I recently joined a panel at the Datacloud Global Congress in Cannes to debate a question that’s been building for some time: is AC or DC power better suited for tomorrow’s high-density computing needs?
Of course, the answer is it depends… but what struck me was how far the conversation has moved. Not so long ago DC was still largely niche in data center circles — an interesting idea, but not yet urgent, the preserve of the telco community. Today, the industry is working hard to get aligned on a future. The debate is no longer whether DC is the right direction. The questions now are: how widely will DC be adopted, how fast will it come, and can we overcome the challenges quickly enough to meet the demand?
The logic for DC is now well established, so let me summarize it briefly. Today’s conventional AC path from grid to chip involves several conversion stages — utility transformer, UPS double conversion, PDU step-down, and server PSU — each one losing energy. With DC technology, there will be an energy efficiency benefit, while small in percentage this could be meaningful when data centers scale to several hundred megawatts.
Beyond efficiency, DC systems require up to 50% less copper, according to DC-Industry project. At 800 V DC, the same wire gauge carries 15% more power than at 415 V AC. DC also simplifies the integration of renewables and battery storage by eliminating additional AC conversion stages — a structural advantage for operators with carbon reduction commitments. And with fewer components and no phase-balancing equipment, DC architectures reduce complexity and minimize potential failure points.
But here’s the thing: knowing where you need to go and being ready to get there are two different challenges.
What we need to solve now
We expect DC power infrastructure to be needed within the next 2-3 years. To do this, we need to solve hard engineering and ecosystem problems, and we need to start now.
Regulation and standards
DC protection standards already exist in the IEC market and are evolving quickly. IEC 60947-10, for example, provides an international framework for semiconductor-based circuit breakers. The direction is clear: toward greater harmonization, regulatory clarity, and a consistent safety framework. But momentum now needs to translate into larger scale. As North America is the highest-growth market for data centers, comparable UL standards are in process and urgently needed to accelerate the transition. Establishing widely accepted standards will be critical to give operators the confidence to invest and to ensure interoperable DC infrastructures.
Supply chain readiness
While 800 V DC components exist, a full DC data center supply chain, from switchgear to busbars to monitoring, is still being built out. Moving from proof of concept to hardened, mature products with economic manufacturability is not trivial. Capacity will scale to meet demand, but this will take some years.
Operations, safety, and risk
The DC conversation tended to focus on efficiency. Now we talk about the operational implications. The global electrical workforce has been trained on AC for decades. DC demands new competencies — in design, installation, commissioning and maintenance. Arc behavior, fault characteristics, isolation requirements, they all change. And therefore, protection also needs to change.
This isn’t just a technology transition; it’s also an operational one. And it demands infrastructure that’s orchestrated as one system, not managed in silos. That’s the challenge we’re solving with our customers every day.
What could be a realistic transition?
The move to DC will not happen overnight, but it has already started. We are likely to see a staged migration over the next five years, with operators adopting hybrid approaches as technologies mature and standards evolve. Operators and investors are trying to see how to intercept this future trend without risking todays investment, this is not trivial and really poses the question – what future should I build for today?
The initial phase is already underway with what we describe as the DC sidecar — localized AC-to-800 V DC conversion racks placed adjacent to IT racks. These allow operators to begin increasing power density while leveraging mature 800 V DC supply chains already established in the EV charging industry.
The North Star will be centralized DC distribution, and we are looking at 2028–29 for these deployments. This involves 800 V DC busway running across the data hall, with plug-in rack feeds. It enables higher-density deployments and reduces per-rack conversion overhead. We will see Medium Voltage Solid-State Transformers (SST) converting directly from MVAC to LVDC at the medium-voltage level, with 800 V DC distribution throughout the entire low-voltage system. This is the architecture that enables maximum efficiency and full renewable integration.
Technology and know-how already exist
This transition requires deep domain expertise across the full electrical value chain — from medium-voltage grid connection to low-voltage distribution in the data hall. And it requires the ability to innovate at the component level while designing at the system level.
That’s exactly where Siemens sits. DC is not new to us. We’ve been engineering DC power systems for decades in the maritime industry, e-mobility and industrial campuses. The data center is the next frontier, but it builds on a foundation we’ve been laying for a long time.
We recently announced a collaboration with Infineon to advance semiconductor circuit breaker technology, one of the critical enablers for DC grids. Our new SENTRON 3QD2 semiconductor circuit breaker (SCCB), using Infineon’s silicon carbide power modules, delivers ultra-fast fault interruption in the microsecond range, up to 1,000 times faster than conventional mechanical breakers. This capability is essential for DC protection, where traditional breakers struggle with the physics of DC fault currents. In AI data centers, where even a slight delay can cause costly downtime, data loss or expensive hardware damage, that speed is not a nice-to-have. It’s a necessity.
This is one example of how we’re actively building the technologies the industry needs for the DC transition — not waiting for the future to arrive but engineering it.
DC will not replace AC tomorrow. But it is clearly becoming the destination architecture where high-density computing is required, like in AI factories.
We’re getting ready.