Microsoft has accelerated the launch timeline for its Maia 300 AI accelerator to September 2025, moving forward from the previously announced 2027 deployment. According to latest reports, the company is now planning to unveil the Maia 300 as early as September, representing a significant acceleration in its AI hardware strategy. This strategic move could significantly reduce Microsoft's reliance on Nvidia, which has held a near-lock on AI chip supply for training models, as the company seeks to establish its own competitive position in the rapidly expanding AI inference market.
The company's Maia 200 chip is already operational, including for the Microsoft 365 Copilot assistant, while the more powerful Maia 300 will launch in September 2025. As reported by Microsoft, the Maia 200 runs models from OpenAI, Microsoft's own in-house MAI models and leading open-source models, and is currently deployed in the United States, Europe and Australia. The chip delivers up to 1.7 times more output per watt than leading GPU-based systems for AI features built into Teams and Copilot, addressing the growing energy efficiency concerns in AI infrastructure. The Maia 300 will squeeze 40 percent more AI work out of the same amount of electricity than leading systems, and more than twice as much as Microsoft's own previous chip, according to company reports.
The Maia 300 will squeeze 40 percent more AI work out of the same amount of electricity than leading systems, and more than twice as much as Microsoft's own previous chip, according to company reports. Each server rack will hold 96 of the chips, twice as many as before, and handle up to seven times the workload. The new chip's computing components will be manufactured by Taiwan's TSMC, while the systems can be cooled with liquid or run in older data centers cooled by air. Microsoft engineers have focused on cutting power wasted moving data back and forth inside the machines, a growing problem as AI tools handle more complex tasks. The Maia 300 features eight HBM4 memory stacks delivering 18 TB/s of peak memory bandwidth at a 950-watt TDP, with a new chiplet design splitting compute, memory control and input/output into separate parts.
Microsoft's accelerated Maia chip strategy comes as the company reported strong Q4 fiscal results with revenue growing 18% to $90.0 billion and Azure achieving 43% growth. The company maintains its competitive advantages through cloud and distribution presence across Azure, Microsoft 365, GitHub, Dynamics, Copilot, and security services. Microsoft is focused on advancing the economics of AI and has reportedly paused some revenue-sharing agreements with smaller AI-cloud firms, while investors increasingly expect AI revenue to quickly facilitate the levels of spending on data centers, chips and power infrastructure. Microsoft will scale its in-house Maia chips to multi-gigawatt deployments starting in 2027, making them the main system for running AI across its products, as announced by executive vice president Scott Guthrie.
