From Planetary Computer to TerraByte: A New Geospatial AI Platform
TerraByte AI, a stealthy startup co-founded by Microsoft veterans Rishi Madhok and Fuxun Yu, is emerging with an ambitious vision: to make satellite data analysis as searchable and intuitive as the web. The company’s Earth Search Engine applies a geospatial AI platform to streams of real-time satellite imagery, automatically identifying assets such as cargo ships, warehouses, solar farms, mining sites, and even areas of deforestation. Instead of relying on slow, manual annotation, TerraByte uses self-supervised learning to understand pixels directly, building a foundation model that can be queried in natural language. That background is no coincidence. Madhok previously led geospatial AI initiatives for Microsoft’s Planetary Computer, while Yu headed the Geospatial Foundational Model project. Together, they are translating that experience into an enterprise-ready layer of AI infrastructure designed to sit underneath the next generation of location intelligence applications.
Foundation Models Unlock Scalable Satellite Data Analysis
At the core of TerraByte’s approach is a multi-sensor foundation model built to power high-volume satellite data analysis. While many existing tools focus on single-sensor data, TerraByte claims to be the first model layer to natively fuse optical imagery, synthetic aperture radar, thermal, and hyperspectral data in one system. This fusion allows the platform to detect a wide variety of features and patterns with a single model, from monitoring container ships entering port to spotting open-pit mines or tracking deforestation. Because the model is trained using self-supervised techniques rather than manual labeling, it can scale with the rapidly growing volume of Earth observation data. Users interact through natural-language queries, turning complex geospatial questions into AI data insights—for example, surfacing more than 2,000 relevant locations that can then be filtered by asset type, risk condition, or proximity.
Enterprise Geospatial Intelligence for Risk, Operations and Infrastructure
TerraByte is positioning its Earth Search Engine squarely at enterprise geospatial intelligence use cases. Early examples include risk assessment for insurers, who can ask the platform to identify high-risk properties based on vegetation encroachment around power lines or infrastructure near wildfire-prone zones. Utilities and first responders can query the system during disasters, such as wildfires, to instantly locate neighborhoods and power-line segments within a specified radius of an active event, enabling faster, more targeted response. Commercial operators can use the same geospatial AI platform to track logistics and infrastructure, such as parking lots near highways that lack EV charging, or cargo ship movements through choke points like major shipping straits. By turning satellite imagery into directly queryable AI data insights, TerraByte aims to help organizations act on geospatial signals that have historically remained siloed, slow, or too complex to operationalize at scale.
Edge Deployment and a Growing Ecosystem Around Satellite Intelligence
Beyond ground-based analytics, TerraByte is designing its models to run on the edge, including directly on satellites. By processing imagery in orbit and filtering what gets downlinked, the platform could reduce both latency and bandwidth costs while still delivering timely intelligence for time-sensitive operations. This edge capability is particularly important for scenarios like natural disasters, where every minute counts. TerraByte has secured pre-seed funding from Ascend, PSL Ventures and angel investors to accelerate development of its AI infrastructure layer and expand its enterprise offerings, initially via large B2B licenses before moving to subscription models. Rather than competing head-on with satellite builders or in-orbit compute providers, the company frames those players as partners. In a landscape that includes tools like Google Earth Engine and BlackSky Spectra, TerraByte is focused on becoming the foundational API for enterprise geospatial intelligence, abstracting away raw pixels in favor of ready-to-use, high-level insights.