Microsoft Debuts Maia 200 AI Chip to Challenge Cloud Rivals

Microsoft launches its latest second-generation AI chip Maia 200, the company describes it as an “AI inference powerhouse”, to power inference workloads across Azure and internal services.

The Superintelligence team at Microsoft will use Maia 200 for synthetic data creation and reinforcement learning to improve its upcoming in-house AI models

The 200, which is a second-generation chip released two years after the Maia 100’s release in October 2023, is set to be deployed at its central datacenter region near Iowa, following which it will be launched in other datacentres such as Arizona and more to follow the company said.

Maia 100 was not available to the public or cloud customers was built mainly for Microsoft’s internal use and testing, and customers could not rent or use it through Azure. With Maia 200 Microsoft plans to make it available to customers over time.

In a blog post on Monday, Microsoft cloud and generative AI solutions chief Scott Guthrie said the new chip will be made available to more customers in the future

The new chipset is part of the software giant’s mixed (heterogeneous) computing setup and will be used to run many different AI models, including OpenAI’s latest GPT-5.2, it said. It will help Microsoft deliver better performance at lower cost for services like Microsoft Foundry and Microsoft 365 Copilot, it added.

Microsoft in a release said Maia 200 is the fastest in-house AI chip developed by any major cloud provider, delivering around three times the FP4 performance of Amazon’s latest Trainium chip and higher FP8 performance than Google’s seventh-generation TPU.

The company said, “Maia 200 is the most efficient inference system Microsoft has ever deployed, with 30% better performance per dollar than the latest generation hardware in our fleet today.” Inference is the stage when an AI model is already trained and is being used to generate answers or predictions. It said that, “Maia 200 can effortlessly run today’s largest models, with plenty of headroom for even bigger models in the future.”

Tech Giants And In-house Chips

Microsoft is following other tech giants in reducing its reliance on the chip maker Nvidia’s GPU’s and producing chips in-house.

Google has been developing its Tensor Processing Units (TPUs) for over a decade, with its seventh-generation “Ironwood” TPU powering advanced AI workloads by 2025 and even training large models without Nvidia hardware.

Amazon Web Services (AWS) has rolled out multiple generations of its Trainium and Inferentia chips, including the newer Trainium 3 by late 2025, as part of its effort to reduce dependence on Nvidia GPUs even while it continues to offer Nvidia systems to customers.

Meta began testing its first in-house AI training chip in March 2025. And OpenAI announced a partnership with Broadcom to develop its first custom AI accelerators, slated for deployment starting in the second half of 2026, while it will still use large Nvidia GPU clusters under existing cloud deals.

Despite this, Nvidia remains dominant in the broader AI hardware market, with major players still using Nvidia GPUs alongside their custom silicon to meet compute demand. But by doing this companies are reducing their reliance on the chipmaker, “which controls an estimated 70% to 95% of the AI chip market, however, this isn’t the whole story,” said Kaoutar El Maghraoui, a Principal Research Scientist and Manager at IBM Research, in a report.

Reducing dependence on Nvidia shifts the centre of power on another giant that is the Taiwan Semiconductor Manufacturing Company Limited (TSMC), said Shobhit Varshney, Senior Partner at IBM Consulting, in a report. Maia 200 is built using TSMC’s advanced 3-nanometer manufacturing process and contains more than 140 billion transistors, Microsoft said.

The chip race has heated up because companies can now design chips tailored to specific AI models, which lowers costs and improves speed, said Varshney. She said, this makes it affordable to run high-volume tasks like fraud detection at scale, like use cases that were earlier too expensive to justify.

Microsoft said that it is now inviting developers, AI startups, and researchers to start testing and optimizing their models and workloads using the new Maia 200 software development kit (SDK).