JPMorgan: Microsoft Maia300 boosts inference performance by 40%, million-chip deployment cuts costs

October 14: A JPMorgan research report dated October 12, 2026, pointed out that Microsoft's Maia 300 delivers 40% more tokens per megawatt than leading GPU systems, and in 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; Reuters reported that Microsoft plans to deploy more than 1 million Maia 300 chips in 2027; Microsoft's capital expenditure for fiscal year 2027 is expected to be about $210 billion, and about $175 billion for calendar year 2026.

JPMorgan judges that Maia becoming the primary inference system will improve Copilot performance and reduce cost per compute unit, and Microsoft's cloud gross margin is expected to improve. It maintains an overweight rating and a target price of $625.