Halcyon Compute
Inference scheduling for multi-model fleets
Builds a scheduling layer that keeps multi-model inference fleets saturated, cutting GPU spend for AI-native platforms.
Portfolio
From pre-seed to Series A, we have led or joined the first round in 64 companies. A public selection is listed below.
Inference scheduling for multi-model fleets
Builds a scheduling layer that keeps multi-model inference fleets saturated, cutting GPU spend for AI-native platforms.
The evaluation layer for production AI
Evaluation and observability infrastructure that lets enterprises ship model updates with measurable confidence.
Autonomous inspection for critical infrastructure
Autonomous aerial and ground inspection for energy grids, rail and heavy industry.
Clinical documentation that closes the loop
Ambient clinical documentation that writes back into the EHR and closes the coding loop.
Autonomous controls for the mid-market CFO
Continuous, autonomous financial controls for mid-market finance teams.
Error-corrected control hardware
Error-corrected control hardware for superconducting qubit systems.
Vector storage built for retrieval at scale
Vector storage engineered for low-latency retrieval at billion-document scale.
Synthetic data for regulated domains
Synthetic data generation for regulated domains where real data cannot leave the perimeter.
Agentic operations for freight forwarders
Agentic operations software for freight forwarders and cross-border logistics.
Specialty insurance priced by machines
Specialty insurance underwriting priced by models rather than committees.
Grid-scale storage chemistry
Long-duration, grid-scale storage chemistry manufactured at commodity cost.
General-purpose manipulation for warehouses
General-purpose manipulation systems for warehouse and light manufacturing.