Data center industry debates whether future facilities should use AC or DC power distribution systems.

As AI transforms data centers with increasingly power-hungry GPUs and LLMs, a century-old question has resurfaced with new urgency: is alternating current (AC) or direct current (DC) the optimal choice for tomorrow's data centers? While AC has dominated commercial and data center power distribution for decades due to its ease of transmission and standardized equipment, DC power is gaining momentum as AI workloads push traditional infrastructure to its limits. According to Peter Panfil, vice president of global power at Vertiv, "DC power has been around in some data centers for about 20 years," with "400V and 800V have been utilized in UPS for ages, but what is beginning to emerge to cope with the dynamic load shifts in AI are [new] applications of DC." Both the vendor and hyperscaler communities are actively developing 400V, 800V, and 1,500V DC architectures, with full-stack DC simulations from utility to rack being conducted to validate performance under AI workloads and high-density compute scenarios.

The technical advantages of DC systems are compelling. Joshua Buzzell, data center chief architect at power management company Eaton, explains that DC offers higher efficiency and eliminates the need for AC-based UPS systems, which frees up rack space and reduces conversion losses. DC-based system designs have already been widely tested and deployed with solar photovoltaic (PV), fuel cells, and battery energy storage systems. However, several significant hurdles must be overcome before DC achieves broad adoption. Nearly every component—from transformers to breakers—must be re-engineered for DC operation, placing a major burden on electrical equipment suppliers. High-voltage DC also raises safety challenges, with arc suppression and fault isolation proving more complex than with AC systems.

Beyond technical challenges, regulatory and standardization obstacles loom large. Bill Kleyman, CEO of Apolo.us and AFCOM Data Center World chair, notes that "Safety and regulatory compliance are significant hurdles, particularly at higher DC voltages due to different arc-flash characteristics and fault-management requirements compared to AC systems." He adds that "There's also a scarcity of commercially available, off-the-shelf IT equipment that accepts native high-voltage DC inputs, leading to reliance on specialized, often custom solutions." Additionally, there is no universal standard for DC distribution in data centers, which complicates interoperability and certification.

The economic case for DC conversion, while requiring substantial upfront capital expenditure, becomes compelling when examined over the long term. According to Kleyman, "Initial capex for converting to DC can be significant, primarily due to the need for specialized infrastructure and power conversion systems," but "the long-term opex savings are notable – often in the range of 7–20% due to reduced power conversion losses and lower cooling demands." DC systems offer clear sustainability advantages: lower conversion losses equate to less wasted energy, fewer conversion stages mean less heat generated and reduced cooling needs, and DC's compatibility with solar PV and battery storage makes it easier to integrate renewables directly to DC buses and reduce long-term carbon costs.

For existing data centers, complete conversion from AC to DC is challenging but not impossible. Data center managers must replace or bypass AC UPS, transformers, and PDUs while navigating compliance and regulatory barriers, since building codes and UL certifications are still catching up. As a result, some data centers are pursuing phased migration strategies—implementing isolated DC power zones for high-density workloads or introducing DC microgrids to achieve efficiency and capacity benefits without overwhelming capital investment. However, those seeking competitive advantage in the AI era are moving beyond incremental approaches; complete DC data centers are already emerging, such as the Mt. Diablo initiative, a collaboration among Meta, Microsoft, and the Open Compute Project developing a ±400 VDC rack power distribution experiment derived from EV power infrastructure.