Microsoft May Deploy 1 Million Maia 300 AI Chips in 2027

According to Microsoft, the company unveiled its third-generation in-house AI chip, Maia 300, on Monday, October 12, 2026, with plans to deploy potentially more than one million units starting next year. The chip is designed to support Microsoft's internal AI services and improve overall system efficiency to reduce computing costs. As cloud service providers and AI developers accelerate their investments in custom silicon, Microsoft's planned large-scale deployment highlights the industry's efforts to reduce reliance on Nvidia's AI processors.

Microsoft Maia 300 Targets AI Computing Efficiency

According to Reuters, Maia 300 is designed to handle AI workloads, including software development and complex business tasks performed on behalf of users. Its predecessor has already been used to run Microsoft's proprietary AI models and OpenAI models delivered through Microsoft Azure cloud services.

Major cloud computing companies and AI developers, including OpenAI, have increasingly turned to custom-designed chips to challenge Nvidia's dominance in the AI computing market. Microsoft is pursuing a different hardware architecture from Nvidia, aiming to strike a balance between computing performance, operating costs, and power consumption.

Chiplet Architecture and Advanced Manufacturing Processes

Nvidia's AI accelerators typically combine large computing chips with high-bandwidth memory. In contrast, Maia 300 adopts a chiplet-based architecture that integrates multiple smaller chips, each designed to perform specific functions.

The chiplets responsible for key computing operations use TSMC's 2-nanometer manufacturing process, while other chiplets that handle data movement and communication between the processor and the rest of the system use the less expensive 3-nanometer process. This approach is intended to balance performance and manufacturing costs while supporting demanding AI workloads.

Maia 300 Server Racks Consume 167 Kilowatts

Microsoft plans to deploy Maia 300 in server racks containing 96 chips each, with power consumption of approximately 167 kilowatts per rack. For comparison, Nvidia's latest GB300 AI server racks contain 72 graphics processing units (GPUs) and consume around 142 kilowatts.

Microsoft claims that Maia 300 can deliver approximately 40% more useful computing output per megawatt of electricity than what it describes as leading GPU systems. However, the company has not disclosed the specific systems used in its comparison, so the figure should not be interpreted as a direct performance comparison with Nvidia's GB300 platform.

Rani Borkar, Microsoft's president of hardware systems and infrastructure for Azure, told Reuters that the company's goal is to optimize the entire system so that every watt of electricity, every dollar invested, and every byte of memory can support more AI computing.

Microsoft Expands AI Infrastructure Investment

Microsoft plans to scale Maia 300 deployment to the multi-gigawatt level starting next year. The company previously indicated that its capital expenditure for 2026 is expected to exceed $175 billion, reflecting its continued investment in AI infrastructure.

As demand for AI computing continues to grow, improving processing efficiency while controlling electricity consumption and hardware costs has become a key priority for cloud infrastructure operators. Microsoft's Maia 300 initiative underscores the growing role of custom AI chips, advanced semiconductor manufacturing, and chiplet integration in shaping the next generation of AI data centers.