OCP Members Tout DC Power in the Data Center to Meet Growing AI Power Demands

Why It Matters

LVDC can slash energy waste and free rack space, enabling data centers to scale to AI’s massive power needs without prohibitive cost or footprint increases.

Key Takeaways

• OCP working group released 170‑page LVDC white paper
• Nvidia predicts AI racks needing up to 1 MW power
• Sidecar design separates power from compute, enabling 400‑800 VDC
• Solid‑state transformers attract $60 M investment for higher efficiency
• Standards effort calls for industry‑wide participation to accelerate DC adoption

Pulse Analysis

Data centers have long relied on a cascade of AC‑to‑DC conversions that erode efficiency and consume valuable floor space. As AI models grow, the power per rack is projected to jump from a few kilowatts to as much as one megawatt, a scale that strains traditional architectures. By eliminating intermediate conversion steps, a pure DC distribution can cut loss percentages by up to 15 percent, translating into lower operating expenses and a smaller carbon footprint—critical factors for hyperscale operators competing on cost and sustainability.

The Open Compute Project’s new working group, featuring engineers from Google, ABB, Siemens and other digital‑infrastructure leaders, has published a comprehensive white paper that maps a path toward LVDC adoption. Central to the proposal is the “sidecar” rack design, which decouples power conversion hardware from compute modules, allowing up to 800 VDC to flow directly into servers and opening the door for future 1,500 VDC deployments. Two transformer technologies are under consideration: the proven transformer‑rectifier unit (TRU) and emerging solid‑state transformers (SSTs), the latter recently buoyed by a $60 million funding round for DG Matrix. SSTs promise superior efficiency and faster response times, though scaling them to megawatt levels remains a technical hurdle.

If industry players coalesce around open specifications, LVDC could become the de‑facto standard for high‑density AI workloads, reshaping equipment design, cooling strategies, and power‑budget planning. Early adopters stand to gain a competitive edge through reduced energy bills and increased rack density, while vendors that embed DC‑ready interfaces will capture new market share. However, widespread rollout will require coordinated standards development, supply‑chain alignment, and validation of solid‑state transformer reliability. The next iteration of the OCP white paper, slated for later this year, will likely set the tempo for this transition, signaling to investors and operators that the data‑center power paradigm is on the cusp of a major shift.