Export bans fueled local LLM renaissance; DeepSeek V4.1 Flash 1017 runs on 4 DGX Spark at home

Zeng Jiajun 曾嘉俊 @zengjiajun_ethOriginally posted 2026.10.27

Open original on X

it's funny that exactly because US government's export ban on chips, force chinese labs to make it efficient to train and serve on every single card they can access to. which lead to the renaissance of local llm. take deepseek v4.1 flash 1017 for example, the model they publish is exactly the model they serve via api. and now you can run it easily with 4 dgx spark at home.