Microsoft unveils Maia 300 chip to power its AI from 2027

Microsoft ↘ 1.93% on Monday unveiled the Maia 300, the third generation of its custom AI chip. The company said its Maia chips will become the "primary inference system for Microsoft products and services" starting in 2027, and that it will scale the Maia family to deployments measured in gigawatts. Reuters reported the rollout could mean more than a million Maia 300 chips next year.thenextweb+1

The plan is Microsoft's clearest step yet toward running its own AI services, including Copilot, on in-house silicon instead of processors bought from Nvidia ↘ 1.43% and AMD Advanced Micro Devices, Inc. ↘ 5.95%. Microsoft has framed it as a way to save on costs.finance.yahoo+1

Efficiency claims

Microsoft presented the chip at the OCP Global Summit in San Jose. It said Maia 300 will deliver 40% more tokens per megawatt than "leading GPU-based systems". It did not say which systems it compared against. A full rack holds 96 chips and draws about 167 kilowatts. Microsoft says that rack produces up to seven times the token throughput of a Maia 200 rack. Reuters noted that Nvidia's GB300 racks hold 72 GPUs and draw about 142 kilowatts.blogs.microsoft+2

Each chip has eight stacks of HBM4 memory, with 288 GB of capacity and 18 TB/s of bandwidth, within a 950-watt power budget. The compute chiplets are made on TSMC's Taiwan Semiconductor Manufacturing Company Limited ↘ 2.28% N2P process. Microsoft says the chip is built for inference, which means running AI models after they have been trained. It is aimed at long-context reasoning, conversational AI and agents that carry out multi-step tasks.microsoft+1

"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, told Reuters.thenextweb

Maia 200 already in service

The previous generation is already running in Microsoft's data centers. Maia 200 serves Microsoft Copilot and runs OpenAI models, Microsoft's own MAI models and open-source models. GitHub Copilot and Microsoft Foundry are due to move onto it by the end of the year. Microsoft says Maia 200 produces up to 1.7 times more tokens per watt than leading GPU infrastructure for AI features in Teams and Copilot.blogs.microsoft

In August, The Information reported that Microsoft was in talks with TSMC to manufacture more than 300,000 Maia 300 chips for 2027.thenextweb

Pressure on Nvidia

Amazon Amazon.com, Inc. ↘ 1.87%, Google parent Alphabet ↘ 2.32% and Meta ↘ 3.72% are also building their own chips, as are OpenAI and Anthropic. That could become a problem for Nvidia, which regularly gets more than half of its quarterly revenue from hyperscalers, according to Yahoo Finance. Microsoft shares rose in Monday trading, while Nvidia and AMD shares fell.finance.yahoo

Analysts caution that the large cloud companies are unlikely to stop buying Nvidia chips altogether. Customers still want a range of processors for different workloads, and for now Nvidia's remain the most in demand.finance.yahoo

Microsoft plans to share more details at its Ignite conference in San Francisco, which runs Nov. 17–20.microsoft