Microsoft Launches Maia 200 Inference Accelerator, Opens SDK to External Users

Big tech rolling its own silicon echoes IBM’s 1964 System/360 (first fully integrated hardware-software mainframe) and Apple’s 2020 M1 pivot—moments when control of the compute stack shifted industry power. Maia 200 fits the long arc toward vertical integration and precision-tuned, low-power accelerators as Moore’s Law flattens and energy, not transistors, becomes the binding constraint. By using TSMC’s N3 and Ethernet-based scale-up, Microsoft is betting that open standards plus proprietary design will out-maneuver Nvidia’s closed NVLink empire, much as Ethernet toppled proprietary minis in the 1980s. Whether the chip itself prevails is less important than the signal: hyperscalers are internalizing critical supply chains. On a century timeline, such moves hint at a future where compute capacity—like electricity—becomes a sovereign infrastructure, not a vendor product.