Key Moments
- Microsoft introduced its third-generation custom AI chip, the Maia 300, designed for large-scale AI workloads on its Azure platform.
- The company indicated it may deploy more than one million Maia 300 chips next year in multi-gigawatt installations for its internal AI services.
- Microsoft said Maia 300 delivers about 40% more useful compute per megawatt than leading GPU-based systems, while using a chiplet design built on 2-nanometer and 3-nanometer process technologies.
New Custom AI Chip Aims to Power Internal Models
Microsoft announced a new generation of its proprietary AI accelerator on Monday, outlining plans to potentially install more than one million units next year to support its own artificial intelligence services.
The chip, named Maia 300, represents the third iteration in Microsoft’s line of custom silicon developed to run intensive AI workloads, including coding assistance and complex enterprise tasks for users. The prior generation is already in production, supporting both Microsoft’s internal AI models and models from OpenAI that are provided through the Azure cloud computing platform.
Challenging Nvidia With a Different Technical Blueprint
The introduction of Maia 300 arrives during a wave of custom silicon initiatives from major cloud operators and AI developers, including OpenAI, that are seeking to reduce reliance on Nvidia in the AI computing ecosystem.
In designing Maia 300, Microsoft adopted a different architecture from Nvidia’s AI accelerators, which center on a single large compute die paired with surrounding high-bandwidth memory components.
By contrast, the main compute engine of the Maia 300 consists of multiple “chiplets” – specialized dies bonded together – that employ 2-nanometer manufacturing technology from Taiwan Semiconductor Manufacturing Co for the most demanding computing tasks.
Additional chiplets in the Maia 300 use less expensive 3-nanometer process technology for roles such as transferring data off the processor and routing it within the larger system.
Power, Density, and Efficiency Comparisons
Microsoft plans to deploy Maia 300 processors in server racks that each contain 96 chips. These racks will consume about 167 kilowatts of power. For comparison, Nvidia’s latest GB300 AI racks, with 72 GPUs installed, draw roughly 142 kilowatts.
| System | Chip Count per Rack | Power Usage per Rack |
|---|---|---|
| Microsoft Maia 300 | 96 chips | 167 kilowatts |
| Nvidia GB300 AI server racks | 72 GPUs | 142 kilowatts |
According to Microsoft, Maia 300-based systems can deliver around 40% more “useful computing” per megawatt than what it described as “leading GPU-based systems,” although the company did not specify which systems were used in its benchmark.
“It’s basically delivering more intelligence for every watt, every dollar, and every byte of memory by optimizing across the entire system,” Rani Borkar, president of Azure hardware systems and infrastructure at Microsoft, told Reuters in an interview.
Scale-Out Plans and Capital Investment
Looking ahead, Microsoft said it intends to begin “multi-gigawatt deployments” of the Maia 300 starting next year as it scales its infrastructure for AI services. The company has previously indicated that it expects capital expenditures to exceed $175 billion in this calendar year, with the rollout of Maia 300 forming part of that broader investment in infrastructure.
TradingPedia.com is a financial media specialized in providing daily news and education covering Forex, equities and commodities. Our academies for traders cover 
