As you read this, it is 2025, and AC remains king. Over the past couple of centuries, alternating current (AC) and direct current (DC) have faced off in the legendary War of the Currents. Now, we’ve entered a new phase – the Current War, if you will – where DC is back on its feet, landing a high-impact recovery punch against AC, which has so far managed to avoid a knockout, even during DC’s 1950s resurgence.

As a spectator to today’s Current War, JP Buzzell, VP and data center chief architect at Eaton, emphasizes the urgent need for industry-wide coordination and standardization in adopting DC power for data centers.

With a storied career spanning nuclear power, building data centers for Meta (formerly Facebook), and designing and running facilities for Oracle, including the introduction of the Stargate Abilene hybridized network topology, Buzzell now leads Eaton’s strategy for overall data center architecture, supporting the broader data center industry.

But above all, he stresses that DC integration is no longer optional: it is a critical enabler for the next generation of AI workloads and high-performance computing. This perspective echoes recent industry moves, such as Nvidia’s 800 VDC architecture, which it highlights as foundational for powering the next generation of AI factories.

Legacy data centers and AC

Looking back to the era of Tesla and Edison battling it out to become power supreme, it’s clear to see why AC emerged victorious in the original War of the Currents. Simply put, AC was easier to distribute.

In the late 1880s and early 1900s, DC lacked the technical tools to efficiently step voltage up or down for long-distance transmission and safe power consumption. AC, by contrast, had transformers: simple, reliable, and highly efficient devices that made distribution practical. Today, that legacy still shapes how data centers operate:

“You bring AC all the way up to the rack, whether it’s 400 or 480 volts (V),” explains Buzzell. “The rack or server power supplies convert that to DC, typically 12V or 48V. Then, onboard voltage regulators step it down further to around one volt – the level that actually powers the chips.”

AC became the standard out of necessity, but its convenience now comes with hidden inefficiencies. As modern workloads call for greater performance and efficiency, there is a renewed interest in DC power at the core of next-generation data centers.

Danger, danger: High voltage

We are no longer living in the 1800s, nor the 1900s. Modern power electronics can now step DC voltage up or down electronically, without massive rotating machinery required.

Still, power is generated as AC at the plant, then stepped up to very high voltages – often tens or hundreds of kilovolts – to minimize energy losses during transmission. For long or subsea transmission routes, utilities rely on DC, while AC remains the standard for regional, short-distance transmission. Local substations then step AC down to voltage levels suitable for safe distribution to businesses, homes, and data centers.

But even at these levels, the power still arrives as AC, meaning data centers must convert it again to DC. Each conversion adds energy losses and heat, even though virtually all modern IT equipment already runs on DC. Servers, GPUs, batteries, and renewable sources are inherently DC-based, making repeated AC-DC conversions inefficient.

The takeaway is clear: AC may dominate transmission outside the data center, but inside, DC reigns – where it’s critical for both energy efficiency and next-generation computing.

The GPU densification problem

Air cooling served data centers well for decades, but as workloads have grown increasingly complex, network topologies have started creating performance bottlenecks – specifically, packet loss that slows down clusters.

To overcome these limits, engineers began densifying GPU deployments, packing more chips into fewer racks to reduce latency and improve performance. However, this push for density has also magnified the challenges of power delivery and cooling.

As density has increased – from eight GPUs per rack to 72 – air cooling has become insufficient, and air fans simply take up too much space. This has necessitated a return to liquid cooling using direct-to-chip methods.

Simultaneously with power conversion shifting outside the racks and DC increasingly adopted to cut losses and improve efficiency, these changes in cooling, power delivery, and network design are reshaping the foundations of modern data center architecture.

According to Buzzell, the move to DC is being driven by network topology and physical constraints, not solely by cost or sustainability considerations:

"The reason performance shifted is because Nvidia’s GV200/300 GPUs had already been pushed to their limits, along with AMD’s 325-series chips. I noticed this limitation myself when designing the Stargate computing clusters."

Something’s got to give

Racks that once drew 10-30kW each now commonly pull 120kW per rack. Pushing the limits, some reach 500kW, with one megawatt per rack just on the horizon – essentially the load of a small power plant packed into a single cabinet.

As GPU clusters for AI training grow, the network connections between racks demand an immense amount of power. The old approach via AC distribution inside the racks cannot keep up. Power delivery architecture must evolve.

“At that scale, the old AC-based distribution wastes too much heat and too much space,” says Buzzell. “New approaches using DC power delivery make it more efficient to feed these racks and take advantage of dead space, so electrical topology is shifting toward direct current.”

Buzzell highlights another challenge, which lies in coordination. Using the analogy of time zones, he explains:

“There are 38 distinct local times around the world, which means people are awake and making decisions while others are asleep. Just the simple fact that people are in different places, thinking at different times, naturally creates varied perspectives. Bringing everyone together physically is tough, and even if you do, it’s not always practical. So the real hurdle is learning how to collaborate effectively across time zones, asynchronously.”

Come together, right now, over DC

Buzzell underscores that the biggest hurdle to DC adoption isn’t technology – it’s communication. Effective collaboration requires global leaders to coordinate and standardize key aspects, such as voltage levels and safety protocols.

As such, Eaton has been actively supporting this effort. For instance, a September symposium with UL will bring together hyperscalers, original equipment manufacturers (OEMs), and other industry participants to align on DC standards. Eaton is also collaborating with Current/OS to define standards for DC microgrids.

Looking ahead, discussions with the Institute of Electrical and Electronics Engineers (IEEE), which sets widely recognized industry standards, are expected to converge with ongoing Open Compute conversations.

“We all want the same outcomes, but right now we tend to operate in silos simply because we don’t know what we don’t know,” says Buzzell. “If we can get everyone in the same room, I believe we can collaborate effectively and have a strong path to do that over the next six months.”

What will be discussed?

It’s easy to get lost in the technical details of new equipment and operational procedures, but Buzzell urges a broader perspective:

“The technical hurdle isn’t just creating the technology – it’s getting global agreement on voltage standards first.”

Without that agreement, widespread adoption of DC power is difficult. Buzzell draws lessons from history, with early trains running on differing track widths, forcing cargo to be unloaded and reloaded when moving between lines. Similarly, before the standardization of shipping containers, goods had to be manually transferred between ships, slowing trade and increasing costs.

Just as standardization solved these past industrial challenges, agreeing on DC voltage standards is essential for scaling modern data centers efficiently and sustainably.

From a safety standpoint, AC has a natural advantage, since it crosses zero volts in each cycle. This “zero-cross” moment allows current to be safely interrupted or switched, providing inherent safety and energy management benefits. DC, by contrast, is “always on” and lacks a zero-cross point, so faster interruption mechanisms, like rapid switches or specialized circuit breakers, are needed to safely control power flow.

While the technology to handle DC safely already exists, the broader success of DC adoption depends first and foremost on establishing a standard voltage. A shared standard allows devices and systems from different companies to work together seamlessly. Without a standard, components risk incompatibility, creating inefficiency, higher costs, and slower adoption.

Knowing the exact voltage also lets engineers design proper insulation, circuit breakers, and protection systems, critical for DC. If companies don’t align on voltage, each would need to develop multiple versions of the same technology, further slowing progression.

Essentially, a standardized voltage ensures that once the technology is ready, it can be deployed worldwide without redesign or incompatibility issues, making commercialization practical. Buzzell adds:

“Eaton has significant technology in this area, and from a technical standpoint, there’s a clear path forward. However, we won’t move to commercialization until we have that agreement – there’s no point in developing a product if no one else is willing to adopt it because the standard changes.”

What is cost without sustainability, and vice versa

It’s easy to separate the benefits of a product or system into different buckets, labeling them as either ‘sustainable’ or ‘cost-effective.’ But when asked about real-world trade-offs between upfront conversion costs and long-term total cost of ownership (TCO), Buzzell offers a practical reminder, connecting sustainability and cost in a business-oriented way:

“Sustainability and cost aren’t separate – when you design something to be sustainable and reusable, it reduces the energy and resources needed to create it, making it cheaper. Done right, sustainability isn’t an added burden; it adds value by extending asset life and increasing utilization, which creates a clear business case.”

In the US, the focus often leans narrowly toward cost. Incentives like tax credits have sometimes reinforced a divide between sustainability and financial efficiency. Yet, with a broader perspective, solutions can be designed to achieve both.

Europe has led the way, driven largely by sustainability goals. This policy-driven focus encourages the adoption of energy-efficient technologies, such as DC motors, which can reduce power consumption by tens of percent, according to Buzzell:

“Unlike AC motors, which rely on braking resistors that dissipate energy as heat, DC systems take advantage of counter-electromotive force – similar to how an electric car slows down and regenerates energy – making them much more efficient.”

DC implementation has already proven successful in European applications, particularly automotive factories, where Eaton’s DC technology has demonstrated measurable improvements in power utilization, supported by empirical data.

The power of three fights back

All good things come in threes, right? For Buzzell, the three key advantages of DC power in data centers are clear:

1. Supporting future compute needs

Large computing clusters, whether for AI or high-performance computing, demand massive power and connectivity. DC helps manage these clusters more efficiently, avoiding the congestion issues that current AC systems can struggle with. It also opens the door to new data center locations by reducing material use and minimizing community impact, making sites feasible where they previously weren’t.

2. Improving power efficiency

DC delivers energy in data centers more efficiently, maximizing what Buzzell calls ‘watts to tokens.’ It also makes future-proofing easier by aligning voltage levels with other industries, like solar photovoltaics, and adopting medium-voltage systems makes data centers more scalable, sustainable, and easier to manage.

3. Reducing materials and space

DC systems need fewer wires than traditional three-phase AC (two instead of three), reducing copper use and enabling more compact designs. Consolidating power components, like UPS systems and converters, further simplifies architecture. The payoff is lower operating costs, less maintenance, reduced noise and emissions, and a more sustainable facility overall.

Don’t create bottlenecks by design

Failing to properly integrate and standardize DC in AI data centers sets the stage for congestion and inefficiency. Buzzell points out that software engineers, who design for logic and algorithms, often approach problems very differently from mechanical, electrical, and plumbing (MEP) engineers, whose expertise comes from decades of facilities and power management experience:

“Both perspectives are valid, but without cross-disciplinary coordination, the result is a system where delivering electrons to chips becomes a bottleneck, limiting performance and scale.”

The stakes are high, and even small missteps, like neglecting DC optimization, can cascade into major operational headaches, preventing data centers from reaching their full potential. Ignoring proper DC design is like highway congestion: the route may appear simple, but without careful engineering, it leads straight to gridlock, inefficiency, and failure.

The next round

The next phase of the DC takeover is just around the corner. The focus is shifting to post-fight rehabilitation, with next-generation retrofit technologies, like direct current sidecars, ready to take center stage.

For Edge deployments and AI factories, the same rules apply: build facilities to last the distance. That means rethinking old-school ‘stick-built’ structures with fixed lifecycles and moving toward modular designs that can adapt, scale, and keep delivering knockout performance round after round.

Buzzell issues a PSA: stay tuned over the next six months and get ready to step into the next era of compute.

For more information about how Eaton helps its customers design, build, and operate AI-ready data centers, visit eaton.com.