Microsoft has unveiled Maia 300, its third-generation custom AI chip, and said it could deploy more than one million units in 2027 as it expands its in-house AI infrastructure.
Announced on October 12, 2026, the processor is designed for demanding workloads including software development, data processing and tasks handled by autonomous AI agents. Microsoft said it will support the company’s internal AI services and Azure-based models.
Maia 300 uses a chiplet-based design, with its core computing functions built on Taiwan Semiconductor Manufacturing Company’s 2-nanometer process. Other components use 3-nanometer technology to manage data transfers and balance performance with manufacturing costs.
Microsoft said server racks fitted with 96 Maia 300 chips will consume about 167 kilowatts, compared with 142 kilowatts for Nvidia GB300 racks containing 72 GPUs. Despite the higher total power draw, the company claims Maia 300 delivers about 40% more useful computing per megawatt than leading GPU-based systems. Microsoft did not identify the systems used in the comparison, and the claim has not been independently verified.
The planned rollout will target multi-gigawatt computing installations in 2027, helping Microsoft increase capacity while seeking greater control over AI infrastructure costs.
Source: EntArabi
