A new data center has recently opened in Middenmeer, Netherlands, reflecting a shift in how utilities are managing electricity demand. Following years of stable power usage in developed economies, energy providers are now upgrading their forecasts, as data centers seek access to grid connections amid rising consumer and business needs. In Northern Virginia, for instance, the wait time for a new data center to connect to the power grid can extend up to 14 years. The urgency is evident, as concerns grow that increasing reliance on artificial intelligence (AI) could overwhelm existing electrical infrastructure.
It's important to recognize a crucial aspect of electric systems: the growth in demand often spurs innovation and adaptation. Electricity consumption is not a static figure that utilities can satisfy solely by adding more generation capacity. While new power generation is essential, modern electric grids are capable of evolving through various means, such as boosting efficiency, enhancing storage solutions, ensuring demand flexibility, and optimizing the current infrastructure.
To explore the energy sector's responses to AI growth, I spoke with executives from Delta Electronics, Eaton, and DNV. Their insights reveal that the solution goes beyond constructing new power facilities; it involves increasing efficiency before energy reaches semiconductor chips, rethinking cooling systems for data centers, and addressing the grid bottlenecks that affect the speed of new demand integration.
Currently, the process of connecting data centers to the electricity grid can take several years. According to Ali Ghorashi, Senior Vice President at DNV, the current landscape of energy investment is markedly different from past trends. Data centers now find themselves at a crucial junction of several traditionally separate sectors: real estate, technology, utility markets, and infrastructure financing. "Few people are familiar with the entire industry," Ghorashi pointed out, highlighting that many stakeholders come from one domain without comprehensive knowledge of the rest. This disconnect is increasingly apparent during grid interconnections. While major tech companies possess the financial resources and strong demand signals needed for large electrical loads, power infrastructure operates on its own timetable. Utilities require certainty that the estimated demands for hundreds of megawatts will materialize and maintain economic viability. Regulatory bodies are responding by raising application fees and considering penalties for projects that reserve power without follow-through. In certain regions, such as British Columbia, data center projects are outright forbidden. Ghorashi emphasizes that overcoming these coordination issues is just as critical as expanding generation capacity. Without a reliable way to deliver electricity due to transmission limitations or delays in permits, the energy challenges posed by AI will only become more pronounced.
The quest for efficiency also plays a crucial role within the data centers themselves. Before electricity can execute tasks, it undergoes multiple conversion processes, each contributing to energy loss. As such, it's essential to optimize these steps to meet overall energy needs. Franziskus Gehle, Vice President at Delta Electronics, stressed the importance of system-wide efficiency. “When we handle power conversion, striving for optimum efficiency is vital,” Gehle explained. At the scale of today’s AI infrastructures, even minor losses can accumulate into significant energy waste. Delta reports that its efficiency enhancements have saved approximately 45.5 billion kilowatt-hours of electricity from 2010 to 2023—enough power for millions of households over an extended period. This focus on efficiency extends beyond individual companies to encompass their entire infrastructure footprint. Delta's initiatives include investments in renewable energy procurement, creating net-zero buildings, geothermal systems, waste heat recovery, and enhanced energy monitoring capabilities. Gehle believes that improving the infrastructure utilized by clients will have a far-reaching impact, surpassing the benefits of just focusing on internal energy efficiency.
As AI processors grow increasingly powerful, the focus shifts from merely delivering electricity to managing heat. Any computational activity inevitably generates heat, and the high density of modern AI systems is leading to new design considerations for data centers. Paul Ryan, Vice President of Energy Transition at Eaton, points out that rising power density is one of the significant trends. "With 140 kilowatts per rack, air cooling isn't an option," he stated. Instead, liquid cooling systems are moving in closer proximity to the processors. Enhanced cooling not only reduces energy expenditures but also boosts operational performance, which is crucial for AI infrastructure that can represent multimillion-dollar investments. Furthermore, the heat generated can be repurposed: district heating, greenhouses, swimming pools, or various industrial applications can all benefit from this byproduct. However, there are technical challenges, as data center heat typically ranges from 50 to 60°C, whereas many industrial processes require temperatures exceeding 100°C. Still, Ryan notes that the potential for energy reuse is significant. “One gigawatt at 50°C is an enormous amount of energy,” he remarked, suggesting that there are numerous opportunities to harness this resource.
