How to Evaluate an AI Colocation Site

AI deployments put unusual pressure on every layer of a data center. A site that works well for enterprise colocation may not be suitable for a dense training cluster, and a site marketed as 'AI-ready' may describe a plan rather than installed infrastructure. A disciplined evaluation separates what exists from what is intended.

1. Start with capacity status

Before comparing specifications, establish whether the capacity is operating, commissioned, under construction, in development, or still a concept. Each status carries different risk. Ask for the status in writing, and ask what milestones remain before your capacity is energized and accepted.

2. Power: source, topology, and delivery

• What is the power source—utility, on-site generation, or hybrid?
• What is the electrical topology and redundancy for your phase?
• Where is the demarcation point?
• What is the delivery schedule, and what does it depend on?

3. Cooling: posture and parameters

Ask whether the hall is air-cooled, liquid-ready, hybrid, or direct-to-chip capable—and whether liquid distribution is installed or only provisioned. Request supply temperature, cooling ratio, and supported density for your phase, not for the site in general.

4. Network: inside and outside the cluster

Evaluate both the cluster fabric (pathways, row layout, distance limits) and the external network (entrances, carriers, route diversity, cloud access). Carrier neutrality is site-specific; confirm which carriers are present today.

5. Operations and accountability

• Who operates the facility, and who is your contractual counterparty?
• What does the remote hands catalog include?
• How are incidents communicated and reviewed?
• Is there a responsibility matrix for every operating layer?

6. Contract structure

Review how the MSA, service order, site exhibit, and SLA fit together. Service commitments should appear in the agreement, not only in marketing materials.