The growth in data centers — fueled by the explosive demand for AI, large language models (LLMs), cloud storage and computing — has put a real strain on natural resources and power grids. Calls for more sustainable operations, or even outright protests against data centers, are growing at a seemingly similar pace.
To quote a recent Siemens white paper: “A sustainable data center is a tradeoff between speed and money — tokens per watt — and long-term viability of a data center facility.” Every data center operator wants to know how quickly they can deploy a data center and begin generating tokens, and how many tokens they can generate per watt consumed. If it is starting to feel like a race, that’s because it is. The race to be the fastest and biggest, and it’s also a race to make money off the demand.
What can realistically be done to reduce the drain on resources while feeding the demand for the services these data centers provide?
Data center cooling strategies for AI workloads
Cooling is one of the most important factors for successful data center operations. The large racks of CPUs running 24/7 generate a lot of heat that needs to be removed quickly and efficiently before it impacts performance. Evaporative cooling has been shown to be the fastest and easiest way to achieve rapid cooling; however, it also uses staggering amounts of water.
One of two solutions Siemens offers, in cooperation with some of its business partners, is air cooling with redundant backups or a liquid cooling architecture. Each is controlled by software and controllers that can respond instantly to changes in the environment or performance.
How air cooling can reduce water use in data centers
At the Novva data center complex in Colorado Springs, Siemens migrated the data center from water to air cooling, saving water and 2 million kilowatt-hours annually. The kilowatt savings were achieved mainly by eliminating downtime and the need for weekly service.
The new cooling system was set up with redundant duplicate components. The flat architecture designed for this facility, with enhanced redundancy, allows the facility to run four chillers, two chiller plants, cooling towers, pumps and fans simultaneously. Should one chiller develop a problem or fail, the other three adjust instantly because that’s how the system software is programmed. This allows the facility to operate fully 24/7 with optimal temperatures at all times, without the risk of overheating.
Why liquid cooling is gaining traction in AI data centers
Liquid cooling is a closed-loop cooling system that doesn’t use extra water beyond what it’s originally filled with. Heat transfer capabilities for liquid are, on average, 3,500 times higher than air, because liquid is denser than air. This solution is included as part of a 100 MW Hyperscale AI Blueprint reference architecture developed by Siemens in cooperation with nVent and NVIDIA, designed specifically for next-generation data centers. The white paper for this blueprint can be found here.
The application is scalable and can be seen as a framework that forms the basis for infrastructure providers to build the system they need. It is designed with Siemens industrial-grade electrical systems and nVent liquid cooling devices, which are offered as indirect liquid-cooled, direct liquid-cooled or a hybrid version, depending on the specifications of the data center infrastructure. According to the white paper, the framework accelerates time-to-compute and maximizes tokens-per-watt.
Energy management software for data center efficiency
A few years back, Siemens provided a package of software-based solutions to a data center in Estonia. This single-vendor solution included building management software, energy and power management software, and white space cooling optimization software (or WSCO for short). They applied AI-powered software to optimize cooling for energy efficiency and to support the consistent operation of critical infrastructure, including customizing power distribution to ensure safe, reliable power throughout the data center.
Much of this particular data center is powered by renewable energy and has the ability to sell waste heat, which, in a colder climate, is a good use of the heat produced by data center operations. When a data center is located far from any urban areas, it becomes more difficult to consider selling waste heat to utilities.
At the Estonian data center, a dense mesh of sensors across the center’s white spaces provided the means for white space cooling optimization. By analyzing the effective cooling, the AI can create an influence map from which to optimize cooling distribution at the rack level, thereby limiting energy usage to power only what’s necessary, instead of powering each section with the same amount of power.
How digital twins improve data center energy and resource planning
A digital twin is already a well-established process for testing settings and adjustments to see how they will perform in real time before implementing them in a facility. It can have a similarly useful application in data center operations, where fine-tuning energy and water usage will save money and resources.
Adding renewable energy sources to the mix makes a digital twin setup useful, as it allows for better control of each source and can help a facility adjust quickly if one source of energy is producing less than anticipated.
Balancing AI data center performance, sustainability and ROI
Ultimately, making data centers sustainable requires a balance between the use of resources and the speed at which the facility operates. The examples shared show what can be done and how a facility can be greener, but each step takes time and money to implement and won’t yield an instant ROI.
Making a data center sustainable comes down to what each individual data center operator wants. With the current demand, getting one built and operational quickly appears to be a priority over sustainability. It’s about getting those tokens per watt.
Sustainability is a choice and one that each data center owner will need to make. What are the trade-offs involved in opting for sustainability? What is the upfront cost versus the long-term ROI? How much does public opinion matter, and what are the federal and local regulations regarding water and power usage? In the United States, the rules vary by region.
Quiz