Artificial intelligence (AI) is a technology that enables computers and machines to simulate human intelligence and problem-solving capabilities. Over the past few years, AI has evolved significantly, and its new category, generative AI, refers to creating new data with a wide range of outputs including texts, images, and music. Machine learning models and other transformer models are key components in generative AI which generates data that is similar but not identical to original data (Suleimenov et al., 2020) .
However, AI users are less familiar with data centers and their vital role toward this influential technology. Data centers have become essential components for any tech company around the world. Nvidia, which is a big player in developing infrastructure for data centers, said its data centers business grew 427% by last quarter.
A data center is a place or a physical location where large companies related to banking, technology, finance, and health among others) put their IT infrastructure such as servers, data storage drives, and other servers in a secure environment. These data centers allow businesses to run. The larger the size and data footprint of a company, the larger the size of the data center required. Thus, the larger the data center, the larger the power required to run it.
Rise in Energy Demand To Run Data Centers
The computationally intensive tasks that an individual performs using any AI model requires power. A recent study revealed using an AI model to generate an an image uses as much power as it takes to charge your smartphone. This power is provided to machines by graphic computing units (GPU), which are semiconductor chips. When running at maximum power and performing at their peak capacity, these processors heat up and reach up to 80°C. Therefore, it necessitates a solution to prevent it from overheating, preferably a cooling system that can reduce the temperature of piping-hot AI data centers. This cooling system is more than a simple air conditioning unit. Usually, a method such as liquid cooling is preferred, which is more efficient than air cooling, but is also more expensive. Overall, these cooling systems installed in data centers require significantly higher power to perform their functions.
The rise of companies using generative AI has created more demand for data centers which require a significant amount of power. This presents a problem as there is not enough electricity to run them. As of March 2024, there are approximately 10,978 data centers worldwide. Table 1 represents the list of the top 10 countries with the most data centers.
The utility companies, which are already in a doldrum due to continuous pressure on the electrical grid, have limited ability for this growing need for power for AI data centers. However, the demand is not slowing down. In the US, Northern Virginia has the largest concentration of data centers, and energy utility companies operating in there forecast 100% growth in energy needs over the next 15 years. Furthermore, the International Energy Agency (IEA) forecasts that by the year 2026, data centers globally will use an amount of energy per year equivalent to Japan’s electricity consumption.
A Big Problem for Tech Giants
In 2023, data centers consumed 7.4 GW of power, a 55% increase from 4.9 GW of power last year. Large technological companies like Microsoft, Meta, and Amazon are concerned with utility companies not being able to meet the energy demand for data centers to keep up with the growth of AI. This is forcing them to increase the number of natural gas power plants, and in some cases, coal-fired power plants. Many of the same companies have committed to reducing their carbon emissions. It has been reported that the carbon dioxide emissions of Microsoft have increased 30% from 2020 to 2023, making its goal of becoming carbon negative by 2030 more difficult.
Tax Exemption for Data Center Facilities and Outraged Residents
With large facilities required to provide the necessary amount of power, a big draw for tech companies is the tax incentives that governments offer in return for creating more jobs for residents. More than half of the states in the US offer tax incentives to tech giants for setting up big data center facilities and companies are getting millions of dollars of tax exemptions in exchange for hiring 50–100 people.Microsoft has announced plans to spend $3 billion to build an AI data center in Wisconsin, citing at least 2,000 jobs for local residents.
Building data center facilities has become a lucrative real estate business while factors including climate, energy costs, and the risk of natural disasters affect where companies choose to build data centers. Still, it is yet to be determined whether these data center facilities are proving to be a benefit or burden on the local community. Some state administrations in the US where tax incentives are given to tech companies are passing new laws for limiting or suspending the tax exemptions.
Alternate Energy Option for Data Centers
Many larger tech companies have committed to becoming carbon neutral in the next 10 years with many considering running their data center facilities in carbon-neutral fashion with built-in nuclear reactors. Why nuclear energy? Reactors producing electricity in nuclear power plants have larger capacity as compared to fossil fuels and other renewable energy power plants. Additionally, nuclear power plants are built in large areas where surrounding land is vacant.
Amazon recently purchased a $650-million nuclear-powered data center campus facility in northeastern Pennsylvania that will use up to 40% of electricity generated from an on-site reactor, eventually enabling the company to reduce its reliance on the grid. Meanwhile, Microsoft has hired nuclear experts to spearhead its quest for this alternate power source. The company has also made contract agreements with nuclear power plant operators to provide power to one of its data centers in Virginia.
Conclusion
Data centers require a continuous and consistent supply of power to operate servers and other core operational equipment. If the energy supply is not consistent, downtime can cause significant financial losses to corporations and other end users. The focus on reducing power consumption for data centers is not a new concept, and it is understood that more sustainable ways of operating are required. However, energy pricing and commercialization of future technologies have their own challenges. To create data centers that are greener and energy efficient, the use of renewable energy and new technology in power reduction has a role to play.
With data center energy requirements expected to increase, obtaining power from renewable energy sources is also not a complete solution, as developers are also looking for other technologies to increase energy efficiency in the operation of data centers. Some of these new technologies include the use of snow, seawater, liquid cooling, or other geothermal techniques instead of conventional (Gullbrand et al., 2019; Tsuda et al., 2015) cooling systems, which accounts for a vast portion of data center total energy consumption. Data center developers who can make this transition toward a more efficient and greener system will anchor themselves as key players in this growing industry.
For Further Reading
Liquid Cooling of Compute System by J. Gullbrand, Intel; M. Luckeroth, M. Sprenger.
Reimagining Future Energy Systems: Overview of the US Program to Maximize Energy Utilization via Integrated Nuclear-Renewable Energy Systems by S. Bragg-Sitton, R. Boardman, C. Rabiti; Idaho National Laboratory.
Artificial Intelligence: What is it? by I. Suleimenov, National Engineering Academy of Republic of Kazakhstan, Y. Vitulyova, A. Bakirov.
Using Naturally Cold Air and Snow for Data Center Air-Conditioning and Humidity Control by K. Tsuda, S. Tano, J. Ichino, Tokyo City University.
