Microsoft's Maia 300 AI Chip to Boost Inference Performance by 40%

Written by: Rita

On October 12, Microsoft announced that the Maia accelerator family will expand to multi-gigawatt deployments starting in 2027, making Maia the primary inference system for Microsoft’s products and services. The third-generation Maia 300 is scheduled for production in 2027, delivering 40% more tokens per megawatt than leading GPU systems and achieving seven times the rack-level token throughput of the Maia 200 in a full 96-accelerator configuration.

In a report released on October 12, 2026, J.P. Morgan noted that Reuters reported Microsoft plans to deploy over one million Maia 300 chips in 2027 to support its internal AI services. J.P. Morgan maintains an Overweight rating on Microsoft with a price target of $625. The Maia 200 is already supporting Copilot production workloads, with GitHub Copilot and Microsoft Foundry expected to follow by year-end.

Microsoft plans to expand the Maia family for multi-GW deployment, with production starting in 2027. Maia will become the primary inference platform for Microsoft’s products and services. The Maia 200 already powers Microsoft Copilot and supports the MAI team in developing next-generation models. GitHub Copilot and Microsoft Foundry are expected to follow by the end of 2026.

Microsoft has prior experience supporting OpenAI models, MAI models, and leading open-source model families. J.P. Morgan believes that Maia’s transition from an auxiliary role to a primary inference system represents a key shift in Microsoft’s AI infrastructure strategy.

The deployment of DigiWa means that Maia will not only serve internal workloads but also support Microsoft Cloud’s overall inference demands. J.P. Morgan expects Maia’s deployment to primarily focus on internal use cases, without pursuing external compute revenue. The most direct impact investors will observe is on capital expenditure intensity and cloud software cost of sales.

Maia 300 delivers 40% more tokens per megawatt than leading GPU systems. With a full 96-accelerator configuration, rack-level token throughput is 7 times that of Maia 200. Maia 200 delivers 1.7 times more tokens per watt than GPU infrastructure with built-in AI capabilities.

Performance improvements come from software-defined data flows and innovations in computing architecture. Hardware innovations include chiplet architectures, enabling higher bandwidth per watt for high-bandwidth memory (HBM), as well as vertically scaled networking at data center scale.

JPMorgan noted that the increased token throughput of Maia 300 directly enhances Copilot performance while reducing cost per compute unit. Microsoft Cloud’s gross margin currently ranges between 60% and 70%, and the deployment of Maia is expected to further improve this margin.

The software-defined data flow of the Maia 300 optimizes data movement efficiency for inference tasks. Computational architecture innovations include a chiplet design that enhances HBM bandwidth per watt. A scalable network supports inference clusters at data center scale. Together, these innovations drive performance improvements and cost reductions.

Reuters reports that Microsoft plans to deploy over 1 million Maia 300 chips by 2027. Microsoft also aims to more than triple its global data center capacity by 2032. J.P. Morgan expects Microsoft’s data center capacity to exceed 30 gigawatts by 2030.

Large-scale deployment of the Maia family of chips, combined with expanded data center capacity, is expected to drive a more favorable cost structure. J.P. Morgan anticipates Microsoft’s capital expenditures to be approximately $210 billion in fiscal year 2027 and about $175 billion in calendar year 2026.

JPMorgan believes that Maia's deployment, focused on internal use cases, impacts capital expenditure intensity and cloud software sales costs. A million-chip-scale deployment, combined with data center capacity expansion, will optimize Microsoft's overall infrastructure costs.

The deployment of millions of chips corresponds to multi-gigawatt capacity; Maia 300 increases the number of tokens per megawatt by 40%, significantly boosting inference throughput for the same power input. JPMorgan expects Microsoft’s data center capacity to exceed 30 gigawatts by 2030, with capital expenditures reaching approximately $210 billion in fiscal year 2027. Economies of scale and improved chip efficiency are jointly driving down the cost per inference.

Maia, as the primary inference system, will enhance Copilot's performance while reducing cost per compute unit. Microsoft Cloud's gross margin currently sits in the high-end range of 60% to 70%. J.P. Morgan believes that the deployment of Maia has the potential to improve gross margins.

Maia's deployment focuses on internal use cases, with external computing revenue not being a priority. Investors are concerned with capital expenditure intensity and the cost of cloud software sales. J.P. Morgan expects that Maia's deployment will optimize Microsoft Cloud's cost structure and improve gross margins.

JPMorgan maintains an Overweight rating on Microsoft with a price target of $625. The performance improvements of Maia 300 and the plan to deploy millions of chips support Microsoft’s long-term competitiveness in AI infrastructure. Declining inference costs and improved cloud gross margins are key metrics to monitor next.

Disclaimer

This article is a compilation and interpretation by Chaoxiang Research of a third-party brokerage research report (JPMorgan, October 12, 2026), combined with publicly available market information. The ratings, price targets, earnings forecasts, and related judgments cited herein reflect the views of the brokerage’s analysts and represent the position of their respective institution only; they do not reflect the views of Chaoxiang Research nor constitute any investment advice.

The market carries risks; make decisions independently. This article should not be used as a basis for buying or selling any securities.