Chips
The hardware that performs the computing work.
← Back to Compute College
Compute College
What is an AI data center? Power, cooling, and GPU capacity
The facility stack that turns accelerators into operating AI compute supply.
Power + cooling + networkFacility stack
The building is valuable when support systems can sustain AI workloads.
Stage mattersSupply status
Planned, built, energized, commissioned, and operating capacity mean different things.
Plain-English definition
An AI data center is the powered, cooled, networked facility that turns accelerators into usable compute capacity. The market impact of a data-center announcement depends on whether the site can host dense AI racks, operate reliably, and make capacity accessible.
Memory trick: A data center is the powered workshop around the chips, not the chips alone.
Why it matters
- Servers and accelerators that perform the computing work.
- Electrical systems that deliver reliable power.
- Cooling systems that remove heat.
- Networking that connects systems inside and outside the facility.
- Storage, security, monitoring, and operations that keep the site running.
Simple example
A GPU is like a machine on a factory floor. The data center is the factory: the building, electricity, cooling, wiring, networking, controls, and people that allow many machines to operate together.
Power
The electricity required to run dense systems.
Cooling
The systems that remove heat from high-power equipment.
Networking
The connections that let systems work together.
Operations
The monitoring, controls, and people that keep the site running.
When those pieces work together, hardware becomes usable compute capacity. Any figures shown are illustrative calculations, not current quoted market prices.
Market signal
How to read the market signal
- AI workloads often require many accelerators to work together.
- High-density deployments need more power and more heat removal than ordinary office computing.
- The physical site determines how much capacity can be deployed, where, and how quickly.
- Data-center expansion links AI demand to land, power, construction, equipment, and utility planning.
Market read: ready AI facilities can bring supply online sooner; missing power, cooling, or networking leaves capacity prospective. Figures here are illustrative unless explicitly sourced and dated — see our methodology.
Common mistake
A site can exist before it is fully useful for AI. The building may be ready, but without enough power, cooling, networking, or installed equipment, it may not yet provide the capacity buyers care about.
Shell
Building
The physical shell.
Systems
Infrastructure
Power, cooling, and networking systems.
Market
Usable capacity
What exists only when the site can support real workloads.
Practical takeaway
What you can do with this
Use the data-center checklist when assessing an AI capacity offer or buildout: power, cooling, network, rack readiness, equipment, operations, and access all matter.
- Procurement teams: verify that the site supports the hardware density and service terms required.
- Analysts: separate building announcements from facilities able to host operating AI capacity.
- Ask whether a site is planned, built, energized, commissioned, or serving customers; those stages imply very different effects on near-term supply.
- When two sites quote similar accelerator time, include network access, reliability, and any constraints on moving data into and out of the facility.
Decision check: count capacity as usable only when the facility and equipment can deliver the intended workload.
Compute College
Turn the lesson into a number
Use the GPU-Hour Cost Calculator, AI Training Cost Calculator, or Model Serving Cost Calculator.
Compute College track
Power & Data Centers
Step 2 of 17: What is a data center