Sources: Microsoft plans to unveil its next-gen AI chip, the Maia 300, potentially as soon as September, and is in talks with TSMC to make 300K+ chips for 2027
Microsoft is planning to significantly increase production of its internally designed next-generation AI chips next year in hopes …
Context & Ripple Effects
Microsoft’s in-house accelerator program began with Maia 100 testing for Bing and Office AI tools, then faced a reported production delay before the Maia 200 reached Azure’s US Central region in January. The reported Maia 300 capacity discussions suggest Microsoft is trying to turn that initial Maia 200 deployment into a substantially larger supply program.
The scale matters because prior reporting characterized the delayed predecessor as trailing Nvidia’s Blackwell. A larger Maia 300 run would therefore be a test of whether Microsoft can make its own silicon a meaningful part of Azure’s AI compute mix rather than a limited deployment.
First-order effects
- Microsoft is preparing the Maia 300’s introduction while seeking more than 300,000 units of 2027 manufacturing capacity from TSMC.
- TSMC gains a prospective large-volume customer commitment tied to a third successive Microsoft AI-accelerator generation.
Second-order effects
- A larger Maia 300 production run would give Microsoft a bigger internal-compute base to deploy alongside other accelerators, following Maia 200’s initial Azure rollout.
- It would increase pressure on Microsoft’s internal silicon roadmap to close the gap highlighted by the earlier report that its delayed chip would underperform Nvidia’s Blackwell.
Third-order effects
- If the discussions result in production, custom AI chips are becoming a multiyear capacity-planning issue for cloud providers and foundries, not simply an internal product-development effort.
- Successive Maia generations would make Microsoft’s AI infrastructure more heterogeneous, with deployment economics increasingly shaped by its ability to secure foundry output for its own designs.
The trend: Cloud platforms are pairing AI-infrastructure spending with larger, longer-horizon commitments to proprietary accelerator supply.