Microsoft Plans to Deploy One Million AI Chips in Major Infrastructure Push
Microsoft plans to deploy more than one million artificial intelligence chips of its own design over the coming year, aiming to support its internal AI services, reduce computing costs and strengthen its position in an increasingly competitive market for custom-designed processors that could reduce reliance on Nvidia.
The Maia 300 Chips
Microsoft recently announced its next-generation AI chip, the Maia 300, the third generation of its custom-designed processors for AI workloads. These include tasks such as writing code and carrying out complex business operations on behalf of users.
Microsoft already uses the previous generation of Maia chips to power its internally developed AI models and OpenAI models through its Azure cloud computing platform.
Reducing Reliance on Nvidia
The new chip arrives as cloud computing companies and AI developers, including OpenAI, expand efforts to develop custom silicon and challenge Nvidia’s dominance in the market for processors that run AI applications.
Microsoft has also adopted a different chip architecture from Nvidia. Nvidia’s AI accelerators traditionally rely on a large central computing die surrounded by high-bandwidth memory chips.
By contrast, the main computing processor in the Maia 300 is built from units known as chiplets. These are specialized semiconductor components linked together to function as a single processor.
The chiplets responsible for core computing functions also use a 2-nanometer manufacturing process developed by Taiwan Semiconductor Manufacturing Company (TSMC).
Other components also use the less expensive 3-nanometer process to handle functions such as transferring data out of the processor and to other parts of the system.
Power Consumption and Computing Efficiency
Maia 300 processors are integrated into server systems containing 96 chips, with power consumption of approximately 167 kilowatts.
For comparison, Nvidia’s latest GB300 AI server systems consume around 142 kilowatts and incorporate 72 graphics processing units.
Microsoft says the Maia 300 can deliver approximately 40% more useful computing work per megawatt of power than leading GPU-based systems, although the company has not identified the systems it used for comparison.
“It simply delivers more intelligence per watt, per dollar and per byte of memory by optimizing the performance of the entire system,” Rani Borkar, Microsoft’s head of hardware systems and infrastructure for Azure, told Reuters in an interview.
Large-Scale Deployment Planned for 2027
Microsoft said it plans to begin deploying Maia 300 chips at a multi-gigawatt scale starting next year, reflecting the substantial computing infrastructure the company is preparing to support its AI services.
The company also expects its capital expenditure to exceed $175 billion this year as it continues investing in computing infrastructure, data centers and artificial intelligence.
Related Topics:
LEAP 2025 Conference: Launch of the Arabic Computing Observatory
Ray Dalio Warns AI Boom Is Approaching a Potential Breaking Point
Elon Musk Launches Grok 3: World’s Smartest AI Merges with Blockchain to Disrupt Crypto Sector



