Microsoft Unveils Maia 300, Challenging Nvidia’s AI Chip Dominance

Microsoft’s Azure Maia 300 AI chip, designed to improve inference performance and reduce reliance on Nvidia’s GPUs. (Microsoft) Microsoft is mounting a direct challenge to Nvidia’s market dominance with the unveiling of its next-generation proprietary artificial intelligence (AI) chip, the Maia 300. As one of Nvidia’s largest customers, Microsoft plans to significantly increase its use of in-house chips, a move expected to reshape supply chain dynamics across the AI semiconductor market. On Oct. 12 (local time), Microsoft unveiled the Maia 300, its third-generation AI accelerator, and announced plans to use it as the primary AI inference system across its products and services. With full-scale production and deployment set to begin in 2027, more than 1 million units of the Maia 300 are expected to be supplied that year alone. Scott Guthrie, executive vice president of Microsoft’s Cloud and AI division, said, “We are entering a new phase of infrastructure as the era of superintelligence approaches,” emphasizing that the Maia 300 would process 40% more tokens per megawatt (MW) than leading graphics processing unit (GPU)-based systems. The Maia 300 incorporates cutting-edge technical specifications. Its computing chiplets, which handle core processing tasks, use TSMC’s 2-nanometer-class (N2P) process, while chiplets responsible for functions such as data movement use a 3-nanometer process, forming a multi-chiplet architecture. A key feature is the integration of eight HBM4 stacks, the sixth generation of high-bandwidth memory, delivering memory bandwidth of 18 terabytes (TB) per second. Power efficiency and processing capacity have also been substantially improved. A single rack housing 96 accelerators consumes approximately 167 kilowatts (kW) of power while delivering up to seven times the token throughput of its predecessor, the Maia 200. Microsoft designed the chip around an inference-first architecture optimized for AI workloads requiring long-context processing and memory-intensive computation, focusing on reducing service operating costs. Microsoft is not alone in developing proprietary AI chips. Google already operates Tensor Processing Units (TPUs) optimized for both training and inference workloads, while Amazon supports major partners such as Anthropic through its proprietary Trainium chips. OpenAI, the developer of ChatGPT, is collaborating with Broadcom to develop its own inference-specialized chip, Jalapeño, while Meta and Anthropic have also begun internally exploring chip design initiatives. These companies are focusing particularly on inference because the proliferation of AI agents is driving a sharp increase in the computing power required to generate responses. Their reasoning is that processing more responses using the same amount of power can improve the profitability of AI service operations. Rani Borkar, president of Microsoft Azure Hardware Systems, said that securing a competitive advantage requires designing the entire system together, encompassing not only chips but also memory, networking and power consumption. Despite growing competition from Big Tech companies, Nvidia’s influence is expected to remain strong for the time being. Nvidia continues to hold a dominant position in the market for chips used to train large-scale AI models, while companies including Microsoft continue to purchase its latest GPUs alongside their own chip development efforts. Nvidia’s second-quarter revenue rose 106% year-on-year to $96.2 billion, maintaining its rapid growth trajectory. Over the longer term, however, analysts expect Big Tech companies to gradually reduce their dependence on Nvidia as they strengthen control over the underlying technologies supporting their AI services. Evercore ISI projected that Maia would account for approximately 10% of Microsoft’s total AI semiconductor spending by the first quarter of 2027. Microsoft plans to unveil further details of its AI infrastructure strategy at its Ignite 2026 event in November as it accelerates the expansion of its proprietary chip ecosystem.