From day one, Compass has been committed to sustainability in data center development. We use advanced building technology, and continuously refine our designs and processes for efficiency.
When Compass was founded in 2011, evaporative cooling had emerged as the industry’s preferred way to cool servers in data centers. However, we viewed it as a wild waste of water. We refused to churn through millions of gallons of water per day, especially since we build data centers in communities like Texas and Arizona, where water is a precious resource. Instead, we used air cooling.
However, in our continuous improvement culture, as AI drives a surge in demand for data storage and processing, we find ourselves reconsidering cooling. AI models require enormous amounts of power to serve the computing capacity they demand. Gartner estimates that by the end of 2024, machine-generated synthetic data will comprise 60% of the data used in AI-training.
At Compass, we’re well on our way to providing hyperscalers a secure place to plug in to this new reality. Because we’ve always been a company that looks to the future, we’ve built our processes and designs to be highly flexible. Our modular design, standard kit parts and longstanding relationship with Vertiv, our cooling partner, means we can meet our tenants’ evolving needs in this new reality even faster.
Too many cooks in the kitchen: how AI impacts data center cooling
Today, data centers primarily store central processing units (CPUs). CPUs are like the head chef in a kitchen, serving as the main processor that can handle a wide variety of tasks. Graphics processing units (GPUs), on the other hand, are like sous chefs who focus on more specific tasks, quickly whisking, chopping and tossing massive amounts of specialized data.
A single AI language model can require 1E12 (that’s one million million) data points. GPUs, which are necessary for AI models, therefore expend large amounts of energy, which creates a lot of heat in the kitchen.
After spending too much time in that hot kitchen, those sous chefs would need to cool down. But cooling GPUs isn’t as simple as cooling CPUs. Stepping outside for some fresh air on a 70-degree Fahrenheit night would be refreshing. But jumping into a 70-degree Fahrenheit pool will cool them down a lot faster.
This illustrates why liquid cooling can be more efficient than air cooling in the AI era. The density of water surrounds the device that needs to be cooled, effectively lowering its temperature very quickly.
Substituting ingredients for the best recipe: flexibility, by design
Most of our data centers still use air cooling today. But our standard kit of parts means we can easily substitute new solutions and do it fast. This flexibility will enable customers to customize equipment within the data hall as more GPUs enter the mix.
As we transition to liquid cooling to meet the needs of our customers, sustainability is still top of mind. While liquid cooling is more energy-efficient than air cooling, it still uses a significant amount of water. We’ve always been committed to minimizing our impact on communities’ water resources. That’s why we will only use a closed-loop system that does not allow the water to evaporate.
With a closed-loop system, the system only needs to be filled once. Water is pumped outside, where heat exchangers and chillers still use air to cool it, without any form of evaporation. We treat water with alcohol and glycol to prevent it from freezing in the winter so we can use it continuously.
Looking to the future: hybrid cooling systems
As data centers adapt as technology evolves, the only thing certain is that things will continue to change. For example, we may see a future where a hybrid approach to cooling is the norm, with some racks of GPUs paired with CPUs for maximum efficiency. While high-powered chips needed for AI might require liquid cooling, other components could remain air-cooled. That’s why we’ve always approached data center development through the lens of long-term flexibility, continuous improvement and sustainability. When changes come, we’re ready to accommodate our customers, now and into the future.
Whatever the future holds, we’re committed to embracing it carefully and respectfully, keeping the needs of our communities in balance with those of our customers.
