Improving Energy Efficiency in AI Data Centers - Lorbel

Improving Energy Efficiency in AI Data Centers

AI infrastructure is power-hungry. Data centres that support large language models can consume more than a small town can pull in a day. The costs of running these high-performance computing systems are far greater than traditional facilities, making energy costs a major operational expense.

Improving energy efficiency is becoming an important part of building sustainable, competitive AI infrastructure.

Where the Power Actually Goes

Many people picture servers when they think about the main power draw in a data centre, but that’s only a part of the story. Cooling, power distribution, and backup systems also account for a significant portion of total energy use.

A typical high-density facility’s power usage may look something like this:

• Compute and servers: The biggest single chunk, especially with GPU clusters
• Cooling: Rises fast as rack density climbs
• Power conversion losses: Every transformation from AC to DC and back costs energy
• Backup power systems: Often ignored, but they run constantly, not just during outages

That last one surprises a lot of facility managers. A UPS isn’t just sitting there waiting for a blackout. Rather, it’s actively conditioning power around the clock. If it’s outdated or oversized, it’s wasting energy every single hour.

Why UPS Performance Matters More Than You Think

Older UPS systems run in double-conversion mode all the time. That mode is reliable, but it’s not efficient. Every watt gets converted twice before it reaches your servers, and each conversion loses energy as heat.

Modern UPS units offer eco-mode or high-efficiency mode instead. Power flows more directly to the load, with the system stepping in only when it detects a problem. Efficiency numbers jump from the low 90s into the high 90s. That gap adds up fast at data centre scale.

Sizing matters too. A UPS running at 30% capacity is far less efficient than one running within its designed efficiency range. Facilities that scaled quickly for AI workloads often end up mismatched, with some units stretched thin and others sitting underutilized.

Practical Ways to Improve Energy Efficiency

You don’t need to rip everything out and start over. Small, targeted changes make a real difference.

1. Audit your current UPS fleet. Know the age, capacity, and efficiency mode of every unit before you decide anything.
2. Right-size new installations. Match capacity to actual load, not projected load from five years ago.
3. Switch to eco-mode where it’s safe to do so. Not every application can tolerate the brief switchover delay, but many can.
4. Schedule regular maintenance. A poorly maintained battery system forces the UPS to work harder than it should.
5. Consider modular UPS systems. They scale with demand instead of running oversized from day one.
6. Pair UPS upgrades with cooling improvements. Heat from inefficient power equipment adds directly to your cooling load.

None of these require a total redesign. They just require someone paying attention.

Improving Energy Efficiency in AI Data Centres

Energy efficiency isn’t just an environmental consideration. It also has a direct impact on operational costs. Since AI infrastructure is expensive to build and maintain, even small reductions in power consumption can go straight back into your budget.

There’s also a resilience angle, as efficient systems generate less heat, place less stress on components, and tend to fail less often. Fewer failures mean fewer emergency calls and less downtime. For facilities running AI workloads around the clock, downtime is costly in ways that are hard to walk back.

Where to Start

The fastest way to find savings is to get eyes on your current setup. A proper assessment of your UPS systems, batteries, and power distribution reveals exactly where you’re losing efficiency.

Lorbel works with facilities across California and neighboring states on this kind of assessment, from installation through service and maintenance. If your data centre is scaling up for AI and your power infrastructure hasn’t kept pace, now’s a good time to take a closer look.

Get in touch with Lorbel to talk through your facility’s power needs.