Pacific

Pacific builds factory produced, modular AI data centers and sells access to the compute inside them. The company's thesis is that data center projects wait years for grid power while stalled construction sites sit on unused grid connections, so its pods are placed on those sites and can begin earning within days. The reference module, Pod 32, combines 32 Supermicro HGX B300 nodes, 256 NVIDIA B300 GPUs and rail optimized 400G Ethernet, with 983 TB of raw NVMe storage per module and direct to chip liquid cooling with redundant coolant distribution units. Clusters are single tenant with bare metal access. Pacific offers reserved GPU capacity for training and inference, on premises managed pods for organizations that must keep data in place, including defense and CMMC workloads designed to help reduce the scope of CUI environments, colocation for customers' own hardware, and bridge capacity while a permanent site is completed. The vendor also describes a franchise model in which operators run a Pacific factory. Advanced cooling using diamond wafers is described as active research, not a shipped feature. The audience is AI teams and regulated organizations that need dedicated compute or control over where it sits. Pacific operates the pods remotely with contracted regional field service. No prices are published, and each deployment starts from a conversation about the workload, site and acceptance plan. The site also invites visitors to rent compute through a contact route.