Companies Building the First GPU-Native AI Engine for Planetary Intelligence
Planet and NVIDIA Are Collaborating on the Program
An initiative has been announced that could fundamentally change how satellite imagery is processed, enhanced, and analyzed. Planet and NVIDIA are collaborating on ways to adopt GPUs to accelerate processing imagery and delivering insights. Planet is leveraging NVIDIA’s Blackwell and NVIDIA IGX Thor platforms to transform raw pixels into analysis-ready insights in a few seconds instead of hours. Planet will showcase these technologies at NVIDIA GTC in San Jose.
“By developing a GPU-native AI engine, we are unlocking the full potential of our petabyte-scale archive.”
Will Marshall, Planet
The collaboration focuses on three primary pillars of innovation:
Pipeline Acceleration via NVIDIA CUDA: Planet is working with NVIDIA to demonstrate GPU-accelerating the heavy lifting components of satellite data ground processing—including compositing, orthorectification, and atmospheric compensation entirely onto NVIDIA GPUs. This flexible NVIDIA CUDA-based implementation allows Planet to deploy its processing power anywhere in the stack: in the cloud, at edge ground stations using NVIDIA Blackwell, or directly in space on the NVIDIA IGX Thor platform. This shift delivers orders of magnitude acceleration over legacy CPU-based ground processing.
Generative AI Super-Resolution with NVIDIA CorrDiff: In a first-of-its-kind collaboration, Planet is applying NVIDIA CorrDiff, a generative AI diffusion model, to its PlanetScope imagery. Unlike standard digital filters, this physics-informed generative AI trains on Planet’s vast archive to unlock super-resolution capabilities, delivering visually superior products and extracting new levels of detail from legacy data.
Global Embeddings for Intelligent Search: Planet and NVIDIA are defining new architectures to convert the entire planet’s daily data stream into scalable AI embeddings. By creating a multi-dimensional “vector map” of the Earth, Planet is enabling capabilities for customers to perform intelligent, semantic searches across the entire globe to find patterns and anomalies in near real-time.
“Planet is solving a massive, planetary-scale data problem by imaging the Earth every single day,” said Will Marshall, Co-founder and CEO of Planet. “The traditional methods of batch processing on legacy CPUs can no longer scale with the speed of global change. By developing a GPU-native AI engine, we are unlocking the full potential of our petabyte-scale archive, delivering orders of magnitude speed improvements and physics-informed generative AI products that were previously impossible. It’s cost-efficient, faster, and benefits our customers: from disaster response to security, time saves lives.”
Planet has successfully tested the NVIDIA IGX Jetson Thor module for use in space applications and is integrating the GPU on its next-generation Pelican satellites as well as the upcoming Owl constellation to enable Real Time Insights directly from space.
“The integration of NVIDIA AI infrastructure with Planet’s petabyte-scale imagery is driving a fundamental shift in geospatial intelligence,” said Dion Harris, senior director of HPC and AI infrastructure at NVIDIA. “From the data center to the orbital edge, we’re enabling a new class of AI-powered workflows that transform vast volumes of Earth observation data into actionable planetary insights at the speed of change.”