Microsoft announces Maia 300 AI chip rollout for 2027

In short: Microsoft introduced its third-generation Maia AI chip, and it plans to start using it for real, day-to-day AI work in 2027.

What happened

Microsoft announced Maia 300, a computer chip it designed itself to run AI in its cloud services. A chip is the small piece of hardware that does the heavy lifting, like the engine in a car.

Microsoft says Maia 300 is meant to be the main system for “AI inference,” which is the step where an AI model gives answers after it has already been trained. Inference is what happens when you ask Copilot a question and it responds.

The company says Maia 300 is built to run AI more efficiently. Microsoft executive Scott Guthrie said the chip should deliver 40 percent more “tokens per megawatt” than leading systems based on GPUs, which are popular chips often supplied by companies like Nvidia and AMD. Tokens are small chunks of text that AI models read and write, so more tokens per unit of energy means more AI output for the same electricity.

Microsoft also said that in a setup using 96 accelerators (special chips for AI), Maia 300 can deliver up to seven times the “rack-scale” throughput of Maia 200. A rack is like a tall shelf in a data center that holds many computers.

Why it matters

If Microsoft can run more AI on its own chips, it can depend less on outside suppliers and may reduce costs and power use in its data centers. Over time, that could affect how quickly Microsoft can add AI features to products people use, such as Microsoft Copilot and GitHub Copilot.

Source: The Verge AI