Microsoft has stepped up its push into custom silicon with the launch of Azure Maia 300, its third-generation AI processor, as the company moves to shift most of its AI workloads onto the Maia platform from 2027 and reduce its dependence on Nvidia and AMD.

The chip was unveiled at the OCP Global Summit 2026 in San Jose, where Microsoft described it as a key part of the AI inference infrastructure behind its products and services. Inference — the stage of AI computing where trained models generate responses for users rather than where new models are trained — is the workload behind services like Cop and Teams, which is why Microsoft is concentrating its custom-silicon effort there.

According to Yahoo Finance, Microsoft says Maia 300 can process 40% more tokens per megawatt than leading GPU-based systems — a claim that places the chip in direct competition with Nvidia, whose GPUs dominate the AI accelerator market. In simple terms, it means Maia 300 can handle more AI workloads using the same amount of power.

Microsoft added that a rack equipped with 96 of the new chips can deliver up to seven times the throughput of a rack running the previous-generation Maia 200. Starting next calendar year, the company plans to scale the Maia family to multi-gigawatt deployments and make it the primary inference system for its own products and services.

Microsoft CEO Satya Nadella described the processor as “our biggest bet yet on our own silicon” in a post on X:

Today we are announcing our biggest bet yet on our own silicon. Starting next calendar year, we will scale our Maia family to multi-gigawatt deployments, making it the primary inference system for our own product and services. And we’re introducing Maia 300, which delivers 40% more tokens per watt compared to leading GPU-based systems. We are already deploying Maia 200 across the US, Europe, and Australia, delivering up to 1.7x more tokens per watt across experiences in Teams and Copilot. Our goal is to build on the full-stack efficiency gains we’ve already seen, while continuing to bring the latest innovations from across the ecosystem into our fleet.

— Satya Nadella (@satyanadella) October 12, 2026

Maia 200 already runs Copilot

Yahoo Finance reported that Microsoft’s Maia 200 chip already runs Microsoft 365 Copilot and supports a range of AI models, including those from OpenAI, Microsoft’s own MAI family, and other open-source developers. The company says the chip can process 1.7 times more tokens per watt than rival GPU systems when running Teams, Copilot, and Cowork.

Microsoft announced its first Maia chip in 2023, but delays pushed the launch of Maia 200 to January 2026. By this summer, it was running in just two US data centers, The Information reported. Microsoft is now expanding Maia 200 across data centers in the US, Europe, and Australia, with GitHub Copilot and Microsoft Foundry also expected to run on the chip before the end of the year.

Billions in projected savings, and a late start

GeekWire reported that the shift could save Microsoft billions of dollars and ease its reliance on Nvidia, while more efficient chips could also cut electricity draw at a time when power supply is one of the tightest constraints on all forms of AI expansion.

Evercore ISI analysts estimate that Maia will make up about 10% of Microsoft’s AI chip spending by early 2027, according to CNBC. The figure implies that, even as Maia scales, most of Microsoft’s AI chip spending would still go to outside vendors such as Nvidia and AMD in the near term.

Microsoft is also entering the custom AI chip market well after several rivals have made progress. Google sells its Tensor Processing Units to external customers, while Amazon’s Trainium chips have attracted AI companies such as Anthropic. Meta has shipped its own chip designs, and OpenAI unveiled its first custom chip, developed with Broadcom, in June. Maia, by contrast, stays entirely in-house and cannot be rented by any third parties.

Microsoft’s push to develop its own chips could have implications for Nvidia, which gets more than half of its quarterly revenue from hyperscalers, according to Yahoo Finance. The two companies continue to work together, however: Nadella and Nvidia CEO Jensen Huang held a joint appearance last week to unveil Nvidia-powered Windows PCs.

Microsoft has also said it will be among the first cloud providers to deploy Nvidia’s Vera Rubin and AMD’s Helios systems. The company plans to share further details at its Ignite conference in San Francisco next month — the next scheduled checkpoint for how Microsoft balances its in-house Maia ramp with its continued deployments of Nvidia and AMD hardware.