What we know

A lawyer has assembled a private AI server for $9,450 using 12 used Nvidia V100 GPUs. This setup is now running every DeepSeek V4.1 Flash expert model at 4-bit NVFP4 precision. The server uses 12 second-hand Nvidia V100 GPUs. It runs DeepSeek V4.1 Flash experts at 4-bit NVFP4 precision.

Why it matters

DeepSeek V4.1 Flash is a large language model that can be operated locally if the hardware is sufficient. Nvidia V100 GPUs, although older, still offer strong performance for AI tasks when combined in quantity. Using 4-bit precision like NVFP4 lowers memory demands, enabling large models to run on more affordable or used hardware.

What is still unknown

No independent verification of model accuracy or performance has been reported. Details on supported languages, software updates, or plans for wider availability remain unclear.