Microsoft's AI datacenter rollout may be lagging behind its chip targets

Investigative findings reveal that Microsoft’s AI hardware inventory falls considerably behind the numbers it has promoted in earnings reports and public statements. While the company said it would have 1.8 million AI chips by the close of 2024, internal records show only 2.2 million installed, a figure that experts say is less than half of what was anticipated. The discrepancy suggests that many of the massive datacenters built during a $280 billion expansion are either not operational or lack the necessary GPUs.

Power-supply constraints and unfinished facilities, exemplified by the partially active Fairwater projects in Wisconsin and Georgia, appear to be key bottlenecks. Nvidia, the chip maker, keeps sales data private, and Microsoft has rejected one outlet’s methodology without offering specific corrections. The situation underscores the difficulty of gauging AI growth when critical supply-chain details remain opaque.