AC, DC or Delusion? Part 1: Lessons from the Last Power Revolution

I have a confession. When I agreed to speak at DCR Live 2026 on the future of AC and DC power architecture, I only had about a week to pull the presentation together.

That turned out to be a good thing.

With too much time, it would have been tempting to make the topic overly polished and corporate. Instead, I found myself going back through the history of the debate, through some of my own experience, and through the very practical engineering questions now being forced on the data center industry by AI.

And the more I looked at it, the more obvious one thing became: this is not really a debate about whether AC or DC “wins”.

It is a debate about how we build the next generation of digital infrastructure when power is constrained, rack densities are rising, and the useful output of a data center is no longer measured simply in megawatts, but in how much compute those megawatts can actually produce.

Or, to put it more bluntly: how many useful tokens per second can we get from every watt?

We have seen this debate before

The AC-versus-DC debate is not new. In fact, it is much older than the data center industry.

The original “War of Currents” between Edison and Westinghouse was settled more than a century ago. AC became the dominant grid architecture because it solved the major problem of the time: moving power efficiently over long distances. Transformers made it practical to step voltage up for transmission and down again for local use.

But DC never went away.

It moved closer to the point of use.

Telecoms standardised around 48 VDC. Electronic systems relied on tightly controlled DC rails. Switch-mode power supplies allowed computers to take AC input and convert it into the low-voltage DC that processors, memory and storage devices actually require.

By the time I entered the data center industry in the late 1990s, the pattern was already familiar. The building ran on AC. The equipment inside ultimately ran on DC. The industry accepted the chain of conversions between the two because it was proven, safe, standardised and supported by a mature ecosystem.

I saw the DC discussion come back in the early 2000s, driven less by the aftermath of the dot-com crash and more by a growing focus on efficiency. Peter Gross, arguably the most prominent voice in our industry, has consistently aligned himself early with its major trends, including through work with Kfir Godrich on the concept of the total DC integrated data center, which they outlined in their 2005 INTELEC paper, Total DC Integrated Data Centers.

Researchers at Lawrence Berkeley National Laboratory and EPRI also explored higher-voltage DC architectures, helping to move the conversation from theoretical possibility toward practical application. Entrepreneurial ventures such as Validus DC Systems, led by Rudy Kraus, brought serious commercial attention to the idea of DC power distribution in data centers. ABB later invested in Validus, which at the time felt like another sign that the idea might be ready to move from research into mainstream deployment.

The concepts were technically interesting. Some of the arguments were sound. But the ecosystem was not ready.

Protection systems, standards, maintenance practices, equipment availability, insurance assumptions, consultant knowledge and operator familiarity were all built around AC. The electrical theory was not the problem. The industry machinery around it was.

That is the first lesson worth remembering: good ideas do not win because they are elegant. They win when the ecosystem can adopt them safely and repeatedly.