NVIDIA has announced that its DGX Spark platform can now be configured as an 8-way cluster, expanding the previous 4-way limit. The update allows 64 GB configurations to aggregate up to 512 GB of memory and 128 GB configurations to reach 1 TB of pooled memory, enabling larger models and longer context handling. This scaling supports frontier-class models such as GLM 5.3 and the 74B Bolt variant. A new Sync Model Launcher will streamline model deployment by automatically matching containers to system configurations and linking to OpenCode documentation. The October software update also introduces a Wi-Fi Assistant for easier network switching in appliance mode and enhanced Spark Skills for troubleshooting and maintenance. Seven enterprise management integrations, including tools from Cisco, Microsoft, and Perforce, will be added to DGX Station for IT deployment workflows. DGX Station will launch with Windows support in December, targeting manufacturing applications and agent harnesses. New open-source robotics workflows are being released to aid in training manipulation skills and generating synthetic data. NVIDIA Cloud Functions will enter early access on DGX Station, allowing teams to share model endpoints and update models without altering application code. Performance improvements include a 30% speed gain in Qwen 3.8 27B inference on Spark, a 20% boost in GLM 5.3 Flash on Station, and a 40% uplift in Qwen 3.8 Flash-Next on RTX PRO 6000 Blackwell. Additional Day-0 models from Nemotron, DeepSeek, and others are now supported across RTX and DGX systems. Optimized model checkpoints improve RAM efficiency, leaving more memory available for KV cache. These updates position DGX Spark and RTX platforms as scalable solutions for local AI development, particularly for agentic workloads and distributed inference tasks.