Microsoft has unveiled Maia 300, its third-generation AI chip, and plans to install more than one million units in 2027. The company wants to run its AI services faster and at a lower cost. The chip was announced on Monday, October 12.
Maia 300 is built for heavy AI work, such as writing code and carrying out complex business tasks for users. It follows Maia 200, which already powers Microsoft's own AI models and OpenAI models on its Azure cloud platform.
Inside Microsoft's Maia 300 AI Chip Design
Microsoft chose a chiplet design instead of one large processor. The main computing parts use TSMC's 2-nanometer technology. Other parts, which move data around, use cheaper 3-nanometer technology.
Microsoft says Maia 300 can give about 40% more useful computing per megawatt than top GPU-based systems. Each server rack will hold 96 chips and use around 167 kilowatts of power. Nvidia's GB300 racks hold 72 GPUs and use about 142 kilowatts.
Rani Borkar, Microsoft's president of Azure hardware systems and infrastructure, told Reuters the company is tuning the whole system. The goal is better performance for every watt, dollar, and unit of memory.
Also Read: Microsoft Decision-1: Faster, Cheaper AI Model Built for Structured Business Decisions
How Maia 300 Could Reshape Microsoft's AI Costs
Big tech firms are building their own chips so they depend less on Nvidia, whose GPUs lead the AI market. Microsoft is following the same path. Making its own chips gives it more control over its computing setup and helps it keep costs down.
The tech giant will begin multi-gigawatt rollouts of Maia 300 in 2027. This is part of a larger push to add computing power as demand for AI keeps rising. The company expects its capital spending to pass USD 175 billion in 2026.
Maia 300 shows how seriously Microsoft takes its own chip plans. If the company meets its target of a million chips, it could cut its AI costs and rely less on Nvidia. Other tech giants are on the same road, so the race to build better custom chips is only getting started.