Jensen also likes Chinese models. I can sympathize with him
The sympathy is practical. Big open models like DeepSeek V4-Pro demand around 862 GB of GPU memory, and NVIDIA ships day-zero RTX support for Qwen3.8-27B.