4 domestic AI models, Dynamo deployment recipes released… B200 configuration presented

NVIDIA on Wednesday published Dynamo deployment recipes for four South Korean AI models—SK Telecom's A.X K2, LG AI Research's K-EXAONE 2.0, Upstage's Solar Open 2, and Motif Technologies' Motif-3—specifying B200 GPU counts and aggregated versus disaggregated execution configurations for each. Dynamo is an open-source inference framework that coordinates request routing and inter-GPU communication to serve large models across distributed compute. NVIDIA said the recipes deliver up to 2x higher performance under default settings, though the blog post did not disclose baseline configurations or benchmark workloads. The release ties Korea's sovereign AI models to NVIDIA's B200 hardware and Dynamo stack, giving domestic developers prescriptive serving guidance. What to watch next is whether independent benchmarks confirm the claimed 2x gain and whether the recipes expand to additional models and GPU configurations.