Data Centers & Power

The cloud is heavy. Global data centers consumed approximately 460 TWh of electricity in 2022, nearly 2% of total global electricity demand.

Server racks in a modern hyperscale data center

Hyperscale vs Colocation

A "Hyperscale" data center (built by AWS, Google, Meta) usually exceeds 5,000 servers and 10,000 sq ft, often drawing 50-100+ Megawatts of power. Colocation facilities (Equinix, Digital Realty) act as hotels for servers, where multiple companies rent space and cross-connect to Tier 1 networks.

The geographical placement of these centers is dictated by three constraints: cheap power, ambient cooling, and proximity to submarine cable landing stations.

Power Usage Effectiveness (PUE)

PUE is the industry standard metric for energy efficiency. It is the ratio of total facility energy to IT equipment energy. A perfect PUE is 1.0, meaning 100% of power goes to computation, and 0% to cooling, lighting, etc.

Facility Type	Average PUE	Characteristics
Legacy Enterprise (Pre-2010)	2.0 - 2.5	CRAC units, poor airflow management, high waste
Modern Colocation	1.4 - 1.6	Hot/cold aisle containment, optimized chillers
Hyperscale (Google/Meta)	1.10 - 1.15	Evaporative cooling, custom servers, AI-managed HVAC
Immersion Cooling (Edge)	1.02 - 1.05	Servers submerged in dielectric fluid

Tool: PUE Calculator & Cost Estimator

Calculate the PUE of a facility and see the financial impact of cooling overhead.

JavaScript Required: Formula: PUE = (Total Facility Power) / (IT Equipment Power). E.g., 1000kW total / 600kW IT = 1.66 PUE.

Common Mistakes in Data Center Planning

Ignoring Water Usage Effectiveness (WUE): Achieving a 1.1 PUE often requires massive amounts of evaporated water for cooling. Facilities in arid regions (like Arizona) trade power efficiency for water consumption, causing local political friction.
Overprovisioning UPS: Running Uninterruptible Power Supplies at 30% load is incredibly inefficient. Modern architectures push battery backup to the server rack level.

FAQ

Why are data centers built in cold climates?

To leverage "free cooling" (economization). If outside air is below 20°C, you can use heat exchangers to cool the servers without running energy-intensive mechanical chillers. Sovereignty laws, however, force data centers to be built in warmer climates.

How do generative AI workloads impact this?

Traditional server racks consume 5-10kW. AI training clusters with dense H100 GPUs push 40-100kW per rack. Air cooling physically cannot remove this much heat, forcing the industry to adopt direct-to-chip liquid cooling.