Microsoft is expanding its efforts to reduce the cost of running artificial intelligence services by deploying its own custom-designed chips at a much larger scale, as demand for AI computing continues to rise.
The company unveiled Maia 300 on October 12, its third-generation AI accelerator, and plans to begin large-scale deployments in 2027. Microsoft expects the chips to support services such as Copilot and its Azure cloud platform, potentially reducing its dependence on processors supplied by Nvidia.
According to Microsoft, Maia 300 is designed to deliver around 40% more useful computing per megawatt than leading GPU-based systems. The chip uses a chiplet architecture that combines advanced manufacturing technology for computing tasks with less expensive components for other functions. The company says this approach could improve efficiency while helping manage the costs of expanding AI infrastructure.
Microsoft plans to deploy the Maia 300 system at a scale measured in multiple gigawatts of computing capacity next year. Each rack will contain 96 accelerators and consume approximately 167 kilowatts of power. The company also expects to spend more than $175 billion on capital expenditure in 2026, underscoring the scale of investment required to support its AI ambitions.
The strategy reflects a broader shift among major technology companies towards developing their own processors to gain greater control over computing costs, performance and infrastructure. However, Microsoft will still face challenges in manufacturing and deploying its chips at scale, while competition from established chip suppliers remains strong.
Leave a comment