Case Study: LGND scales geospatial AI to 160 GPUs with Union.ai

LGND scales to 160 GPUs with Union.ai

LGND, a geospatial AI company on a mission to make Earth data searchable, faced significant engineering challenges in scaling its operations. They needed to scale distributed model training from a single node to dozens of GPUs, eliminate latency from Kubernetes cold-starts, automate ETL pipelines, and maintain full infrastructure ownership on their own AWS account without the high costs of a managed service. To overcome this, they turned to vendor Union.ai and its enterprise Flyte platform.

By implementing Union.ai, LGND deployed a Python-native orchestration solution that ran entirely within their own cloud environment. The platform's Reusable Containers eliminated cold-start latency, while its integration with Ray enabled them to scale their core training workload to a 40-node, 160-GPU cluster without changing their code. This resulted in a 111.5% growth in compute usage and a 4x expansion of their GPU footprint. Union.ai provided full cost transparency and partnership-level support, allowing LGND to successfully launch their public API.

LGND