Microsoft to launch Maia 300 AI chip in 2027

Microsoft (MSFT) said on Monday it plans to introduce its Maia 300 AI chip in 2027, as it expands its in-house hardware to handle growing demand for AI services.

The chip is designed to run AI models after they have been trained, a process known as inference, including those used in chatbots, coding assistants, and systems that carry out tasks in multiple steps.

"Maia 300 is designed with an inference-first architecture optimized for long-context, memory-intensive and decode-heavy AI workloads. This involves balancing compute, memory bandwidth, networking and power efficiency at cloud scale," Microsoft said.

Its design combines multiple chiplets, with computing components built using Taiwan Semiconductor Manufacturing Co's (TSM) N2P manufacturing process and eight stacks of high-bandwidth memory.

The company said a rack containing 96 Maia 300 accelerators could deliver up to seven times the token throughput of a Maia 200 rack. Tokens are units of text processed or generated by AI models. Microsoft also expects Maia 300 to generate 40% more tokens per megawatt than leading graphics processing unit-based systems.

In January, Microsoft released AI accelerator chip Maia 200, built for inference and to improve the efficiency of token generation, built by Taiwan Semiconductor Manufacturing. Microsoft introduced the Maia chip ⁠in November 2023 but has lagged ​its peers in ramping it up ​to scale as it looks to reduce its reliance on Nvidia's (NVDA) costly processors.

The company's Maia 200 chip is already running production workloads, including Microsoft Copilot. Microsoft said it plans to extend its use to GitHub Copilot and Microsoft Foundry by the end of 2026.

Microsoft said it plans to begin scaling its Maia systems to multi-gigawatt deployments in 2027, with the chips intended to handle a larger share of AI workloads across its products and services.