Microsoft expands use of its homegrown Maia AI chips in data centers, aiming for a multi-gigawatt rollout

Microsoft is increasing the share of its own silicon in the hardware stack that powers its artificial intelligence systems, according to a report shared via Yahoo Finance. The company says it is moving toward installing the newest version of its Maia AI chips across data centers, with the chips positioned to support AI software delivered through its cloud offerings.

Maia is Microsoft’s custom chip design for AI workloads. By pairing specialized hardware with its software, Microsoft can target lower latency and potentially improve efficiency compared with relying entirely on third-party accelerators. In the Yahoo Finance report, Microsoft frames the push as a way to run more of its AI compute on “mostly homegrown hardware.”

The report also says Microsoft expects to stock “multiple gigawatts” of Maia chips. Gigawatts is a measure of power capacity, not chips in a count, and the phrasing implies Microsoft is planning for a large-scale buildout of data center electricity and cooling infrastructure to sustain AI demand over time.

Microsoft did not provide additional detail in the cited post about the specific performance targets of the latest Maia generation, the mix of Maia chips versus other accelerators in its fleet, or the exact timing of when the multi-gigawatt chip supply would fully translate into deployed AI capacity.

The company, meanwhile, has increasingly emphasized that AI infrastructure is a strategic bottleneck across the industry, tying chip availability, power procurement, and data center capacity together. Custom silicon, where feasible, is one lever for reducing dependency on external supply while aligning hardware with the company’s own model training and inference software pipelines.

A broader backdrop is that hyperscale cloud providers are competing on both software and compute efficiency as AI workloads expand. Microsoft’s emphasis on homegrown hardware suggests it wants more control over the cost and availability of the compute used for tasks such as model inference and other AI services that run in the cloud.

Even with the direction described in the Yahoo Finance report, key specifics remain undisclosed in the material cited. It is not clear what portion of Microsoft’s AI training and inference will be handled by Maia chips as opposed to other accelerators, nor is there disclosed information on whether Microsoft’s “multiple gigawatts” expectation reflects committed supply contracts, internal production, or a combination of planning scenarios.

For investors and customers, the next practical question is how Microsoft’s Maia rollout maps to measurable outcomes, such as data center power utilization, AI service capacity growth, and any changes to pricing or performance for AI features running on Azure. Monitoring Microsoft’s subsequent disclosures, including any technical announcements and infrastructure updates, is likely to offer more clarity on the pace and impact of the buildout.