Microsoft Plans to Unveil Maia 300 AI Chip in September: Report

Microsoft plans to unveil its next-generation Maia 300 AI chip as soon as September, according to a report by The Information, as the company seeks to reduce dependence on Nvidia's GPUs. The chip follows the Maia 200, introduced in January and manufactured by TSMC on a 3-nanometer process, and Microsoft has been in talks with TSMC to secure capacity for more than 300,000 units with delivery targeted for 2027. Andrew Wall, general manager for Azure Maia, said the reported production figures don't reflect the scale of the program.

August 11, 2026 , Inside AI — Microsoft is preparing to unveil its next-generation AI chip, the Maia 300 , as soon as September, according to a report by The Information citing people with direct knowledge of the plans. The launch marks a critical step in the company’s multi-year effort to build custom silicon and reduce dependence on Nvidia ’s expensive GPUs. The Maia 300 follows the Maia 200 , which Microsoft introduced in January, manufactured by TSMC on a 3-nanometer process. That chip packed substantial SRAM memory to accelerate AI inference for large-scale user requests. The new chip is expected to push performance further as Microsoft races to catch up with rivals Alphabet and Amazon , which have already commercialized their custom AI silicon. Microsoft first entered the custom AI chip arena with the original Maia in November 2023 , but its in-house efforts have lagged. Google’s Tensor Processing Units TPUs began generating direct sales revenue in the quarter ending June, while Amazon’s Trainium chips have seen growing adoption. Both companies have moved faster to integrate custom silicon into their cloud offerings, putting pressure on Microsoft to accelerate. The company has been in talks with TSMC to secure manufacturing capacity for more than 300,000 Maia 300 units, with delivery targeted for 2027 , according to the report. Microsoft also aims to persuade major cloud customers, including Anthropic , to adopt the chip. However, component supplies and ongoing capacity negotiations with TSMC could constrain its ambitions, even as it ultimately seeks capacity for over 1 million chips. Andrew Wall, general manager for Microsoft’s Azure Maia, pushed back on the reported production figures. “Microsoft continues to invest in custom silicon as part of our long-term AI infrastructure strategy. While we don’t share production volumes, the figures reported don’t reflect the scale of our program,” Andrew Wall, general manager, Azure Maia, Microsoft. TSMC did not respond to requests for comment outside regular business hours. The Taiwanese foundry is a linchpin in the global AI chip supply chain, manufacturing advanced processors for Nvidia, AMD, and now Microsoft. Securing adequate capacity at TSMC has become a strategic bottleneck, with lead times stretching years and competition intensifying among tech giants. Microsoft’s push into custom silicon is part of a broader industry shift. Cloud providers are increasingly designing their own chips to optimize performance for specific AI workloads and to mitigate the soaring costs and supply constraints of Nvidia’s GPUs. Amazon’s Inferentia and Trainium, Google’s TPUs, and now Microsoft’s Maia family all reflect this trend. However, building a competitive chip ecosystem requires not just hardware but also a robust software stack to attract developers and customers. Can Microsoft Close the Custom Silicon Gap? Despite the Maia 300’s imminent debut, Microsoft faces significant hurdles. Rivals have years of experience refining their architectures and cultivating developer communities. Google’s TPU v5p, for instance, is already powering large-scale training and inference for models like Gemini . Amazon’s Trainium2, announced in 2023, promises up to four times faster training than its predecessor. Microsoft must demonstrate that Maia can deliver comparable or better performance per dollar to win over cost-conscious cloud customers. The reported target of 300,000 units by 2027 suggests a cautious ramp-up. By contrast, Nvidia shipped an estimated 2.5 million data center GPUs in 2024 alone. Even if Microsoft hits its goal, Maia will remain a small fraction of the AI chip market. The company’s strategy appears to focus on inference workloads where custom designs can offer clear advantages in latency and energy efficiency, rather than directly challenging Nvidia’s dominance in training. Anthropic, a potential early adopter, is an interesting test case. The AI startup has been using Google’s TPUs alongside Nvidia GPUs for its Claude models. If Microsoft can convince Anthropic to shift some workloads to Maia, it would validate the chip’s competitiveness. However, such migrations are complex and risky, requiring extensive software retooling. Supply Chain Realities Temper Ambitions Microsoft’s ultimate aim of securing capacity for over 1 million Maia 300 chips is ambitious but faces real-world constraints. TSMC’s advanced packaging capacity, particularly for CoWoS Chip-on-Wafer-on-Substrate , is heavily booked by Nvidia and others. Expanding production requires years of planning and billions in investment. Even with preferential treatment, Microsoft may struggle to scale as fast as it wants. The Maia 300’s reliance on SRAM, while beneficial for speed, also makes the chip larger and more expensive to manufacture. SRAM does not shrink as easily as logic transistors with each new process node, limiting cost reductions. This could make Maia less attractive for price-sensitive customers compared to alternatives using high-bandwidth memory HBM . Microsoft’s custom silicon journey is entering a critical phase. The Maia 300 will be a litmus test of whether the company can translate its massive AI investments into competitive hardware. Success could reshape the cloud AI landscape; failure would leave Microsoft even more dependent on Nvidia’s roadmap and pricing. The September unveiling will offer the first concrete clues.