Earthmover is betting the next big data platform won’t be found in ad clicks or financial logs, but in the sky. The startup is aiming to be the Snowflake of weather and geospatial data — transforming vast, unwieldy Earth observations and forecasts into governed, cloud-native data sets that analysts, quants and operations teams can explore and share within seconds.
The pitch is simple: Weather moves markets and infrastructure, but the data feeding those predictions are still scattered across file formats, archives and model grids. Earthmover, launched by climate scientists and open-source veterans Ryan Abernathey and Joe Hamman, is creating a single platform for array-based data — the rasters the be-geospatial’d love, or tensors if you want to get all fancy like the AI kids — so that teams can store once, compute anywhere, and collaborate without wrestling with petabytes.
From the climate scale to daily operational choices
Earthmover started in climate tech, and then became more specific about weather — the time scale when decisions are made. Long-term climate indicators form the strategy, but daily and sub-intraday signals drive power dispatch, crop inputs, insurance exposure, and transport routing. In other words: weather is where probability meets P&L.
That shift reflects industry demand. Utilities and renewables developers are demanding sub-hourly wind and irradiance forecasts; insurers consume fire weather, drought, and convective risk grids; commodity traders monitor jet stream shifts and heating-degree days. Outfits rely on reanalysis products, such as ERA5 from the European Centre for Medium-Range Weather Forecasts, to essentially cold-cock their opponents so they can pivot to high-resolution forecasts and fight.
Cloud-native, petabyte-scale rasters for fast analysis
Weather and Earth observation data are huge. Well over 100 petabytes are under the purview of NASA’s Earth Observing System Data and Information System. NOAA’s geostationary satellites produce terabytes each day, and national centers output new model runs every couple of hours. Conventional file-based workflows crumble under that pace.
The core of Earthmover is an array-native data model for reading and writing chunked, columnar-like multi-dimensional grids. It builds on the open-source tools that its founders helped popularize, such as Xarray for labeled arrays, Pangeo for scalable science in the cloud, and Icechunk/Zarr for cloud-optimized storage. The upshot: Users can select time ranges, variables and bounding boxes without having to drag entire files around, and they can push compute to data on AWS, Google Cloud or Azure, or on-premises object stores.
Typical users handle tens to hundreds of terabytes, and some up to the petabyte level as they merge model output, satellite imagery and sensor feeds. Consistent, cloud-native layout enables those teams to spin up dashboards, batch runs or machine learning pipelines without bespoke ETL for each new dataset.
Open source as practical vendor risk insurance
The portability also minimizes data gravity costs. Teams can keep archives closer to existing analytics stacks, stand up governed sharing policies and move only the slices they need for a query or model run. In regulated sectors, that hybrid approach opens up modern cloud workflows while not giving up data residency or provenance controls.
Who’s buying: insurance, energy and trading
Early adopters shine a light on the breadth of the platform. Kettle, an insurance startup, uses Earthmover to analyze wildfire risk by fusing meteorology with fuels, topography and historical burn scars. The Germany-based energy giant RWE applies weather-induced forecasting to stack the ladder of renewable generation and financial positions. Throughout the industry, grid operators monitor ramp risk from fast-moving weather fronts, and commodity desks are watching ensemble guidance to price uncertainty.
In all three cases, the unifier is speed to insight. Instead of having to spin up Python scripts for each new question, they seek shareable maps, time series and anomaly layers rendered on demand, with the ability to drop back into code when necessary. The Snowflake-likeness of rasters — query, share, govern — checks that box.
Competition and the role Earthmover targets
It is a crowded but broken landscape. The Weather Company (IBM), Tomorrow.io, and Meteomatics offer forecasts as services; planetary-scale catalogs like Google Earth Engine and Microsoft’s Planetary Computer dominate imagery and vector analytics; data clouds such as Snowflake are increasingly looking to support geospatial joins through partners like Carto. What is unusual, however, is an enterprise-grade array-native layer built for operational weather models and reanalyses alongside satellite rasters with governed sharing across clouds.
So if Earthmover standardizes access to, say, model grids from NOAA, ECMWF and national meteorological services; blends them with EO imagery; and exposes a marketplace-style sharing model, then it’s fulfilling an actual gap between scientific archives and businesses.
The path to a secure, scalable data cloud for weather
To deserve the “Snowflake of weather” shingle, a vendor needs to get fucked three ways from Sunday and emerge safe and sound through it all: seamless data onboarding at scale, governed collaboration, elastic compute at the point of data. Look for deeper support of real-time streams, ensemble-aware APIs, GPU acceleration for deep learning and cross-tenant data sharing while maintaining lineage and access controls.
Challenges remain. Standardization across models and grids is difficult, and data egress costs can blow a hole in the budget if the architectures are not tuned. But the macro tailwinds are blowing hard. The stakes are getting higher for all types of forecasts as extreme conditions become more intense, according to the World Meteorological Organization. A platform that democratizes petabyte-class weather and geospatial data, every bit as queryable as a table, could become indispensable infrastructure.
