Overview

Harbor Compute designs and runs physical AI infrastructure in three tiers: Core Harbor for large-scale distributed training, Middle Harbor for regional GPU capacity, and HarborOne for inference at qualified sites. It provides GPU compute, private networking, a CDN, and AI delivery networks within purpose-built data centers, using high-density power, advanced cooling, and interconnects like NVLink, NVSwitch, and InfiniBand. The leadership team combines deep experience from OpenAI’s infrastructure program with veterans from Apple, Meta, Microsoft, Amazon, and Tesla, backed by investors such as a16z Speedrun and NVIDIA Inception. Its goal is to deliver reliable, scalable, and responsible AI infrastructure that supports massive training, low-latency regional inference, and seamless model hosting and workflows for enterprises and labs.