Microsoft announced its latest custom artificial intelligence processor on Monday, unveiling the Azure Maia 300 as the company moves to power more of its cloud services with internal silicon and reduce its dependence on outside chipmakers such as Nvidia.
Introduced at the OCP Global Summit 2026 in San Jose, the third-generation chip is designed to serve as the primary inference engine across Microsoft’s internal product catalog, according to company executives. The company plans to begin production rollouts in 2027 across multi-gigawatt data center deployments, The Verge reported.
Efficiency gains and architecture
Built on Taiwan Semiconductor Manufacturing Co.’s N2P process node, the Maia 300 utilizes a chiplet design that separates compute, input-output, and memory control functions. Microsoft claimed the architecture allows it to deliver 40 percent more tokens per megawatt than leading graphics processing unit systems, Yahoo Finance reported. In artificial intelligence systems, tokens represent basic processing units for text, images, and other media.
Scott Guthrie, executive vice president of Microsoft’s cloud division, wrote in an announcement that a 96-accelerator rack configuration can achieve up to seven times the token throughput of the prior Maia 200 generation. Microsoft also stated the chip provides more than 2.5 times the performance per watt and 33 percent more memory capacity than its predecessor, while scaling into clusters of up to 12,288 accelerators.
Shifting away from external suppliers
The announcement marks Microsoft’s latest effort to lower infrastructure expenses and gain greater autonomy over its computing hardware. For years, major tech firms have relied heavily on Nvidia’s graphics processing units to train and run large generative AI models. That dynamic has fueled soaring profits for Nvidia, which regularly draws more than half of its quarterly revenue from major cloud providers, Yahoo Finance noted.
However, major cloud operators including Microsoft, Amazon, Alphabet’s Google, and Meta have increasingly designed proprietary silicon to mitigate supply bottlenecks and curb equipment costs. Finance Chief Amy Hood noted at an investor conference in September that using proprietary chips has contributed to margin improvements over the past decade, CNBC reported.
Microsoft initially launched the Maia 100 in 2023 and introduced the Maia 200 in January. Chief Executive Satya Nadella told investors in July that the Maia 200 offered 30 percent better performance per dollar compared with existing fleet hardware while handling OpenAI models. The Maia 200 already supports services such as Microsoft 365 Copilot and GitHub Copilot, according to company reports.
Next steps in AI deployment
Microsoft AI Chief Executive Mustafa Suleyman emphasized that proprietary silicon is crucial for the company’s internal model development as it competes with rivals like Google DeepMind, OpenAI, and Anthropic. “At that scale we need a platform that we own, and that we co-design with our models,” Suleyman said, according to The Verge.
Despite the internal push, Microsoft is not abandoning commercial processors entirely. Cloud providers still require chips from vendors such as Nvidia and Advanced Micro Devices to meet diverse customer requests. Alistair Speirs, a general manager at Microsoft, declined to specify rigid deployment schedules, telling CNBC that the company reserves flexibility to match specific silicon with rapidly evolving models.
Microsoft plans to share additional details about its computing road map and infrastructure strategy next month at its Ignite conference in San Francisco.
