Microsoft plans to scale its custom AI chips across a multi-gigawatt compute buildout, aiming to cut the cost of running artificial intelligence workloads, Investing.com reports.

The move puts Microsoft's in-house silicon strategy — anchored by its Maia accelerator line — at the center of one of the largest infrastructure expansions in the company's history. The reported scale is measured not in server counts but in gigawatts of power capacity, the metric that now defines the ceiling on AI data-center growth.

What does the custom-chip push change?

Custom silicon shifts part of Microsoft's accelerator supply away from third-party vendors and toward chips it designs itself. The company already runs its Maia chips in Azure data centers alongside GPUs from external suppliers. Scaling that program multi-gigawatt-wide would extend in-house silicon across a much larger share of Azure's AI fleet.

The economics drive the decision. AI inference and training consume power at a scale where small efficiency gains per chip compound into major savings across gigawatts of capacity. Every watt saved at the chip level frees capacity for additional workloads without new construction.

The reported plan covers:

• Deployment of custom AI chips across multi-gigawatt data-center capacity
• Cost reduction as the primary stated goal
• Expansion of an in-house silicon program already operating inside Azure

Why does scale matter now?

Gigawatt-class buildouts have become the benchmark for hyperscaler AI investment. At that scale, the choice between merchant accelerators and proprietary chips determines billions of dollars in capital and operating cost over the life of each facility.

Custom chips give Microsoft tighter control over performance-per-watt, the metric that governs how many tokens a data center can serve per unit of energy. Reducing cost per workload directly affects the pricing Azure can offer AI customers in a market where compute bills dominate model economics.

Microsoft has not disclosed final deployment figures, chip generation details, or a completion date in the report. The plan's defining parameter is capacity: multiple gigawatts of AI infrastructure to be served increasingly by Microsoft-designed silicon.

What is the takeaway?

Microsoft is betting that designing its own AI chips at gigawatt scale will cut its compute costs enough to matter in the AI infrastructure race.

By Amara Osei, Die Signal