“Greener” Data Centers in an Increasingly Digital and Data-Driven World

Lockdowns, travel restrictions, and the shift towards working from home are among the significant changes that arose due to the COVID-19 pandemic. Despite countries beginning to lighten restrictions and open their borders, the changes in culture and habits continue to remain. The rapid digitization of companies and growing work-from-home flexibility led to shifts in data consumption patterns, driving growth in data centers and data center services. The world’s largest data center operators, led by Amazon, Google, Facebook, and Microsoft, spent US$37 billion in the third quarter of 2020 on data centers. The byproduct of this rapid growth, however, is the significant increase of energy consumption and resulting CO2 emissions. According to the International Energy Agency (IEA), data centers consume approximately 200 terawatt-hours (TWh) of electricity, or nearly 1% of global electricity demand, contributing to 0.3% of all global CO2 emissions. To make things worse, one estimate sees data center energy consumption reaching 8% of global electricity usage by 2030 as digital transformation increases. Despite this, the use of artificial intelligence (AI) and machine learning (ML) has been one key element of achieving greater overall operational efficiencies in data centers.

Data centers have advanced over the years, with the energy consumption per gigabit, as of 2020, being twelve times lower than in 2010, according to the Borderstep Institute and the eco Association. However, this is somewhat negated by the fact that computing power demand has increased tenfold over the same period. And although the rise of renewable energy has also created an alternative and sustainable source of electricity for data centers, the complete transition remains to be seen. Therefore, in order for data centers to be more sustainable, it is important to focus on the cooling systems within data centers, a major consumer of energy and a key contributor to operating costs. According to Xudong Zhao, a professor of engineering at the UK’s University of Hull, cooling data centers can take up to 40% of their total energy consumption. The implementation of AI and ML with its predictive analytics capabilities has allowed for identifying energy wastage, performance issues in equipment, and areas where additional cooling is required, all of which optimize the performance in data centers and increase their efficiency.

In the U.S., the use of DeepMind ML in Google’s data centers to directly control the cooling systems has resulted in a substantial 40% drop in energy usage for cooling and a 15% reduction in overall energy consumption. Through its ability to adapt and learn data relating to temperatures, power, and pump speeds, the AI system could predict future temperatures and pressures, preventing over-utilization of electricity. Another example is Intel’s Memory Failure Prediction (MFP) which utilizes AI to analyze historical data to predict future memory failures, boost data center uptime, and lower total cost of ownership (TCO). Not only does the use of AI constitute a significant cost saving, but it leads to a dramatic reduction in carbon emissions for companies that rely on data centers. According to a study done by PwC, the deployment of AI in business operations could reduce global greenhouse gas emissions by 4% by 2030, equivalent to the combined emissions of Australia, Canada, and Japan.

However, it also important to note the energy consumption of utilizing AI in data centers. Due to the large amounts of data that require processing and computation, the AI systems also require significant amounts of energy to run. To tackle this, AI chipset manufacturers are directing their developments not only towards greater performance but doing so without growing their carbon footprint. For example, this year IBM introduced the world’s first quad-core AI accelerator chip at the forefront of low precision training and inference built with 7nm technology. It is one of the first chips to have power management features and maximize performance by slowing down computation phases with high energy consumption. This new AI core and chip technology can be used for many new cloud to edge applications across multiple industries.

As digital transformation continues to be at the forefront of companies’ developments, in addition to growth in global data consumption with the advances in connectivity (e.g., 5G), the demand for data centers will continue to rise. With that, it is crucial for industries to direct their attention towards sustainability and the environmental impacts of data center energy consumption. There is a need to design, build, and operate “green” data centers with not only maximum energy efficiency in mind, but also the need for the energy consumption to be less than what’s needed to cool them. This is especially important with new applications such as bitcoin mining and blockchain technology, 5G-driven distributed and edge computing, and more demanding latency requirements coming into the mix. According to the IEA, bitcoin mining uses more energy than some countries, including Austria and Colombia. The incremental usage of renewable energy alone will not be sufficient to address the adverse environmental impacts of data centers. Optimization of data center operations, effective management of cooling systems with the help of technologies such as AI and ML, and development of chipsets to attain high levels of performance while limiting energy consumption will be key to the future sustainability of data centers. With the growing reliance on data centers to support the continuous global digitization, these considerations and implementations will help contribute to the overall climate change movement.