Foxconn says factory AI cut production-scheduling labor by 25%, while its DGX SuperPOD with DGX GB200 systems produced a Foxconn-measured 2.48x model-training speedup, tying computing performance to factory operations.
NVIDIA's case study says Foxconn measured the gain against its H100 baseline using 120-billion-parameter models. Separately, telemetry from DGX SuperPOD training jobs recorded one trillion tokens over 12 days. The DGX GB200 systems use NVIDIA's Blackwell architecture.
Foxconn attributed a 50% reduction in engineering work for injection-molding changeovers to setpoint optimization. On a surface-mount technology (SMT) line, its root-cause analysis addressed more than 20 major defect root causes, and the correction rate exceeded 90%, the case study said.
