Microsoft plans to scale custom AI chips across data centers starting in 2027, targeting cost reduction

Microsoft is preparing to expand its use of custom artificial intelligence chips across its cloud data center footprint, aiming for what it describes as a multi-gigawatt scale starting in 2027. The plan, first reported through a market-news post, reflects a broader industry push to control the power, hardware and supply-chain costs that come with training and running increasingly large AI models.

According to the report, Microsoft’s strategy is to scale “custom-designed” AI processors rather than rely solely on third-party chip supply. Custom silicon is often pursued to improve performance per watt and to better match hardware to the workloads used by a company’s own AI systems, though the post does not specify chip design details or performance targets.

The same report frames the multi-gigawatt figure as a capacity milestone. In practical terms, a multi-gigawatt scale indicates the level of electrical power expected to support AI computing operations across data centers once the rollout is underway. For cloud operators, that matters because energy and infrastructure costs increasingly influence the unit economics of AI, especially when demand is met by building or expanding capacity.

The announcement also indicates that Microsoft is treating chip planning as a long-term component of its cloud roadmap. A 2027 start date suggests the company is aligning semiconductor development and procurement cycles with data center buildout timelines, rather than reacting late to AI demand.

While the market-news report emphasizes cost reduction, it does not provide a breakdown of where those savings are expected to come from. Potential drivers in similar efforts across the sector include improved efficiency, reduced dependence on externally priced accelerators, and the ability to optimize software stacks around specific hardware. Microsoft did not outline those mechanisms in the cited post.

Microsoft has previously discussed building and operating large-scale infrastructure for AI through its Azure cloud platform and its broader AI services. In that context, custom accelerators can act as a lever to manage both capability and cost as AI workloads scale. The company’s newsroom is the primary place it typically publishes product and infrastructure updates, but the cited market-news item does not quote a new Microsoft statement that includes technical or financial specifics.

The report does not disclose whether Microsoft will use a single chip design across all workloads, whether it plans multiple generations through the 2027 rollout, or how much of its overall AI compute will be supported by these custom processors versus externally sourced alternatives. It also does not specify whether Microsoft has already secured supply arrangements for the multi-gigawatt deployment horizon.

Looking ahead, the key thing to watch is whether Microsoft follows up with more concrete details around the timeline, the types of workloads targeted, and the expected efficiency or cost impact. Investors and customers will also be watching for how quickly Azure capacity built for AI demand reflects this hardware strategy, and whether the company’s reported cost objectives translate into measurable changes in pricing or margins in future updates.