Powering the demand for GenAI sustainably

Yam spent more than 26 years in the information technology industry and held senior regional sales and management roles for multinational. He is responsible for driving the growth of Eaton’s electrical business in the East Asia commercial organization. He manages the region to drive demand, orders, customer service, logistics and warehousing.

Generative artificial intelligence (GenAI) has been all the rage, with companies jumping on the bandwagon to roll out services making use of the technology. However, amidst all the buzz around GenAI, what’s often overlooked in the conversation is the immense power required to support its computational prowess. Even before the rapid proliferation of AI, it was already predicted that global compute demand would exceed the total amount of electricity generated by 2028.

The growing adoption of GenAI is thus spearheading a new wave of data and capacity requirements, and driving data center demand. This increased data load has serious environmental and financial implications — in fact, studies have shown that the training of large language models for ChatGPT creates up to 552 metric tons of carbon emissions. If current developments continue, research has also forecasted that the costs of developing and operating GenAI data center infrastructure could exceed $76 billion in the next few years.

As we enter a new era of digitalization and development, technologies like GenAI are here to stay. How can we ensure that data centers - the backbone of the digital economy - will be able to support this surge in power consumption efficiently and sustainably?

Finding the right conditions for GenAI success

From digital payments to health records, data centers provide the infrastructure required to store, process and distribute vast amounts of data that powers a wide range of applications and innovations.

Over the past five years, we have already seen how trends such as cloud computing and 5G have fueled data center growth. This prompted governments from around the world to take a closer look at how they can sustainably support the industry, resulting in the implementation of measures such as green standards and data center moratoriums to manage growth.

As GenAI becomes more accessible, more data will have to be processed and transmitted to deliver results in milliseconds. Compared to basic IT actions, such as retrieving a file from the company’s server, GenAI is much more compute-intensive. For instance, training ChatGPT can require 300 to 500MW of power, significantly higher than the typical 30 to 50MW of power consumed by a typical data center. Conventional air cooling methods, which relies on fans and air circulation, may be inadequate in dissipating the heat efficiently from densely packed server racks. This can result in overheating and power failure.

It is important to note that GenAI’s impact on power and cooling is not just restricted to hyperscale data centers, which are typically the go-to for companies with massive compute, storage and networking requirements, such as GenAI and Large Language Model (LLM) training.

Experts anticipate that as GenAI demand surges, developers will build small language models optimized for the edge. This could help to reduce reliance on hyperscale data centers and power required for AI inference.  Data is moved closer to the user, potentially cutting network latency and addressing data privacy concerns. However, this will not negate the amount of power required to sustain and cool GenAI overall.

Scaling the data center for digital innovation

While GenAI offers unprecedented possibilities, it also presents an important question on whether we are equipped to support the power demands of a digital, AI future.

In recent years, there have been advancements in data center technologies to improve energy efficiency and reduce their environmental footprint. For example, studies have shown that liquid cooling solutions can reduce facility power use by 27% even when adopted partially. Companies like Microsoft have also been exploring ways to make use of dormant energy assets in backup power systems by turning them into grid-interactive energy storage, enabling participation in the energy transition.

The gap now lies in whether organizations are building and retrofitting their facilities fast enough to meet growing energy consumption levels. Often, teams are hesitant to rock the boat for fear of disrupting systems and processes that are already working well. Especially where newer technologies are involved, costs of adoption may be higher than conventional solutions, and there may not be sufficient information to support the case for immediate investment. Continued industry collaboration to facilitate the exchange of best practices and development of commercially viable solutions will remain key if we are to achieve an industry-wide transition towards sustainability.

Building the talent pipeline of data center professionals who are equipped to manage the modern data center effectively will be crucial as well. This includes equipping them with skills related to cybersecurity and data analytics in order to harness automation, visualization and analytics tools. With research estimating that nearly 2.3 million staff will be needed to run the world’s data centers by 2025, the digitization of data center operations can help to reduce pressure on teams in day-to-day operations.

Ushering in a new age of data centers with GenAI

The success of GenAI is dependent on many factors ranging from the reliability of data infrastructure, computing resources to optimization techniques. As adoption ramps up, there will be increasing pressure on data centers to be high-performing and efficient. I’m positive that GenAI is just scratching the surface of innovations we will see in the coming future, and data centers will continue to play a crucial role in making these ideas come to life. By keeping energy costs in mind, we can continue to fuel digital innovation for a long time.