Advantech uses this booth tour to connect NVIDIA Blackwell GPU servers with industrial PC design and edge AI boxes, showing a full pipeline from training large models to deploying them on the factory floor. Marco Zampolli walks through a Blackwell-based GPU workstation for LLMs and video AI, then contrasts it with compact Jetson-powered systems and a smart camera that can both ingest data and run inference directly on the line. https://www.advantech.com/en/products/edge-ai-computer/sub_3dcb0054-ba17-441a-925a-0869f03e4691
The video focuses on how digital human interaction and “AI helpers” are built on top of these platforms. A digital assistant front-end taps into LLMs and enterprise knowledge bases, while Blackwell GPU servers handle training, fine-tuning and heavier multimodal workloads in the data center. The result is a stack that can support chat-like interfaces on top of real-time machine data, video streams and historical production logs rather than just static documents.
From there, the discussion moves to edge AI, where fanless Jetson systems are mounted close to machines for low-latency inference, ruggedized operation and reduced backhaul traffic. These boxes ingest sensor and video data, run optimized models, and close the loop with actuators or MES/SCADA systems. The architecture separates model development and retraining on powerful GPU servers from deterministic inference on compact embedded devices, which is critical for OT environments with strict availability and thermal constraints.
A key highlight is the Jetson-based smart camera that combines optics, lighting and compute in a single housing. Instead of streaming raw video back to a rack PC, the camera hosts the model, performs on-device pre-processing, and can be quickly retrained with new image samples. This enables very fast deployment cycles for computer vision use cases such as traceability, defect detection and assembly verification, where engineers can iterate models without rewriting large amounts of traditional vision code.
Filmed at SPS 2025 in Nuremberg, the conversation underlines how Advantech positions itself across the full AI lifecycle: Blackwell GPU servers for training and heavy LLM/video workloads, configurable edge IPCs for inference at the machine, and integrated AI cameras for vision-centric tasks. Together with open-source models and customer-specific fine-tuning, the portfolio targets manufacturers who want to shorten go-to-market time for AI projects and keep both data and inference under tight industrial control. ([Advantech][1])
Advantech from Blackwell GPU workstation to Jetson Orin smart camera inference
Advantech Blackwell IPC and Jetson Orin smart camera for traceability and visual inspection



