Microsoft to unveil Maia 300 AI chip in September, orders 300,000 units from TSMC

Microsoft is preparing to unveil the Maia 300, its next-generation custom AI chip, as early as September 2026, according to a report from The Information citing people with direct knowledge of the plans. The announcement would mark the most significant escalation yet in the company's effort to build an alternative to Nvidia's dominant AI hardware.

The company has engaged Taiwan Semiconductor Manufacturing Co. (TSMC) to secure manufacturing capacity for more than 300,000 units of the Maia 300, with deliveries expected in 2027, the report said. Microsoft is also in talks to persuade major cloud customers, including AI startup Anthropic, to adopt the chip for their AI workloads.

The Push to Break Free from Nvidia

Microsoft's custom silicon journey began in earnest with the original Maia chip, announced in November 2023 and built on a 5-nanometer process. The Maia 200 followed in January 2026, designed for large-scale AI training and inference on Azure. But production volumes remained limited relative to internal targets, and the company lagged behind rivals that have already taken their custom chips to market.

Google's Tensor Processing Unit v5p is broadly available on Google Cloud. Amazon's Trainium2 is in active ramp-up for large-scale training workloads. The Maia 300 is Microsoft's bid to close that gap — and the scale of the TSMC order signals this is no longer a pilot program.

Andrew Wall, Microsoft's general manager for Azure Maia, acknowledged the chip program in a statement: "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."

Anthropic Adoption Would Signal a Shift in the AI Chip Market

If Anthropic — one of Nvidia's largest customers and the company behind the Claude family of models — were to move workloads to the Maia 300, it would represent one of the first significant defections from the Nvidia CUDA ecosystem. Such a move would also validate Microsoft's chip beyond its own services like Azure OpenAI and Copilot.

That transition is far from guaranteed. Nvidia's CUDA software stack is deeply entrenched, and porting large transformer models to a new architecture demands significant engineering investment. Microsoft will need to demonstrate not just raw performance but a mature software ecosystem — support for PyTorch, TensorFlow, and the tooling developers actually use — to win over wary AI labs.

The AI Chip Arms Race Intensifies

The Maia 300 arrives as AI chip demand continues to outpace supply across the industry. Nvidia's H100 and H200 GPUs have faced extended lead times for over two years, and cloud providers are increasingly investing in in-house silicon to control costs, availability, and strategic positioning. A successful Maia 300 launch could give Microsoft a meaningful edge in attracting AI-first customers to Azure, particularly if it offers competitive performance at a lower cost per token than Nvidia's hardware.

Microsoft declined to comment beyond Wall's statement. TSMC did not respond to requests for comment. The Maia 300 launch date has not been officially confirmed, but multiple sources cited by The Information point to September as the target window, as first reported by Inside AI.