Artificial intelligence is rewriting the rules of data center design, and power architecture is right at the center of the conversation. As rack densities surge and energy efficiency becomes a defining constraint, operators are navigating a transitional era. Hybrid AC and DC power systems are becoming pervasive in the near term, while higher-voltage DC architectures are emerging as the long-term solution once standards and equipment fully mature.
The Reality Today: Hybrid Is Here to Stay for Now
Most data centers were built on AC power distribution, and for good reason. AC is standardized, widely supported, and deeply embedded in utility infrastructure, UPS systems, and facility operations. But AI workloads, especially GPU-dense clusters, are exposing the inefficiencies of multiple AC-to-DC conversion steps inside servers.
To address this, many operators are adopting hybrid AC and DC approaches:
- AC at the facility level, leveraging existing utility feeds and UPS systems
- DC closer to the load, reducing conversion losses inside power supplies
- Rack-level or row-level DC distribution, particularly in high-density AI pods
This hybrid model allows data centers to incrementally support AI without replacing their entire electrical backbone. It provides a practical path forward while balancing performance, risk, and speed of deployment.
Why DC Matters More as AI Scales
AI changes the math. Traditional enterprise racks once averaged 5 to 10 kW. Today’s AI racks routinely exceed 30 to 60 kW, with roadmaps pushing well beyond that. At these densities:
- Conversion losses compound quickly
- Copper usage increases dramatically
- Thermal management becomes inseparable from power design
DC distribution, especially at higher voltages, directly addresses these challenges:
- Fewer power conversions improve overall efficiency
- Higher voltage reduces current, lowering losses and conductor size
- Better alignment with native DC loads such as GPUs and accelerators
This is why hyperscalers and silicon vendors are increasingly vocal about DC’s role in AI infrastructure.
What’s Slowing Full DC Adoption?
The benefits of DC power are well understood, but large-scale adoption depends on more than technical merit alone. Data centers operate within a highly standardized and risk-averse environment, and changes to power architecture require confidence at every level of the ecosystem.
The biggest constraint on full DC adoption is not performance. The bottleneck is predictability at scale. Data center power systems are expected to operate continuously, safely, and repeatably across thousands of deployments. Any change to the electrical model has to prove itself not just in pilots, but in long-term, real-world operations.
High-voltage DC introduces new considerations around protection, fault isolation, maintenance procedures, and interoperability between vendors. While these challenges are solvable, the industry is still building shared confidence in how DC systems behave under every operating condition, from normal load fluctuations to failure scenarios.
The Near Term: Coexistence and Choice
For the foreseeable future, hybrid AC and DC systems will dominate AI deployments. They offer:
- Faster time to market
- Lower risk during rapid AI expansion
- Compatibility with existing facilities
We will continue to see AC-based data centers with localized DC distribution, modular power blocks designed specifically for AI halls, and power and cooling planned together as a single, integrated system. Vendors such as Schneider Electric are already delivering solutions that support this hybrid reality while preparing operators for what comes next.
Final Thought: Designing for What Comes After Today
Powering AI infrastructure is not about committing to a single architecture overnight. It is about designing systems that can evolve as workloads, densities, and technologies continue to accelerate. Hybrid AC and DC power t enables scale today while creating a clear on-ramp to future architectures.
As standards mature and high-voltage DC equipment becomes more widely available, the shift toward DC will happen naturally, driven by efficiency gains, simplified power paths, and the physical realities of ultra-dense AI racks. The data centers that succeed will be the ones making forward-looking decisions now, selecting power architectures that support flexibility, modularity, and long-term adaptability