Accelerating AI model production at Hexagon with Amazon SageMaker HyperPod
Industrial giant slashes model training time from months to days using AWS's resilient Blackwell GPU clusters.
Hexagon, the $5.4B industrial measurement giant, has dramatically accelerated its AI model production pipeline through a strategic partnership with Amazon Web Services. By implementing Amazon SageMaker HyperPod, Hexagon now trains specialized computer vision models for 3D point cloud processing in days rather than months—representing a 10x speedup in development cycles. The infrastructure leverages NVIDIA's latest Blackwell architecture GPUs (specifically P6-B200 and P6e-GB200 instances) with Elastic Fabric Adapter networking for optimal multi-node communication during distributed training.
The technical implementation focuses on resilient architecture where SageMaker HyperPod provides automated node health checks, self-healing capabilities, and checkpoint resumption—critical for training runs that span weeks. This allows Hexagon's AI teams to develop targeted segmentation models that clean sensor noise from LiDAR scans, classify land types in complex environments, and filter moving objects from static scenes. These models process massive point cloud datasets (collections of 3D spatial data points) that Hexagon captures across aerospace, construction, and mining applications.
Contextually, this move represents industrial AI specialization over general-purpose models. Instead of using one massive model, Hexagon builds purpose-built AI for specific tasks like creating digital twins of entire cities from mobile mapping data. The practical implication is faster deployment of AI that improves safety and productivity—from autonomous vehicle navigation to heavy construction site monitoring. By reducing infrastructure management overhead, Hexagon's data scientists can focus on domain-specific innovations rather than cluster operations.
- Hexagon achieved 10x faster AI model training (months to days) using AWS SageMaker HyperPod with NVIDIA Blackwell GPUs
- Infrastructure features self-healing clusters with automated checkpoint resumption for weeks-long training jobs
- Specialized models clean 3D point cloud data for digital twins in construction, mining, and autonomous vehicles
Why It Matters
Accelerates industrial AI deployment for safety-critical applications like autonomous navigation and infrastructure monitoring.