Microsoft’s AI Capacity Is Still Difficult to Verify

Microsoft’s AI Capacity Is Still Difficult to Verify

On 17 August, the Guardian questioned Microsoft’s disclosures about its AI infrastructure using internal documents and publicly stated power capacity. The documents indicated about 2.2 million AI chips, while calculations based on the company’s announced power capacity suggested that there should be more. Microsoft said the estimate relied on incorrect assumptions but did not identify which figures were wrong. In June, the company announced that the first Fairwater facility in Wisconsin was fully operational, and Wisconsin Public Radio also reported that equipment had been brought online.

The important issue is not one chip count. Announced capacity, powered capacity and computing actually in use are different measures. The AI race is increasingly described in gigawatts and capital expenditure, but without verifiable utilisation rates and consistent ways of counting chips, outsiders cannot tell whether scale has become usable capability. The evidence currently establishes a disclosure gap; it does not prove that Microsoft lacks chips or that its construction program has failed.

Sources:
https://www.theguardian.com/technology/2026/aug/17/are-microsofts-ai-plans-being-held-back-by-a-shortage-of-chips
https://news.microsoft.com/source/2026/06/23/microsoft-completes-construction-on-first-datacenter-facility-in-mount-pleasant-wisconsin/
https://www.wpr.org/news/microsoft-finishes-construction-mount-pleasant-data-center


Discover more from Geoffrey Chen

Subscribe to get the latest posts sent to your email.