the science data cloud

A Zarr-compatible transactional storage engine, written in Rust. ACID transactions, git-like version history, best-in-class cloud I/O, and zero-copy virtualization of NetCDF, HDF5, GRIB, and TIFF archives. Apache 2.0, serverless, and yours to run anywhere.

For the data that belongs in rows — station observations, event catalogs, training labels, and the tabular results of tensor queries. Iceberg tables live in the same Arraylake catalog as your Icechunk repositories, under the same permissions and in the same buckets, and are served through a standard Iceberg REST catalog, so DuckDB, Spark, Polars, pandas, Trino, and Snowflake read and write them with no Earthmover-specific connector. In public beta.

A multi-cloud catalog for Icechunk repositories and Iceberg tables: single sign-on, teams, and role-based access control; credential vending so users never handle cloud keys; webhooks on every commit; and background optimization that expires old snapshots and garbage-collects unreferenced data.

Subscribe to analysis-ready datasets from trusted providers and they appear as repositories in your organization — versioned, updated incrementally, and queryable immediately, with nothing copied. Providers list free or paid data straight from their own repositories and buckets. More than 60 PB from 22 providers today.

Managed Compute runs next to your data on Zax, which plans the computation for you — read strategy, I/O scheduling, and resource sizing — so you never tune chunk sizes or worker counts. Zax-SQL queries any repository from a PostgreSQL or Arrow Flight SQL client, and tiles, EDR, DAP2, and openEO services serve the same data over open standards. Or bring your own compute: because the data is in open formats, NumPy, Xarray, PyTorch, Dask, Spark, DuckDB, and more read and write it directly.