Direct Current and the Data Center Case Study

This is the third and final report from our visit to the Direct Current Experience Center, where we captured both the enthusiasm and the latest updates on the emergence of direct current technology as a viable and accessible solution for a wide range of applications.

You will find the full article in the March–April issue of POWERTRAIN International.

We now turn to the application of DC in data centers—the ‘crown jewel’ of stationary applications, particularly those driven by Artificial Intelligence. The core driver is, as always, energy efficiency. Even a single percentage point gain in efficiency within large-scale data centers (operating at the Gigawatt scale) translates into massive savings on electricity bills and a reduced environmental footprint. This benefit justifies the transition to direct current, despite the initial cost disadvantage.

Power Density, Direct Current and AI

The power density of a single server rack has surged from tens to hundreds of kilowatts. For AI training, it is expected to reach one Megawatt per rack in the near future, likely by 2030. This growth demands a paradigm shift in how servers and GPUs are powered. Currently, distribution is AC-based up to the server level; inside the server, converters transform single-phase AC into 54V DC for the GPUs. The next step will see the introduction of the ‘sidecar’: an external module where the AC-to-DC conversion takes place before feeding the GPUs. Looking further ahead, the entire internal distribution of the data center will transition to DC, eliminating the need for sidecars altogether.

Regarding the critical issue of energy supply, the national grid remains the primary source. Localized renewables (such as solar) are insufficient to cover the massive energy requirements of large data centers due to intermittency and the sheer scale required. As for smart grids, they would require a battery energy storage system of unsustainable capacity.