Microsoft reported fiscal 2026 third-quarter results on April 29 for the period ended March 31, confirming what industry watchers have suspected: Copilot is no longer a line-item feature but a consumption-based AI utility woven into the company's biggest platforms. Revenue hit $82.9 billion, up 18% year-over-year, with CEO Satya Nadella telling analysts that \"metered AI usage is the new growth engine.\" The numbers reveal a company successfully transitioning from software licenses to intelligent cloud consumption, and Copilot sits at the center of that pivot.

Q3 Financials: Cloud and AI Drive Growth

The headline figures are staggering, but the composition tells the real story. Azure revenue grew 33%, with AI services contributing 12 percentage points of that expansion—up from 8 points just two quarters ago. Microsoft 365 commercial revenue rose 16%, propelled by Copilot subscriptions and consumption add-ons. Windows OEM revenue grew a modest 4%, yet Windows commercial products and cloud services jumped 22%, thanks largely to Copilot integration in Windows 11 Enterprise.

CFO Amy Hood noted during the call that AI-related revenue has now crossed $23 billion on an annualized run rate, double the rate reported in the previous fiscal year. More importantly, more than 60% of that AI revenue now comes from metered or consumption-based agreements rather than flat per-user licenses. That shift changes everything about how Microsoft sells AI.

The Metered AI Layer Explained

Metering means customers pay for what they use—prompts processed, tokens generated, agent actions triggered—rather than buying a fixed number of seats. It's the model that made Azure an $80-billion-plus juggernaut, and now it's being applied to productivity and operating system layers. The concept is simple: Copilot functions as an intelligent middleware that intercepts user intent across applications and routes it to the appropriate AI model, charging in micro-transactions along the way.

Three product areas show this transformation most clearly.

Copilot in Microsoft 365: Pay Per Prompt

Copilot in Word, Excel, PowerPoint, Teams, and Outlook originally launched as a $30-per-user monthly add-on. That fixed-fee model still exists, but enterprise data from the Q3 report shows that 70% of large customers now use a hybrid plan: a reduced base subscription plus metered consumption. A law firm might pay a flat fee for basic Copilot features across 5,000 employees, then incur additional charges every time a paralegal uses Copilot to draft a 50-page contract analysis or generate a regulatory compliance report.

Microsoft doesn't disclose exact per-prompt pricing, but partner sources indicate rates start at $0.03 per standard query and scale up to $2.00 for complex multi-model tasks that invoke GPT-6, legal-domain fine-tunes, and real-time data connectors. This consumption model opens Copilot to frontline workers without the sticker shock of a blanket $30 license, while giving Microsoft a direct lever to capture value from heavy users.

Azure AI: Pay Per Token

Azure OpenAI Service and the broader Azure AI portfolio have been metered from day one. What's new is the scale. Q3 saw Azure token consumption climb 140% sequentially, as enterprises shifted from experimentation to production. Manufacturing firms now run Copilot-powered digital twins that simulate supply chains in real time, paying by the inference call. Retailers use Azure AI Studio to build custom shopping assistants, paying per generated response.

Nadella highlighted a Fortune 500 healthcare company that moved 80% of its clinical note summarization to a metered Copilot model inside Azure, cutting costs by 35% versus the previous per-seat clinical AI tools. The same pattern repeats across industries: metered AI lowers the entry barrier and then grows with the customer.

Windows Copilot: The Desktop Meter

Windows 11 version 24H2 made Copilot a system-level component, but the Q3 report reveals the behind-the-scenes business model. Windows Copilot now meters premium AI actions separately from the OS license. Fetching a file across OneDrive with a natural-language prompt? Free. Asking Copilot to rewrite your resume using that file and then schedule a LinkedIn post? That might cost $0.05. Deeper agentic workflows—like instructing Copilot to monitor a folder for new contracts, extract key terms into an Excel sheet, and email alerts—run on a consumption basis tied to the user's Microsoft account.

This approach lets Microsoft keep Windows licenses relatively stable in price while building a new AI revenue stream atop the 1.5 billion active devices. The Q3 numbers show Windows AI revenue hitting $2.8 billion, mostly from metered enterprise desktops. Microsoft also introduced a Copilot Basic tier for free users, limited to 50 prompts per month, with additional prompts sold in blocks.

Enterprise Adoption and the Consumption Model

The metered shift explains why enterprises are buying in faster than analysts projected. Procurement teams that once balked at multiplying a $30 Copilot add-on by 50,000 seats now see a more manageable equation: deploy widely, pay modestly for routine use, and let the heavy users drive the bulk of costs. Early data from Q3 suggests that the 90th percentile of users accounts for 60% of Copilot prompt volume, matching the classic Pareto distribution seen in cloud compute.

Microsoft's own sales strategy has adapted. The company now embeds Copilot consumption quotas into Azure Enterprise Agreements, much like Azure credits. A typical EA might include $100,000 in annual Copilot usage across Microsoft 365 and Windows, with overage billed at a small discount. This packaging makes Copilot feel like infrastructure, not a luxury add-on.

Partners are building practices around this model. System integrators now audit clients' Microsoft 365 telemetry to project Copilot consumption costs, similar to how they optimize cloud spending. FinOps teams are extending their tools to track Copilot \"prompt spend\" by department, leading to a new discipline some call \"AI-ops.\"

What It Means for Windows Users

For everyday Windows users, the metered layer is both empowering and a potential minefield of micro-charges. The free Copilot tier handles simple questions, but doing real work—reformatting a document, analyzing a spreadsheet, generating a presentation—will often trigger a paid prompt. Microsoft says it surfaces clear pricing before executing chargeable actions, but user forums are already filling with complaints about unexpected costs, especially in shared family accounts where one person's heavy usage drains a common budget.

On the plus side, Windows is becoming more capable without an upfront subscription. Students who only occasionally need AI help can get by on the free tier, while professionals can expense their prompts. Microsoft is also experimenting with a \"Copilot Pass,\" a $9.99 monthly bundle that includes 500 prompts across Windows and Office, aimed at freelancers and consumers. The Q3 earning's slide deck showed early adoption at 12 million subscribers, a promising start for a product launched quietly in February.

The Competitive Landscape

Metering AI also positions Microsoft to compete differently with Google, Amazon, and Apple. Google has kept its Duet AI mostly as a per-user add-on for Workspace, though reports suggest a consumption model is coming. Amazon's Q has followed a similar seat-based path. Microsoft, by metering Copilot across Office, Azure, and Windows, can offer a unified AI experience where a developer writing code in Visual Studio, querying a database in Azure, and creating a slide deck in PowerPoint all draw from the same consumption account. That cross-silo frictionlessness is difficult to replicate.

Apple, meanwhile, has avoided metering in its Apple Intelligence features, bundling them into device purchases. Microsoft's counter is that true enterprise AI requires cloud compute that can't be commoditized into a one-time hardware sale. The Q3 results suggest the market agrees: Microsoft's AI revenue now dwarfs estimates for Google and Amazon combined.

Looking Ahead

Microsoft's metered AI strategy will get a stress test in the coming months. The company faces questions about cost transparency for consumers, potential regulatory scrutiny over AI pricing practices, and the challenge of maintaining model quality as usage scales. But with $82.9 billion in quarterly revenue and AI growth accelerating, the shift is paying off spectacularly.

Nadella ended the earnings call with a telling phrase: \"We are building the world's first AI utility.\" The Q3 numbers show that utility is already powering a significant chunk of the global economy—and sending a bill for every watt.