NVIDIA says DGX Spark now supports up to eight units in manual configuration, positioning clusters as shared model backends for small teams running concurrent development sessions. This could let teams share local inference rather than provision separate model instances for each developer. https://blogs.nvidia.com/blog/local-ai-gtc-rtx-dgx-spark-station-sync/
NVIDIA says its Sync Model Launcher is due later this month, while DGX Station for Windows is expected from manufacturing partners in December. Those remain planned deliveries, not confirmed availability; eight-unit clustering currently requires manual configuration. https://blogs.nvidia.com/blog/local-ai-gtc-rtx-dgx-spark-station-sync/
NVIDIA’s shared-backend positioning does not establish performance for a particular team’s workloads. Concurrent-session latency, deployment costs and delivery dates remain unverified, so the practical capacity and rollout timing are still uncertain. https://blogs.nvidia.com/blog/local-ai-gtc-rtx-dgx-spark-station-sync/