AI in the Data Center: Building Partnerships for Success
AI is bringing about fundamental changes in virtually every industry, and data centers are no exception. As demand for AI continues to skyrocket, data center developers and operators must find ways to increase capacity. However, many are running up against constraints that limit their ability to build new facilities, access the power they need and ensure reliable operation.
At the same time, AI itself places unique demands on the data center environment. It’s not simply a matter of performance. AI workloads require extremely low latency, forcing data centers to increase density to minimize distance. Of course, greater density means more heat, straining cooling systems.
The right partnerships can help data center developers and operators address these challenges through innovative approaches to power generation, cooling, infrastructure design and more. Data centers need to forge strong relationships with suppliers and service providers who can help them meet changing requirements while reducing costs and risks.
External Barriers to Data Center Growth
To keep up with the demands of AI, data centers need to add exponentially greater compute power to already strained IT infrastructure. Space is at a premium and electrical and cooling systems are struggling to keep pace. However, data center developers and operators are running up against barriers to growth. Data centers will need new strategies to scale capacity and capitalize on the opportunities presented by AI.
Space Limitations
Land is a finite resource. Several communities, including Amsterdam, Dublin and Loudoun Co., Va., have imposed moratoriums or restrictions that have effectively halted new data center construction. Salt Lake City and Santa Clara, Calif., have also limited data center development. Some data center developers have opted to construct their facilities outside traditional technology hubs. However, smaller communities such as Burns Harbor, Ind., in the Chicago Metro area, and Fayette Co., Georgia, southeast of Atlanta, are also pushing back against data centers.
In light of these limitations, data centers need to maximize space in their existing facilities. This will require even greater density, which in turn requires new configuration strategies. Sustainability will have to be baked into every design decision to ensure availability, resilience and performance.
Power Constraints
Many communities that are limiting new data center construction are concerned about new demands on a power grid that’s already strained. JLL’s Data Centers 2024 Global Outlook report predicts that average rack density will increase 7.8 percent through 2027 to 48.7kW. In some areas, the power grid simply cannot support these demands.
Limited supply isn’t the only power-related challenge data centers face. The power grid is increasingly fragile at a time when data centers need reliability more than ever. In addition to causing system downtime, power outages and load shedding disrupt cooling systems. Data center temperatures can rise up to three degrees per minute, quickly causing thermal shutdown. Data centers will need to diversify their power sources, using microgrids to seamlessly transition between them.
Regulatory Requirements
In addition to moratoriums on new construction, data center developers and operators face an increasingly complex array of federal, state and local regulations. Air quality regulations place limits on emissions, with the EPA limiting the use of emergency generators to 200 hours per year. In some areas, low-emissions fuel is required along with Best Available Control Technologies such as filters or scrubbers.
Noise and water use are also facing regulatory scrutiny. Most communities limit noise to 50dB to 60dB, and data center vent fans and onsite generators can exceed those levels. New data center development may require a site assessment to determine environmental impact, including water usage. Data centers need to plan for these requirements, stay abreast of evolving regulations and develop a community engagement strategy.
AI’s Impact on Data Center Design
AI workloads are not monolithic. Development is divided into three phases, each with different power demands. The training phase requires the most compute power and thus consumes the most energy. Multiple training cycles may be needed to fine-tune the model, further increasing power demands. The inference phase is the second most power-intensive, requiring significant power to generate responses to user queries. Large AI models with millions of users consume vast amounts of power for inference.
Smaller-scale AI models designed for specific industries or businesses require much less power. The models have far fewer parameters and less data is required to train them. They also have a smaller number of users and queries.
In short, data center developers and operators will need to carefully evaluate the nature of AI workloads to design their facilities. High density will be required to minimize latency, but the power and thermal management requirements will vary based on the phase of the AI development process and the scale of the model.
Forging a Path Ahead
Most of the “rules of thumb” that have dominated data center design evaporate when it comes to AI. However, developers and operators don’t have to fly blind. Strategic partnerships can help them gain the insight needed to plan their facilities and modernize existing data centers to support advanced technologies.
The right partners can also help data centers navigate the complexities of obtaining sufficient power. Some developers are investing in solar farms and partnering with utilities and independent power producers to bring transmission lines to the data center. Others are experimenting with alternative power sources such as hydrogen fuel cells, geothermal energy and even small modular reactors.
Wesco has the right mix of expertise and relationships to help data centers overcome the roadblocks to growth. Our subject-matter experts can provide objective guidance in energy, cooling and other disciplines. We also have a well-developed supply chain to help data centers source the solutions they need to meet their project timelines and budgets. With Wesco as a partner, data centers can capitalize on the AI boom and become strategically positioned in the marketplace.
Alan Farrimond
Alan Farrimond is Vice President, Wesco Data Center Solutions. Alan is a seasoned data center professional with over 35 years of experience in the engineering and sales sectors. Throughout his career, Alan has held key leadership positions across the EMEA, APAC and North American regions. In his current role as VP of Data Center Solutions, he oversees global business development initiatives to expand Wesco's market presence.
To learn more, check out Wesco's new application guide.