Foxconn has deployed an AI factory utilizing an Nvidia DGX SuperPOD cluster based on the Blackwell architecture. The facility specifically employs Nvidia DGX GB200 systems to train proprietary AI models and optimize internal production workflows.
Performance metrics indicate that the GB200 setup trains 120 billion parameter models 2.48 times faster than previous H100-based systems. The cluster achieved a peak throughput of 4,900 TFLOPS and processed 1 trillion tokens in 12 days.
Operational improvements include a 25% reduction in production scheduling time and a 50% decrease in engineering workloads for injection molding. Additionally, the company reports a defect correction rate exceeding 90% on its SMT production lines.
Foxconn continues to manufacture AI server racks for external clients while collaborating with Nvidia on large-scale supercomputer projects in Taiwan. The company allocates between $2 billion and $3 billion annually to AI infrastructure.
