization adopts Copilot?