NVIDIA DGX Spark 64GB Launches at $4,999 for Local AI
NVIDIA's DGX Spark 64GB arrives October 23 for $4,999, running 100 billion parameter AI models locally and clustering two units for larger AI workloads.
Summary
NVIDIA will launch DGX Spark 64GB on Friday, October 23, 2026, exclusively through Acer, ASUS, Dell, Gigabyte, HP and MSI, starting at $4,999. It retains the 128GB model’s GB10 Grace Blackwell Superchip, DGX OS, unified memory, ConnectX-7 networking and CUDA accelerated AI stack, supporting private, cloud independent operation of models with up to 100 billion parameters. NVIDIA Agent Toolkit, CUDA-X AI libraries, Nemotron, llama.cpp, Ollama, vLLM, LM Studio and PyTorch with CUDA are supported; Blender plans a prebuilt installer soon.
Two units connected directly by QSFP over 200 GbE pool 128GB, double memory bandwidth and support models with up to 200 billion parameters. NVIDIA measured up to 1.7 times single system performance in a Qwen3.8 27B test. NVIDIA Sync Cluster Assistant detects and validates connected systems, configures ConnectX-7 and routes workloads without requiring software reconfiguration.
Due at the end of October, NVIDIA Sync Model Launcher will download and launch Qwen3.8 27B on one system or a cluster, make it accessible from laptops and configure OpenCode for browser based coding. Target workloads include persistent coding and research agents, inference for language and image applications, fine tuning, data science, longer contexts and concurrent agent requests.
Positives
- $4,999 buys local support for models with up to 100 billion parameters using the same GB10 chip and software stack as the 128GB model.
- 128GB of pooled memory lets two connected systems support models with up to 200 billion parameters.
- 1.7 times single system performance was achieved by two units in NVIDIA’s Qwen3.8 27B test.
- 200 GbE ConnectX-7 networking and Sync Cluster Assistant automate clustering without software environment reconfiguration.
- October’s Sync Model Launcher will simplify Qwen3.8 27B deployment and connect OpenCode for browser based coding.
Risks & concerns
- $4,999 remains a substantial entry price for developers, researchers and AI enthusiasts.
- 64GB limits one system to models with up to 100 billion parameters, requiring a second unit for larger models.
- 1.7 times performance in NVIDIA’s test falls short of doubling compute performance, despite doubled memory bandwidth.
- Blender’s installer and NVIDIA Sync Model Launcher are not available at launch, with both promised later.