September 15, 2026: Microsoft resets its AI scorecard

Today is September 15, 2026, and Microsoft’s most important update is a reporting reset that makes the company’s AI strategy easier to see, but also harder to compare with the past.

On September 2, Microsoft said that starting in fiscal 2027 it will report under two segments: Agents and Infra, and Devices and Consumer. That is a meaningful change because it redraws the lines around Azure, Microsoft 365, GitHub, Windows, Xbox, and ads. The old view of the company was already under pressure from AI, but now the financial map itself is changing.

The key point is that future growth rates will not line up cleanly with the old buckets. Microsoft is also moving GitHub cloud and other developer cloud services, plus Security Copilot, out of Azure and into Microsoft 365 commercial cloud. It is also adding paid GitHub seats to Microsoft 365 commercial seat growth. In plain language, Microsoft is trying to measure more of its AI and developer business where the usage is actually happening, instead of leaving it spread across older categories.

That matters because the last hard operating baseline came from fiscal fourth quarter 2026. Microsoft said Azure and other cloud services revenue grew 43 percent, Microsoft Cloud revenue reached 59.3 billion dollars, and Microsoft 365 Copilot passed 30 million paid seats. Those numbers are still the reference point, but the new segment structure means investors will need to be careful about apples-to-apples comparisons from here.

The second big theme is monetization control. GitHub said on September 14 that Copilot auto model selection is rolling out across VS Code, Copilot CLI, and the GitHub Copilot app. Usage is charged based on the model selected, and paid subscribers get a 10 percent discount on auto usage. That is important because it turns model choice into a billing and cost-control mechanism, not just a product preference.

GitHub also added VS Code Agents to Copilot usage metrics on September 11, giving enterprises and organizations more telemetry on adoption and engagement. These changes suggest Microsoft is still building the plumbing for AI monetization. Before the company can credibly talk about stronger returns, it needs better measurement of who is using what, how often, and at what cost.

Azure is making the same point from the infrastructure side. Microsoft’s Azure blog said project-level cost attribution is now in preview for models sold by Microsoft Azure, including Azure OpenAI. The message is simple: see the spend, bound the spend, and prove the return. For enterprise buyers, that is becoming central to AI adoption. For Microsoft, it is a way to make AI consumption feel financeable rather than open-ended.

One mechanic helps explain the constraint here. AI datacenters only scale as fast as the company can secure power, cooling, and grid connections. More chips do not help if the buildings cannot be energized and kept within thermal limits. That is why the reported plan for roughly 38 gigawatts of datacenter capacity by 2032, along with about 175 billion dollars of capital expenditures in calendar 2026, keeps power access at the center of Microsoft’s story. Reuters reported those figures, so they should be treated as corroborated reporting rather than company guidance.

Security is another part of the same operating loop. Microsoft Security said Defender attack disruption now contains more than 81,000 compromised user accounts each month and disrupts more than 45,000 adversary-in-the-middle attacks monthly. Those are company claims, not independently audited totals, but they show how Microsoft is positioning security as both a product line and a proof point for AI-era threats.

Windows is also being nudged toward local AI development. On September 4, Microsoft announced Project Zenith, a developer-class Windows setup with 64 gigabytes or more of unified memory and 250 gigabytes per second of memory bandwidth, aimed at running 30 billion parameter plus models locally and unmetered. That is not a consumer Windows refresh. It is a signal that Microsoft wants Windows to matter in local AI workflows, not just in cloud-connected apps.

Gaming remains active, though not newly quantified in the latest evidence. On September 9, Xbox said it expanded its partnership with Kojima Productions to publish PHYSINT and extend the relationship beyond OD, including film and television collaboration. That tells us Microsoft is still willing to spend on prestige content, even without a fresh gaming financial disclosure.

So the read on Microsoft as of September 15 is straightforward. The company is not just shipping features. It is reorganizing how it measures AI, pricing how it charges for AI, and expanding the infrastructure base required to support both. The open question is whether demand, monetization, and power supply stay in balance as the FY27 reporting structure takes effect.

What to watch next: first, the next FY27 earnings materials for any clean read on Azure growth under the new segment structure. Second, any disclosure of GitHub paid seats, Copilot usage, or Microsoft 365 commercial seat growth that lets us judge adoption more directly. Third, any update that reconciles datacenter capacity, capex, and power access with actual deployment timing. Fourth, whether Project Zenith or the new Copilot billing controls show up in broader enterprise rollout data.