How Access to Power is Hindering Advances in AI

According to computer scientists, the artificial intelligence (AI) we see today pales in comparison to what it could become.

• Imagine technology that could tailor a student’s education based on their personality and abilities
• Imagine if medicine had the means to treat or even eliminate blindness, deafness and even neurological disorders such as Parkinson’s and Alzheimer’s
• Imagine if police were able to use facial and behavioral recognition, object detection and other advanced technologies to prevent crimes before they happen

AI is poised to make tremendous advances in the near- and mid-terms. But a single factor is holding back AI-based technology advancement: Power.

Why the increased need for energy?

Data centers are scrambling to expand capacity to meet exploding demand primarily due to power-intensive AI applications. But they are hampered by utility companies’ ability to provide sufficient energy. Most data centers were built to handle the needs of cloud computing. But AI is taxing their capacity. AI servers need roughly 10 times the computing power¹. They also produce 10 times more heat than cloud servers and require advanced cooling methods such as liquid immersion².

As a result of AI, industry experts believe that the demand for energy may grow by 85% over the next 15 years, and that key data center markets may not have the energy they need within a mere 3-4 years³.

There’s also a development gap. While it can take a utility company 2-4 years to connect a large customer to the grid, a data center can be built in 18 months⁴.

Fortunately, there are steps you can take that can help flatten the demand curve and help make your data center development plans more attractive to your utility.

1. Release your stranded capacity

“Stranded capacity” refers to installed assets that can’t be used to support critical operations. A data center’s stranded capacity can be found in its power, space and cooling infrastructure. For example,

• Older, inefficient servers that are still drawing power and cooling resources but not actually being used
• Underutilized servers that could be consolidated into more efficient equipment
• Legacy and/or oversized lighting and HVAC equipment
• Cooling of non-essential equipment and/or areas
• Poorly sealed openings under servers with raised floors, resulting in inefficiencies

Freeing stranded capacity in your data center not only demonstrates to your utility that you are using as much power as you are currently allocated, but it also helps increase your leasing capacity.

2. Implement an energy storage system

While your data center may need 100 MW to operate, that doesn’t mean you’ll need the utility to deliver that much 24 hours a day, 365 days a year. In many areas of the country, a utility may experience a strain on their grid for only 10-15 days of the year. With a strategically designed cooling system, you can show the utility how you’ll help reduce your usage during those peak loads.

To curtail your peak demand loads, you’ll need to be able to “unplug” from the utility for a certain amount of time and still operate. Think of your data center like a laptop computer. It charges while it’s plugged into power, but still runs for a while just on its battery.

Energy storage works the same way. Using one of several available types of storage technologies, you can help save that energy for use when you need to unplug from the grid. This includes approaches such as electrical energy storage, battery storage, mechanical storage, biomass and more. At Trane we can work with you to find the right storage solution based on need and application. Here are two solutions we’re seeing customers put on the top of their consideration list:   

• Pumped Hydropower Storage (PHS): This technology uses water and gravity to generate power. Water is pumped up a slope to a basin, where it’s stored until you need energy. When the water is released, it runs downhill through a pump turbine, which generates energy.
  
  
• Thermal Energy Storage (TES): A thermal energy storage system such as the Trane Thermal Battery is a chiller plant that stocks energy at certain times, then releases it as needed. For example, your data center could operate on energy stored in a TES during peak loads and/or when utility rates are highest. A TES uses large tanks of water, transforming it into ice during night-time off-peak hours, then releasing it during the day. Designed as “full” or “partial” storage systems, a TES can supplement cooling by the chillers, or it can completely replace the chillers for a certain amount of time.

3. Create a source of renewable energy

Any of the energy storage systems mentioned above can be powered by renewable energy sources, such as wind or solar. Could the energy needed for your new data center be supplemented by solar panels on the roof? Or a small wind turbine? Or geothermal? Remember that you only need to generate enough energy to create power for storage, not to run your entire operation. But demonstrating how you’re going green can help when negotiating with utility companies.

How Trane can help

Trane’s proprietary solutions could redefine how you think about developing new data centers and retrofitting your old ones. When looking at the total grid, our proprietary technologies, energy contracts and ITC credits could help reduce the cost of building a new data center. Flattening the power demand curve could help:

• Reduce new data center costs by 80-85%*
• Minimize the impact on local utility companies and remove barriers to building your new data centers
• Show investors, tenants or your corporate leaders that you’re serious about building more environmentally-friendly data centers

Trane experts can advise you on the development, design and operations of your new data center, and help you create an approach to cooling and energy use that can position you favorably with utilities. By working together, we can play a part in fulfilling the potential of AI to help elevate the quality of life across the globe.

^{* Based on an average of $10-12MM per MW cost for new data centers}