Project
A GPU compute routing engine that queries cloud marketplaces and scores available instances by price and carbon intensity — helping AI workloads find the cheapest or the greenest compute, depending on what you optimize for.
Modular provider architecture — each marketplace integration is an independent module exposing a normalized record schema, making it straightforward to add new providers without touching the scorer.
Queries available GPU instances filtered to rentable, un-rented nodes and ordered by hourly price — surfacing GPU model, reliability score, and geographic location for each result.
Fetches real-time marginal carbon intensity for datacenter locations, enabling the engine to score and rank compute options by their actual grid emissions — not just location-level averages.
A configurable weight parameter from 0.0 (pure price) to 1.0 (pure carbon) blends price rank and carbon index into a single score — letting users dial between cost minimization and emissions minimization.
Normalized results across cloud GPU marketplaces with price, reliability, and geolocation data — presented in a consistent schema regardless of the underlying provider's API structure.
Integrates real-time marginal grid emissions to surface not just the cheapest compute, but the cleanest — a meaningful distinction as AI infrastructure's environmental footprint grows.
A single parameter controls the tradeoff between cost and carbon — useful for organizations with sustainability commitments that need to balance budget constraints against emissions targets.
Questions or feedback?