The influx of artificial intelligence (AI) in data centers creates significant demand on critical power equipment. It will only become more of an issue, as analysts predict that agentic AI will handle 15 percent of day-to-day work decisions by 2028.
All this AI is going to require more power. One estimate is that three New York Cities’ worth of electricity will be added to the grid by 2030, largely due to new AI data centers. Data center power distribution designs must reach new performance and reliability levels as a result. When designing data center UPS systems for this new era, it's important to consider power quality/delivery, availability, sustainability, and total cost of ownership (TCO).
The nature and complexity of AI computation will test the resiliency of a data center’s power infrastructure. Ten times more electricity is used for a search query on ChatGPT compared to a traditional Google search. The extra capacity places considerable pressure on temperature, cooling, and grid resiliency.
Adding complexity is how this capacity is generated. AI models can spin up to 150 percent of their power capacity for a few seconds – even milliseconds – thus putting more pressure on the entire power chain (i.e., grid, switchgear, UPS). Everything must be optimized to work together and properly support these loads.
Fast adoption of AI is quickly changing the conversation around power density. Data centers often had power densities of 30kW per rack a few years ago but, today, many have 50kW per rack, with 150kW densities planned and even 500kW being talked about in the not-so-distant future.
AI systems will require backup systems with well over 1MW of power while having to deal with up to 150 percent power peaks. For these reasons, AI requires a massive, singular, and ideally purposely built data center.
Meeting AI power demands
No power means no data, so power availability needs to be analyzed. Will the source (grid, wind, solar, etc.) be able to support the data center? If not, how often will an alternative power source, such as a backup system, be used?
If the backup system is relied on regularly, cycling batteries are better for the UPS than float batteries. Cycling batteries are better for on-off applications and consistent use. They are not as power dense as float batteries, though, leading to possible oversizing of the system.
Beyond the potential impact on battery chemistry, AI loads are putting additional load and requirements on both the UPS system and even the grid. Major OEMs and utility companies continue to work on optimizing their systems to handle these power bursts.
Achieving sustainability
Everyone agrees that sustainability is important. True sustainability must account for the full life battery cycle – sourcing, operating, and disposing.
A responsible manufacturer who is ethical ensures that a battery is properly sourced. Sustainable practices should be used during design and manufacturing.
Recycling programs vary by company, as well as chemistry. Lead acid has a well-established ecosystem that will pay for battery recycling. Ecosystems for newer technologies, such as lithium-ion, are evolving. There may be a recycling fee as a result.
Determining True TCO
Four factors accurately determine the TCO of a battery backup system:
- Installation – Day 1 costs, such as the batteries, along with cabinets or racks, and labor to connect the system, are installation expenses.
- Maintenance – The biggest factor is the frequency with which maintenance needs to be done. Another consideration is labor cost, as some technologies may require higher-skilled labor.
- Replacement – When comparing technologies, it can be very hard to determine when a battery should be replaced. While there isn’t one approach, it is recommended to start a TCO calculation using the standard warranty for each technology. A warranty is a clear sign of how much a company will support its products, making it the best way to compare batteries. A standard warranty should always be used as the true measure.
- Disposal – Selecting a battery means you are also selecting a partner. Choose a manufacturer who has a battery collection and recycling program that provides an eco-friendly, safe, convenient, and low-cost recycling option.
Conclusion
Many factors go into choosing a battery backup system for AI data centers. A partner who can develop a complete solution that best fits the application is as important as the battery.
As AI continues to impact power requirements, UPS must be designed from the beginning. This will help ensure the extra power demands are consistently met.