Sid Sheth, d-Matrix | CASPA CEO Summit 2026
summary
Claims rack-scale is now the unit of AI inference and backs it with a multi-megawatt Corsair XPU cluster built after acquiring GigaIO and Wallaroo to move from cards to racks. Architecture stacks memory directly atop compute to fix capacity and bandwidth limits—memory stacked atop compute—and plugs into the NVIDIA rack ecosystem to avoid rebuilding cooling, power and supply chains. Target is a gigawatt deployed by 2029, with time to first compute as the key customer metric.
AI-written summary · about
