Wherobots is turning spatial data into a cloud primitive

Wherobots is turning spatial data into a cloud primitive

There is a specific kind of dataset that most cloud analytics engines handle badly: anything where location is the point. A table of addresses is fine. A table of addresses where you need to ask "which of these 2.7 million buildings sit inside a flood zone" is where ordinary SQL falls over, because the question is spatial and the engine treats geometry as an opaque blob.

Wherobots is a company built entirely around that gap. It sells spatial compute as a serverless cloud service, and its pitch is narrow and honest: run spatial queries at planetary scale, up to 5–20x faster than general-purpose cloud engines, without managing the infrastructure. It was founded in 2022 by Mo Sarwat and Jia Yu, the two people who created Apache Sedona, the open-source spatial engine that a lot of the industry quietly already runs on.

Where it came from

The lineage matters because it explains why the company exists. In 2017, Sarwat and Yu released the first version of Sedona under the name GeoSpark, extending big-data systems to handle spatial data. By 2020 it had entered the Apache Software Foundation, supporting both Apache Spark and Apache Flink, with Spatial SQL and Spatial Python as the interfaces. By 2024 it had passed 40 million downloads, growing 200% year on year, and by 2025 past 62 million.

Sarwat and Yu saw that the mainstream analytics engines were adding spatial support in a half-finished way — enough to demo, not enough to run a country on. In June 2022 they left their academic posts to found Wherobots and build first-class spatial support in the cloud. The company raised a $5.5M seed round co-led by Wing Venture Capital and Clear Ventures, then a $21.5M Series A led by Felicis with participation from JetBlue Ventures and P7 Ventures.

The user list is not theoretical. Apache Sedona is used across more than 30,000 organisations, with names like Apple, Amazon, Uber, Nvidia, BMW, Mercedes, Instacart, Zillow, Allstate, SwissRe, and Land O'Lakes. The Overture Maps Foundation — a coalition including Meta, Microsoft, Amazon and TomTom — uses Sedona and Wherobots Cloud for its global mapping work.

What the platform actually is

Three things, layered.

WherobotsDB is the compute engine: planetary-scale spatial compute with more than 300 functions, handling both vector data (points, lines, polygons) and raster data (satellite imagery, elevation grids) in one system. It is compatible with Apache Sedona, so existing Sedona code carries over.

RasterFlow is the Earth-intelligence layer: an inference engine for satellite imagery, now in public preview. This is where the AI sits — running models over pixels at scale to detect crops, rooftops, tree canopy, or flood extent.

A spatial AI coding assistant and a global hub round it out: a coding helper that understands spatial functions, and a shared catalogue of datasets, notebooks, workflows and recipes so you are not starting from a blank cell.

The design choice that matters most is that vector and raster are unified. Most stacks make you join a GIS tool to an imagery pipeline to a database and reconcile them by hand. Wherobots treats both as first-class, which is what makes questions like "which farms are in this satellite scene" a single query instead of a plumbing project.

What you can build with it

This is the useful part, and it is broader than "maps".

Insurance and climate risk. Join a building footprint dataset to peril zones and count exposure. Wherobots' own demo does exactly this: it joins Overture buildings to Colorado parcels and FEMA risk zones and finds 2,172,394 buildings exposed to hail — 78.4% of the state. An insurer can run that per portfolio, per county, in minutes.

Earth observation. Process petabytes of satellite imagery and run model inference to build monitoring products. The US Forest Service reportedly moved from a months-long infrastructure project to fire forecasts updated throughout the day.

Agriculture. Satellite crop monitoring and field-boundary detection at scale, which is what Land O'Lakes does with the underlying Sedona engine.

Telecom and utilities. Process billions of network events, plan 5G rollout, map grid capacity gaps, and assess wildfire risk around infrastructure.

Mobility and maps. Process billions of GPS points into location-intelligence products — the same class of problem Uber, Instacart and Zillow solve with Sedona.

Property and geospatial data engineering. Aarden.ai reported going from a week of babysitting compute jobs to results before lunch, and a 300x speedup on core processing — seven days down to thirty minutes.

The honest caveats

Two things to hold in mind.

First, this is a commercial cloud, not a free tool. Apache Sedona underneath is open source and you can run it yourself, but the convenience — serverless scale, the catalogue, the assistant — is what you pay for. If your data is small, a plain PostGIS instance will do everything you need and cost less.

Second, spatial work is a skill, not a button. Knowing which function to reach for, how to index geometry, and why a naive join over millions of polygons is slow — that is the actual work. Wherobots removes the infrastructure tax, not the thinking.

Why this belongs next to Google's geospatial push

The timing is not a coincidence. On 14 October 2026, Google Research announced a Geospatial Reasoning agent and Earth AI foundation models covering remote sensing, weather, flood hydrology, and population dynamics. Wherobots is the commercial infrastructure layer for the same shift: AI is moving from text on a screen to decisions about the physical world.

For anyone in Malaysia, that is worth watching. Flood modelling, plantation monitoring, urban planning, and disaster response all run on the same primitive — spatial data you can actually query at scale. If you want to see the research side of this, our look at what research.google hands developers is the companion piece.

Sources

• Wherobots — The Context Engine for Physical World Data
• Wherobots — Products
• Wherobots — About us
• Wherobots Raises $21.5M in Series A Funding Led by Felicis
• Felicis's Series A in Wherobots
• Wherobots documentation — Examples