DGX Spark Live: Smart Routing for Hybrid AI

AI tokenomics makes every routing decision a cost decision — and the right answer changes from one job to the next. In this livestream, we demo smart routing for hybrid AI running live on a Lenovo ThinkStation PGX. A lightweight judge model evaluates each incoming request and sends it to the right destination, choosing between local models and the cloud based on what the job actually needs. Three models run concurrently on one ThinkStation PGX, with a live dashboard showing throughput per model and the cost profile of each decision as it happens. A practical look at how hybrid AI matches workloads to the right resource, automatically. Attendees will learn: - How NVFP4 quantization enables three models to run concurrently on a Lenovo ThinkStation PGX, expanding local inference capacity. - How to compare token throughput and cost when sizing a hybrid AI deployment. - How a judge model routes requests between local and cloud inference in real time. - How to use those routing decisions and performance measurements to estimate the work a desktop-class system can handle.