Tractian builds hundreds of AI models faster, at lower cost using OCI

Overview

Tractian, which builds Physical AI solutions for industrial asset monitoring, uses Oracle Cloud Infrastructure bare metal with NVIDIA H100 and GB300 GPUs to train and deploy custom AI models that turn sensor data into real-time maintenance decisions across more than 2,000 plants and 200,000 monitored assets. Tractian cut AI training costs by 20%, reduced training times by 35%, cut inference costs by 15%, reduced inference latency by 50%, and estimates it prevents a customer machine failure roughly every 15 minutes across its installed base.

The challenge

Tractian's sophisticated AI models, which require terabytes of sensor data per machine, can take months to train. To accelerate training the company needed high performance, bare metal cloud infrastructure with high-throughput storage, plus low latency and predictable costs at scale for its inferencing needs.

The solution

Tractian selected Oracle Cloud Infrastructure (OCI), leveraging NVIDIA GPU-powered infrastructure and OCI bare metal with NVIDIA H100 and GB300 GPUs to train and deploy AI models at scale, processing massive volumes of sensor, operational, and industrial data in real time.

Reported business value

Tractian cut training costs by 20% and reduced training times by an average 35%. In production, inference costs were reduced by 15% and latency by 50%. The company estimates its systems prevent a customer machine failure roughly every 15 minutes across its installed base.