Tomorrow.io

Building an "AI-native" satellite constellation (DeepSky) equipped with microwave sounders — they’re not just ingesting third-party satellite feeds but creating a proprietary space sensor layer that feeds their ML stack, which is a materially different data source than most weather APIs that rely on public satellites and third-party models.

Exposing derived dynamical meteorology fields as premium API layers (CAPE, CIN, vertical velocity @500hPa, vorticity @700hPa, divergence, precipitable water, integrated water vapor) — shipping these higher-order, altitude-specific diagnostics as productized endpoints implies a pipeline that does model reanalysis/diagnostic computation, vertical interpolation and quality control at scale.

Fine-grained altitude support across APIs (Aviation and Low-Level altitudes) plus explicit multi-altitude temperature/wind/humidity/dewpoint endpoints — they’re treating vertical profile resolution as a first-class product, which requires continuous vertical remapping across many models and observations and consistent API semantics.

Operational alert semantics expressed in the API (lightning alerts with multi-buffer radius arrays, min/max bounds, first-strike-only logic and an all-clear TTL) — instead of a generic event stream they codify complex stateful alert rules in the endpoint, implying low-latency event processing, strike de-duplication, and per-subscription state management.

Productizing operational indices like Wet Road Index and location-area limits (polygon/polyline area caps as premium features) — they monetize spatial scale and domain-specific indices, which requires scalable geospatial tiling, surface modeling (e.g., road wetness), and access-control tied to spatial footprint.