The growth of artificial intelligence has spurred demand for larger data centers, leading to the rise of multigigawatt campuses worldwide. 

As data centers require more power than ever, the industry conversation is expanding beyond securing enough capacity to ensure that power is reliable, stable and predictable. 

AI factories — a rapidly growing segment of data centers that process AI data into intelligence — need substantial computing power to train and run AI algorithms.

However, operating AI infrastructure at this scale can introduce new stability challenges, including large, rapid load swings and dynamic interactions with the grid. This can stress the infrastructure and compromise operational reliability, said Pietro Serra, global product manager at Hitachi Energy, a global leader in electrification. 

“AI has changed the nature of electricity demand,” Serra said. “We can have hundreds of megawatts changing the AI factory’s power system within tens of milliseconds, which is quite fast. The grid can support such loads, but maintaining reliability and quality of service becomes increasingly challenging.”

To support the continued growth of AI factories, developers and operators need to incorporate a power stabilization strategy into their infrastructure. Serra said power stabilization “transforms available capacity into usable capacity” to help AI factories scale, while protecting uptime and compute productivity. It also helps manage grid impacts associated with large, dynamic loads.

Power stabilization plays a critical role in balancing the operational requirements of AI factories with the reliability needs of the electrical grid. AI factories need high-quality power to maximize their computing performance, while grid operators must maintain reliable and stable service for all customers.

Susan McLeod, vice president of data center market development at Hitachi Energy, said power stabilization is becoming a “foundational enabler” of the AI era.

“The goal is to develop a balanced solution that enables rapid AI growth while maintaining reliable and predictable power,” she said.

Power stabilization should be considered early in the planning and design phases of an infrastructure project, McLeod said. Developing an effective strategy requires early coordination among AI factory developers, operators and grid operators. 

It begins with understanding both the facility’s power requirements and the characteristics of the grid serving the proposed site, as these factors can significantly influence the appropriate power stabilization strategy.

Even as AI infrastructure becomes more standardized, each site must be evaluated within the context of the electrical system that will support it, McLeod said.

Power system studies provide insight into how an AI factory will interact with the grid, identify potential reliability and stability concerns, and evaluate system performance under a range of operating conditions before the facility is built, Serra said. 

“Power system studies translate early coordination into informed engineering decisions,” McLeod said. “Coordinating the parties without doing this technical step is just a conversation about what can be done. Power studies give both sides a shared, quantified basis for making design and interconnection decisions.”

Serra said power stabilization strategies should be based on a coordinated stability architecture rather than any single product or technology. Depending on site-specific requirements and grid conditions, stabilization strategies may incorporate a combination of battery energy storage, power-quality solutions such as static synchronous compensators, also known as STATCOMs, advanced automation systems and digital controls. 

Hitachi Energy works alongside AI factories, utilities and transmission system operators, or TSOs, to support early planning, modeling and design.

Flexible design and architecture, as well as accounting for eventual changes in AI workloads, cooling technologies, rack densities and operating models, can keep power running predictably, McLeod said. 

Incorporating power stabilization methods is easier and less costly when done early rather than redesigning the infrastructure or deploying additional mitigation measures after critical design decisions have been made, Serra said. 

Failing to consider these solutions during the design phase can also impact companies’ operations and bottom lines, McLeod said. 

“Outages can happen based on milliseconds, which, in the data center and IT space, has a huge impact on revenue and brand reputation, especially for their end customers, such as industries like finance and healthcare,” she said. 

Looking to the future, Serra said a few things are evident: Power demand will continue to increase, both real-time and advanced modeling will remain important to ensure reliable operations as facilities grow, and AI infrastructure may become more integrated with the electrical system.

“Right now, we’re seeing large loads and power behaving in an uncontrolled way, but if we can stabilize those interactions and keep the system flexible, what we're seeing today as a grid challenge can become an opportunity,” Serra said.

This article was produced in collaboration between Hitachi Energy and Studio B. Bisnow news staff was not involved in the production of this content.

Studio B is Bisnow’s in-house content and design studio. To learn more about how Studio B can help your team, reach out to studio@bisnow.com.