NVIDIA pushes DGX Spark
NVIDIA is promoting DGX Spark and desktop DGX options to bring high‑density dev compute to institutions as an on‑prem alternative to cloud GPU shortages. The move signals a push to lock in dev workflows earlier in the stack. (highereddive.com)
NVIDIA first unveiled DGX Spark at GTC on March 18, 2025 and said shipments to partners began in October 2025, with a formal shipping announcement dated Oct. 13, 2025. (edge-ai-vision.com) The compact system is built around the GB10 Grace Blackwell Superchip and NVIDIA says it delivers up to 1 petaFLOP of FP4 AI compute with a Grace Arm CPU and unified memory in the desktop form factor. (resource.naddod.com) Published hands‑on reviews and vendor coverage have reported retail price points in the roughly $3,999–$4,699 range depending on configuration and reseller availability. (digitaltrends.com) DGX Spark runs a DGX OS variant with the full NVIDIA AI stack and NVIDIA has introduced NemoClaw and an open‑source/local‑first software push to let developers run agents and community models on premises. (moorinsightsstrategy.com) NVIDIA positioned OEM partners including Acer, ASUS, Dell, GIGABYTE, HPI, Lenovo and MSI to build and resell DGX Spark and DGX Station systems to widen desktop availability. (investor.nvidia.com) Recent firmware and software updates added multi‑node linking—expanding DGX Spark beyond its original two‑node design to support up to four linked units and larger shared RAM configurations. (guru3d.com) NVIDIA’s marketing and early coverage frame DGX Spark as a developer node capable of prototyping models in the 100‑billion‑parameter class locally, with the larger DGX Station aimed at scaling toward trillion‑parameter model work. (techpowerup.com)
