NVIDIA's Arm-based Vera CPU, designed for AI inference, achieved 1.8x higher performance in agentic execution and 1.5x in coding workflows compared to x86 competitors in loaded CPU workloads. The chip features an 88-core monolithic design with SMT support for 176 threads, delivering 1.2 TB/s LPDDR5X RAM bandwidth and 3.4 TB/s core-to-core bandwidth.