Microsoft Unveils Maia 300: Scaling AI Infrastructure with Third-Gen Custom Chip by 2027

On October 12, 2026, Microsoft announced the Maia 300, its third-generation in-house Azure Maia AI accelerator. This new chip is central to Microsoft's strategy to scale its custom Maia silicon to multi-gigawatt deployments starting in 2027, positioning it as the primary AI inference system across Microsoft's extensive product and service portfolio. The Maia 300 is engineered for demanding AI workloads, including long-context reasoning, latency-sensitive conversational tasks, and sustained multi-step agentic operations. For broader context, explore our AI Tools Pricing.

Maia 300: Designed for Next-Gen AI Workloads

The Maia 300 is set to enter production rollout in 2027, marking a significant step in Microsoft's commitment to developing specialized hardware for artificial intelligence. The chip's design focuses on optimizing performance for complex AI applications. Microsoft claims the Maia 300 delivers 40 percent more tokens per megawatt compared to leading GPU-based systems, indicating a focus on energy efficiency for large-scale operations.

Key specifications of the Maia 300 include eight HBM4 memory stacks, providing a total of 288 GB of memory and an 18 TB/s memory bandwidth. The chip is rated at 950 watts. It utilizes a new chiplet architecture built on TSMC's N2P process node, which separates compute, memory controller, and I/O functions to enhance performance and efficiency.

Performance Advancements Over Previous Generations

The Maia 300 represents a substantial leap in performance over its predecessor, the Maia 200. A 96-chip rack of Maia 300 is reported to produce up to seven times the token throughput of a Maia 200 rack. This improvement is critical for handling the increasing demands of advanced AI models and services.

The Maia 200, which is already operational, currently powers Microsoft Copilot workloads across the US, Europe, and Australia. It also supports models from OpenAI, MAI, and various open-source initiatives. Microsoft previously stated that the Maia 200 delivers up to 1.7 times more tokens per watt than leading GPUs when running Teams workloads.

Microsoft's Strategic Shift in AI Infrastructure

Microsoft's investment in its custom Maia silicon underscores a broader industry trend towards specialized hardware for AI. By developing its own accelerators, Microsoft aims to reduce reliance on external GPU providers like Nvidia and AMD, gain greater control over its AI infrastructure, and optimize for its specific software ecosystem. This strategic move is expected to enhance the performance and efficiency of AI services deployed across Azure and integrated into Microsoft products.

The plan to scale the Maia family to multi-gigawatt deployments by 2027 highlights the ambition behind this initiative. This scale suggests that Maia chips will form the backbone of Microsoft's future AI capabilities, supporting everything from cloud-based AI services to integrated features within applications like Microsoft Copilot.

Future Outlook and Implications

Further technical details regarding the Maia 300 are anticipated at Microsoft Ignite 2026 in November. These details are expected to provide deeper insights into the chip's capabilities and its role in Microsoft's long-term AI strategy. The rollout of Maia 300 in 2027 will be a key milestone, demonstrating Microsoft's ability to deploy its custom silicon at a massive scale to support its growing AI ambitions.

This development has implications for the broader AI hardware market, as major cloud providers increasingly opt for in-house chip development to meet their unique AI processing needs. The focus on energy efficiency and specialized architectures like chiplets indicates a mature approach to managing the computational and power demands of advanced AI.

Sources

• Microsoft to expand Azure Maia system to multi-gigawatt deployments - The Official Microsoft Blog
• Microsoft's latest custom chip: Maia 300
• Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs | The Verge