28 August 2026 · Rob Carroll
Taking one node travelling: what running a sovereign AI stack on DGX Spark actually taught us
One node of a production DGX Spark pair becomes a portable dev box: dual models co-resident, speculative decoding at 33 to 69 tok/s, and a promotion path into client NVIDIA infrastructure.
dgx-sparkgb10sovereign-airocevllmaarch64edge-inference