Maia 200 inference accelerator

Microsoft's in-house AI accelerator for inference, built on TSMC 3nm, with 216GB of HBM3e memory. Microsoft says it runs Microsoft 365 Copilot, Foundry and OpenAI models, and it is previewing an SDK for outside developers.

Why it matters to the stock. Each chip Microsoft runs in-house reduces the share of its inference bill that goes to Nvidia GPUs. Microsoft claims about 30% better performance per dollar than its latest hardware, which is a vendor claim until it shows in margins.