Microsoft plans to make its own Maia chips the primary hardware for running AI across its products starting in 2027.
The company outlined the strategy while unveiling Maia 300, the third generation of its custom AI accelerator.
Microsoft expects to deploy Maia systems at multi-gigawatt scale beginning next year. That level of deployment could translate to more than one million Maia 300 chips in 2027.
Built for AI Inference
Maia 300 is designed primarily for inference, which is the process of running an AI model after it has already been trained.
Microsoft says the chip is optimized for workloads such as:
- Long conversations
- AI reasoning
- AI agents
- Multi-step tasks
The company is positioning Maia 300 as a core piece of infrastructure for its growing AI services.
40% More Tokens Per Megawatt
Microsoft claims Maia 300 can generate 40% more tokens per megawatt of power than leading GPU-based systems.
The company did not identify which competing GPU systems it used for that comparison.
Microsoft also says a full Maia 300 rack can deliver up to seven times the output of Maia 200.
96 Chips Per Rack
Each Maia 300 rack contains 96 chips and consumes around 167 kilowatts of power.
For comparison, Nvidia’s latest GB300 racks contain 72 GPUs and draw around 142 kilowatts.
That makes Maia 300 a denser system in terms of accelerator count, although Microsoft’s focus is on improving output per watt rather than simply reducing rack power consumption.
TSMC N2P and HBM4
Microsoft has moved to a chiplet-based design for Maia 300.
Different parts of the processor handle computing, memory, and data transfer.
The compute chiplets are manufactured using TSMC’s N2P process, while each Maia 300 carries eight stacks of HBM4 memory.
Microsoft says the design is aimed at improving efficiency across power, cost, and memory usage.
Already Running Microsoft AI
The previous-generation Maia 200 is already being used inside Microsoft.
It currently runs workloads for Microsoft Copilot and supports both OpenAI models and Microsoft’s own MAI models.
Microsoft also plans to use Maia 200 for GitHub Copilot and Microsoft Foundry by the end of the year.
Reducing Reliance on Nvidia
The Maia program is part of a broader industry shift among large cloud providers and AI companies to develop their own chips.
Microsoft is following a similar path to companies such as Google and Amazon, which already operate custom AI accelerators.
OpenAI has also been developing its own AI chip with Broadcom.
The strategy could give Microsoft greater control over cost, power consumption, and hardware availability as demand for AI computing continues to rise.
Microsoft Could Produce Hundreds of Thousands of Chips
Microsoft has also been working to secure enough manufacturing capacity for the planned rollout.
Earlier reports indicated that the company was in discussions with TSMC to manufacture more than 300,000 Maia 300 chips for 2027.
Microsoft now says its deployment plans will be measured in gigawatts, suggesting eventual volumes could be substantially higher.
The company plans to reveal more details about Maia 300 at its Ignite conference in San Francisco next month.
The post Microsoft Plans to Run Its AI Products Primarily With Its Own AI Chips appeared first on ProPakistani.