Microsoft reports restricted AI capacity through 2026

Microsoft has acknowledged that its AI computational capacity will remain restricted at least until the end of 2026. This comes amid investigations into discrepancies between the company’s reported AI capabilities and its actual hardware deployment.

Internal documents suggest Microsoft has deployed approximately 2.2 million AI accelerators, a figure that contrasts with estimates derived from the company’s announced electrical capacity. Some calculations based on power availability suggest the infrastructure could theoretically support up to 6.4 million GPUs, creating a significant gap between projected and actual inventory.

While Microsoft contests these electrical-based premises, the company confirmed operational limits during its fiscal third-quarter results conference. This restriction may lead to tighter global resource allocation for Azure customers, though there is currently no public data confirming specific regional shortages or impacts on existing contracts.