Microsoft is working with TSMC to manufacture custom-designed chips, extending a partnership that stretches back more than two decades to the original Xbox era. The collaboration centers on Microsoft’s push to build its own AI accelerators rather than buying them off the shelf from Nvidia or AMD.
What Microsoft is building
The centerpiece of Microsoft’s custom chip effort is the Maia line of AI accelerators. The latest iteration, the Maia 200, packs over 140 billion transistors onto a single chip, manufactured using TSMC’s cutting-edge 3nm process technology.
In terms of raw capability, the Maia 200 delivers more than 10 petaFLOPS of FP4 performance. That’s 10 quadrillion floating-point operations per second at four-bit precision, a format specifically tuned for AI inference workloads. These are the tasks that power everything from Copilot responses to Azure AI services running behind the scenes for enterprise customers.
Next-generation Maia variants are expected to arrive around 2026, which means Microsoft is already well into the design cycle for chips that won’t see production for months or longer.
