Microsoft Unveils Second‑Generation AI Chip to Strengthen Cloud Capabilities

TLDR

• Microsoft launched the Maia 200 chip to support demanding cloud AI tasks and compete with Nvidia, AWS, and Google chips.
• The Maia 200 delivers 30% better performance at the same cost and uses a 3nm process for optimized power and speed.
• Microsoft will first install Maia 200 in U.S. Central data centers, followed by U.S. West 3 and other global locations.
• A preview software development kit is available for developers, researchers, and labs working with open-source AI models.
• The Maia 200 will power Microsoft 365 Copilot, Microsoft Foundry, and internal superintelligence teams led by Mustafa Suleyman.

Microsoft has introduced a new artificial intelligence chip called the Maia 200 that targets cloud computing workloads. The chip arrives two years after Microsoft revealed its first AI chip, the Maia 100, which never became widely available for cloud clients. The company said Maia 200 will reach more customers and provide broader availability in the future.

Maia 200 Aims to Compete in Cloud AI Processor Market

Microsoft announced that the Maia 200 could serve as an alternative to processors from Nvidia, Amazon’s Trainium, and Google’s TPUs. Scott Guthrie, Microsoft’s executive vice president for cloud and AI, said the Maia 200 brings “wider customer availability in the future.” Microsoft said the Maia 200 represents its most efficient inference system deployed to date.

Developers, academics, and AI labs can apply for a preview of the Maia 200 software development kit. The preview offers early access to tools for building and optimizing AI workloads on the new chip. Microsoft said the preview will broaden experimentation across open source AI models and enterprise use cases.

Microsoft said its superintelligence team, led by Mustafa Suleyman, will use the Maia 200 for internal and customer workloads. The company also confirmed that Microsoft 365 Copilot and Microsoft Foundry will run on the new chip. These services include productivity software add‑ons and frameworks for building on top of large AI models.

Cloud providers face growing demand from AI model developers such as Anthropic and OpenAI, Microsoft said. Data center operators seek higher computing power while managing energy and cost constraints. In this competitive environment, companies aim to balance performance against operational expenses and energy use.

Technical Details and Deployment Plans for Maia 200 Chips

Microsoft said the Maia 200 chips use Taiwan Semiconductor Manufacturing Co.’s 3‑nanometer process. The company said it places four Maia 200 chips inside each server, interconnected for increased throughput. Microsoft explained that the design uses Ethernet cables rather than the InfiniBand standard seen in Nvidia installations.

The company stated that the Maia 200 delivers 30 percent higher performance at equivalent price points compared with alternatives. Microsoft said each Maia 200 includes more high‑bandwidth memory than AWS’s Trainium or Google’s seventh‑gen TPUs. This architectural design aims to support large‑scale model inference workloads.

Microsoft also said it can link up to 6,144 Maia 200 chips to scale performance even further. The company claimed that this approach helps reduce energy use and total cost of ownership. Microsoft previously showed that the Maia 100 could run GitHub Copilot in 2023.

Microsoft said it will deploy Maia 200 chips first in its U.S. Central data center region. The company said chips will later arrive in the U.S. West 3 region. Additional global deployments will follow these initial rollouts.

The post Microsoft Unveils Second‑Generation AI Chip to Strengthen Cloud Capabilities appeared first on Blockonomi.