High-density GPU colocation

Bring the GPUs.We bring theinfrastructure

Assess power, direct-to-chip cooling, network, and operations for owned HGX hardware without forcing it into a generic colocation footprint.

Dark render of liquid-cooled GPU infrastructure
645 kW
Reference load
Redundant
CDUs
No site
Water requirement
Owned hardware
Fit assessment
Liquid cooled
Direct to chip
400G
Network reference
Managed
Remote operations
Fit

Start with a hardware fit review

Technical fit is the reason to engage. The intake covers your hardware, the site envelope, and how the handoff is verified.

Rack and node

Map hardware, dimensions, and service access against the deployment envelope.

Power and cooling

Model the operating envelope and direct-to-chip cooling interfaces for your nodes.

Network and storage

Confirm the required topology and configuration before anything ships.

Reference

Pod 32 as the reference envelope

Pod 32 is the reference infrastructure envelope, clearly separate from your final buyer-specific configuration. Your fit review defines what applies.

Reference compute
32 Supermicro HGX B300 nodes
Reference load
645 kW operating
Cooling
Direct-to-chip liquid, redundant CDUs
Site water
No site water requirement
Fabric
Rail-optimized 400G Ethernet
Fabric throughput
Configurable 3.2 or 6.4 Tbps per node
Storage
983 TB raw NVMe per module
Tenancy
Single-tenant deployment

Reference values describe Pacific's Pod 32 module, not a guarantee that any specific buyer hardware fits. The engineering intake confirms your envelope.

Acceptance

Acceptance is a handoff, not a brochure

Evidence gates make the handoff criteria explicit before your hardware moves.

Factory

Component and integration evidence recorded through the 24-stage test program.

Site

Utilities, interfaces, and readiness verified against the deployment plan.

Fabric

Measured NCCL criteria at defined gates where the network fabric is in scope.

Operations

Fit, ownership, operations, exit

Pacific operates the infrastructure remotely with contracted regional field service. Who does what for your hardware is defined during discovery - service terms stay in the contract.

Who operates?

Pacific and buyer responsibilities are defined explicitly before deployment, not assumed.

What fits?

The engineering intake confirms the envelope. No implied universal fit.

What changes?

Configuration changes after handoff follow acceptance control, with evidence.

Get the Pod 32 spec
Book

Tell us what you own

Three quick questions - role, hardware, timeline - then the calendar opens. No model numbers or free text required before booking.

1 / 3

Your role

Colocation call

Answer the three questions and the calendar opens here. Prefer to skip ahead? Book directly on Cal.com.

FAQ

Before you book

What hardware fits?

An engineering intake maps your rack and node dimensions, power envelope, cooling interfaces, network topology, and service access against the infrastructure. Fit is confirmed by that review, not assumed.

Who operates what?

Pacific runs the infrastructure through remote operations and contracted regional field service. Responsibility for your hardware, software stack, and maintenance access is defined explicitly during discovery.

Is liquid cooling required?

The reference environment is direct-to-chip liquid cooling with redundant CDUs and no site water requirement. The fit review determines how your hardware's cooling profile maps onto it.

How is acceptance handled?

Handoff follows factory, site, and integration evidence gates, including measured NCCL fabric criteria where the network fabric is in scope for your deployment.

What are the commercial terms?

Terms are discussed privately on a qualified colocation call. Pacific does not publish rates.

Check fit for your GPU hardware

Book a colocation call