Member-only story
Learn how Xoople scales Python AI on Azure with managed Ray on AKS: a two-plane architecture, hybrid CPU and GPU pipelines, and high GPU use.
This article walks through how Xoople runs large-scale Earth observation AI in Python on Azure, using managed Ray through Anyscale on top of Azure Kubernetes Service.
You will see the reference architecture that keeps a managed control plane separate from the compute that stays inside your own subscription, how a hybrid CPU and GPU pipeline turns hundreds of megabytes of multispectral satellite imagery into land use classifications, and why strong GPU utilization comes down to the handshake between data preparation and the accelerators.
By the end, you will have a clear pattern for scaling Python AI workloads on Azure, plus a starting point for building it yourself.
Reference here.
What Xoople is building #
Xoople is a Madrid-based company working on what it calls an Earth system of record: a data infrastructure layer built for the era of AI, designed to let models and increasingly agentic systems reason about what happens in the physical world. Founded in 2019, the company has raised around $225 million to date and is building toward its own satellite constellation for…