Build Daily
Tinley Park · September 9, 2026NVIDIA didn't want me to do this
The creator builds a 1TB VRAM cluster by wiring eight DGX Sparks together, working around NVIDIA's lack of official guidance for scaling beyond two units. The video covers the hardware setup, networking, and the process of creating a large-scale local AI infrastructure.
The outline
Every point the video makes, in the order it makes them.
- 01
Cluster Architecture
The video details the process of connecting eight DGX Sparks to create a single 1TB VRAM cluster.
- 02
NVIDIA Limitations
It addresses the lack of official documentation from NVIDIA for scaling DGX Spark setups beyond two units.
- 03
High-Speed Networking
The build utilizes 400Gbps switches to facilitate the high-bandwidth communication required for the cluster.
- 04
Hardware Components
The creator shares the specific gear used, including the DGX Sparks, switches, and precision tools for the build.
- 05
Local AI Infrastructure
The project aims to create a powerful local environment for running large-scale AI models and LLMs.
- 06
Build-in-Public Process
The video follows the creator's journey of troubleshooting and assembling the custom hardware setup.
Wiring eight DGX Sparks into a 1TB VRAM cluster without NVIDIA's official guidance shows solo builders can assemble local AI infrastructure by combining hardware, networking, and open models. This matters because it lowers the barrier to testing large local models without relying on vendor-supported scaling paths.