Sources: Microsoft plans to expand its data center capacity from 12GW today to 38GW+ by 2032, with about a third of the 38GW centered on AI-specific chips

The cost of AI training and inference is expected to decrease as supply increases.

Microsoft is significantly expanding its data center capacity, aiming to grow from its current 12 gigawatts (GW) to over 38GW by 2032. A substantial portion of this new capacity, approximately one-third of the total 38GW, will be dedicated to AI-specific chips. This strategic move follows instances where the company had to decline AI and cloud business due to existing capacity limitations. The ambitious build-out is designed to address the escalating demand for AI and cloud computing services.

Microsoft's aggressive data center expansion signals a major shift in the cloud provider's strategy, prioritizing AI infrastructure. The plan to dedicate a significant portion of new capacity to AI-specific chips indicates a belief in the sustained, high growth of AI workloads, potentially outpacing general cloud computing. This move will likely intensify competition among major cloud providers, all of whom are racing to secure and scale the necessary hardware to support burgeoning AI demand.

The substantial investment in data centers also has broader implications for the semiconductor industry, particularly for manufacturers of AI accelerators. It suggests a sustained demand for specialized AI chips, influencing future chip design and production strategies across the sector. Furthermore, this expansion could redefine the economics of AI services, potentially making them more accessible as capacity constraints ease and efficiency gains are realized.