What challenges did Hexagon face before AI implementation?

Hexagon needed to scale and accelerate pretraining of segmentation models for AI point-cloud workflows while maintaining high-performance training pipelines and reducing time-to-market for specialized AI models. The company also needed scalable compute resources, access to the latest GPUs, and more streamlined training operations.

What was the impact of the AI solution?

Hexagon cut training time from about 80 days on-premises to about 4 days on AWS. The company achieved its first training deployment within hours. Larger batch sizes improved training throughput and model accuracy. Hexagon reported a 95% reduction in training time.

Source

Evidence: Low35/100

Hexagon accelerates AI model production with Amazon SageMaker HyperPod

Use case typeAI model trainingUpdated Feb 23, 2026

Hexagon, the global leader in measurement technologies, collaborated with Amazon Web Services to scale AI model production for point-cloud workflows. The company built a managed training environment to pretrain state-of-the-art segmentation models for built-environment and geospatial use cases, with an integrated data pipeline, compute cluster management, and MLOps monitoring stack.

Organization: Hexagon
Industry: Manufacturing
Location: Switzerland
Published: February 2026

Reported outcomes

−95%

training timeTime & speed

−95%training time

Strategic outcomes

Speed & agilityFirst training deployment within hoursOther strategic outcomeImproved model accuracy through larger batch sizesSpeed & agilityAccelerated specialized model development

Catalog median for time & speed deployments: −50% across 312 reported metrics. Compare benchmarks →

Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Training time: 95% decrease

AWS Machine Learning BlogFeb 23, 2026Blog postExplicit claimLow evidence strength

Hexagon reduced their training time from 80 days on-premises for a given network and configuration to approximately 4 days on AWS

Normalized claim

Training time: 95% decrease

AWS Machine Learning BlogFeb 23, 2026Blog postExplicit claimLow evidence strength

Hexagon’s collaboration with Amazon Web Services delivered a remarkable 95% reduction in training time through Amazon SageMaker HyperPod

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Hexagon
Provider: AWS
Maturity: Unknown
Linked source: AWS Machine Learning Blog

No explicit deployment-stage evidence found.

Customer identity supportedSource describes one deploymentMaturity evidence evaluated

Primary read

Use case focus

Showing 2 of 2

1AI model training
2Training infrastructure modernization

Hexagon needed to scale and accelerate pretraining of segmentation models for AI point-cloud workflows while maintaining high-performance training pipelines and reducing time-to-market for specialized AI models.
The company also needed scalable compute resources, access to the latest GPUs, and more streamlined training operations.

Hexagon implemented Amazon SageMaker HyperPod as a managed training environment for distributed model training.
The architecture used Amazon S3 for training data storage, Amazon FSx for Lustre with data repository association for high-throughput data streaming and checkpoint export, Amazon EC2 GPU instances for training, Amazon SageMaker Training Plans for predictable GPU capacity, Amazon Managed Service for Prometheus and Amazon Managed Grafana for observability, and MLflow on Amazon SageMaker AI for experiment tracking.
The HyperPod cluster included self-healing, automated job resumption, EFA-backed networking, lifecycle scripts, and integrated monitoring for cluster health and performance.

Technologies

Amazon SageMaker HyperPod Amazon S3 Amazon FSx for Lustre Amazon EC2 Amazon SageMaker Training Plans Amazon Managed Service for Prometheus Amazon Managed Grafana MLflow on Amazon SageMaker AI

Hexagon cut training time from about 80 days on-premises to about 4 days on AWS.
The company achieved its first training deployment within hours.
Larger batch sizes improved training throughput and model accuracy.
Hexagon reported a 95% reduction in training time.

Architecture

Managed training environment on Amazon SageMaker HyperPod with self-healing cluster management, EFA-backed distributed networking, Amazon S3 and Amazon FSx for Lustre data pipeline, Amazon EC2 GPU instances, SageMaker Training Plans, Amazon Managed Service for Prometheus, Amazon Managed Grafana, and MLflow on Amazon SageMaker AI.

Sources & evidence1

Evidence: Low35/100Evidence strength