The rise of AI is redefining every aspect of the data center – from design and build to ongoing operations. Rising rack densities and space densification is reshaping infrastructure strategies and driving demand for unprecedented volumes of power.

To manage these new parameters, rethinking where power is sourced, how it’s managed, and anticipating the demand, is essential for supporting intensive AI workloads. The best approaches will be determined as a result of close collaboration between vendors and operators, plus key players across the supply chain.

With this in mind, in this episode of DCD>Talks, Kyle Keeper, SVP of power management at Vertiv shares how the company is enabling data centers to keep pace with rising power demands by championing strong relationships and expert forward thinking.

Changing power

The unit of compute has evolved beyond individual processors or servers to encompass the full stack of complete, end-to-end systems. Supporting AI and high-performance computing (HPC) has driven the rise of GPUs, which operate at levels that demand a new approach to data center infrastructure.

These systems are driving the exponential rise in both rack and spatial densities, forcing a major evolution in how power is delivered and managed.

“The most notable change we're seeing on the power side is in and around the actual rows and racks, and the high-density GPU clusters themselves,” says Keeper. “Just the nature of the GPU power densities and space density is forcing a rethinking of how power is delivered to the cluster.”

Voltage levels are increasing rapidly, with power delivery extending beyond the rack into side cars, enabling tighter coupling with advanced energy management systems built to handle the unique demands of AI workloads.

When assessing the overall power infrastructure, it raises a central question: What must be put in place so that, as the GPU refresh cycle unfolds, and as clusters and pods expand, the entire system can be re-architected and scaled to support it effectively? Addressing this challenge is essential for operators looking to keep pace in the era of power-hungry compute.

Managing fluctuating workloads

AI workloads are not just denser and heavier, they’re unpredictable. While conventional workloads often follow set usage patterns, AI-driven processes involve complex algorithms that adapt in real-time. This results in inconsistent usage spikes, particularly during phases like training, inference, or exposure to new datasets.

For data center operators, managing these dynamic and resource-hungry shifts presents an urgent and complex challenge – especially as most legacy facilities were never built to handle them. Fortunately, Vertiv has long anticipated this transition.

“We saw our first early views of these workloads a few years ago, even before it was formally public,” says Keeper. “We had some customers that came to us and said, ‘here's what we're seeing in these workloads that we're deploying – help us understand what the implications are going to be on the overall power system.’”

Understanding the implications means evaluating the entire power system – not just its overall capacity, but its responsiveness to sudden peaks and extended troughs. It's about analyzing whether the system can withstand the unique stressors that AI workloads present and determining whether they risk failure in the devices and equipment that support them.

To support its customers, Vertiv has gone a step further by working with them directly to test third-party components – whether integrated into Vertiv’s own products or into the operators’ broader systems. This collaborative approach has helped reveal the real-world implications of different configurations and power loads, providing valuable insights.

Keeper emphasizes a key distinction: different GPU vendors deliver vastly different performance characteristics. While the overall trend of spiking workloads remains consistent, the durations, amplitudes, and system effects differ depending on what's being deployed. This means there is no one-size-fits-all approach.

More power

The proliferation of AI workloads has intensified one universal concern across the industry: Where will the power come from, and can it be delivered at the scale and speed required? As the technology accelerates, this question has become the focal point for data center operators and infrastructure suppliers alike.

Crucially, power availability is now a top-tier consideration when selecting new data center sites. For years, the dominant factors in site selection included network connectivity and land availability. While these remain important, today’s energy-intensive workloads have elevated power availability to the forefront of the planning process.

“A few things are important to understand about this overarching industry problem,” says Keeper. “It’s a very local issue. We see very different dynamics depending on the region of the world in which the customer is operating. And then even within a country or geography, it can be incredibly specific – down to a city or municipality.”

This localized nature of power availability is pushing operators into new markets and previously unexplored geographies previously overlooked – because they offer access to scalable, reliable power.

But the solution isn’t as simple as breaking ground in power-rich locations. The ability to effectively and efficiently manage that power is just as critical as access. Operators must understand the intricacies of local grid constraints, especially during periods of high demand.

Keeper explains: “Beyond just waiting for the grid interconnect to come online, the power availability problem really is related to the peaks involved. There’s generally plenty of power capacity available. The issue is when everyone needs it at once – during peak seasons, there’s a real capacity limit.”

Power on demand

To address these complex power challenges, Vertiv is increasingly guiding customers toward strategies that include supplying their own power. On-site power generation, once considered purely a backup or fail-safe measure, is now becoming a popular go-to solution.

“We're seeing a move more toward this idea of always-on, on-site generation,” says Keeper. “Even if it isn't the full capacity needed to support a site, we ask, ‘is there a possibility to deploy enough to make a meaningful difference?’ So, we're exploring a lot of different technologies that could potentially be applied here.”

These technologies range from traditional fuel-based generators to cutting-edge alternatives like hydrogen fuel cells, microgrids, and modular nuclear options. Vertiv helps customers evaluate whether such on-site strategies are right for them, and if so, how and when to deploy them most effectively, based on individual needs.

This approach demands foresight and strategic planning. It’s not about reacting to today’s power shortages, it's about anticipating tomorrow’s requirements. By encouraging operators to think ahead, Vertiv is helping them prepare their facilities to accommodate surging demand long before it becomes a bottleneck.

Assessing technology readiness

As part of this forward-looking strategy, Vertiv has developed a testing and demonstration facility in Delaware that showcases cutting-edge power technologies in real-world configurations.

The site serves as a valuable resource for operators, allowing them to witness firsthand how various solutions perform under realistic conditions.

At the facility, Vertiv customers can observe fuel cells directly coupled to UPS, and examine a battery energy storage system integrated with a photovoltaic solar array. These installations are designed to help customers understand how emerging energy technologies can be applied to support critical applications within future-ready data centers.

“We're looking very closely at technology readiness – what can be deployed at scale, and what is the right timing for that,” says Keeper. “We’re considering the building block sizes needed for modular deployment as sites and campuses continue expanding their total power requirements. All of these factors determine what technologies might make sense, and when.”

Vertiv’s hands-on approach to testing ensures that its customers are one step ahead. By staying at the forefront of innovation and application, the company is helping bridge the critical gap between where the industry currently stands and where it must be to support the high-powered, dynamic workloads of tomorrow.

As Keeper concludes, Vertiv is not just responding to these key changes – it’s steering them. By helping customers anticipate power demand, customize infrastructure, and explore new technologies, the company is actively powering the future of data centers in the AI age.

To hear more about how Vertiv is supporting operators with innovative approaches to sourcing and managing power in ways that work for them, watch the full DCD>Talks episode with Kyle Keeper, here.