Next-Gen 800VDC data center empowers AI factory

The 800VDC data center architecture emerges precisely in this context. It is not merely a voltage upgrade, but a systematic end-to-end innovation from grid to chip, from power to thermal, and from equipment to building. It uses an 800V "power artery" to connect green energy, efficient power distribution, intelligent cooling, and autonomous robots into an integrated whole, laying a solid foundation for the large-scale deployment and sustainable operation of AI factories.

The solid-state transformer is the starting point of the 800VDC architecture and the first step from traditional AC distribution toward high-voltage DC. It employs silicon carbide (SiC) power devices and high-frequency transformer technology to directly convert medium-voltage AC (10kV to 13.8kV) into 800V DC, replacing the multi-stage equipment of conventional solutions, step-down transformers, switchgear, UPS, etc., and reducing the power delivery chain from 6 conversion stages to 3. The resulting changes are tangible: end-to-end system efficiency jumps from below 90% to over 93%, and equipment footprint is reduced by approximately 50%, freeing up valuable space for computing cores such as GPUs. The SST is not just a power supply device; it is a digital energy hub that integrates intelligent monitoring and dynamic response, paving the way for "green power supply" in AI factories.

If the SST is the "source" of the 800V architecture, then the DC busway is the "power aorta" running through the entire data center. Under the 800V architecture, the transmission current for the same power level is significantly reduced compared to conventional solutions, line losses (I²R) decrease markedly, copper usage is cut, and rack space is freed up. More importantly, this "aorta" is evolving into an integrated distribution platform. This "thermal-electrical co-design" provides a critical path to address the dual challenges of power distribution and heat dissipation for megawatt-level racks, and clears obstacles for high-density deployment in AI factories.

After power is delivered via the DC busway to the rack, the next challenge is: how to efficiently step down 800V high voltage to the ultra-low voltage (<1V) required by GPUs and CPUs? The difficulty at this stage lies in the extremely high step-down ratio - from 800V to below 1V - while carrying thousands of amperes of current in an extremely compact space. The current mainstream approach uses a two-stage conversion architecture: first, an isolated intermediate bus converter (IBC) steps 800V down to 48V or 6V, then a multi-phase step-down solution reduces it to below 1V. The efficiency and density of this stage directly determine whether AI compute power can be effectively "fed," serving as the "last mile" connecting high-voltage distribution and chip power delivery.

From the 48V or 12V bus to the GPU core is the "last mile" of the power delivery chain - and the most challenging segment. During AI training tasks, GPU power consumption fluctuates violently within milliseconds, requiring power solutions that combine ultra-fast transient response with extremely high current density (single GPU current already exceeds 1,000A). Multi-phase interleaved buck converters are the current mainstream architecture, while the trans-inductor voltage regulator (TLVR) topology represents the next-generation direction. It uses multi-winding coupled inductors to enable simultaneous response of all power phases within tens of nanoseconds, greatly reducing reliance on output capacitors. At the device level, gallium nitride (GaN), with its extremely low switching losses, is gradually replacing traditional silicon MOSFETs as the new choice for the core switching devices in GPU power delivery.

While the 800V architecture improves power delivery efficiency, it also drives a sharp rise in thermal density, with single-rack power exceeding 100kW and moving toward the megawatt level, pushing traditional air-cooling solutions toward their physical limits. Liquid cooling is transitioning from "optional" to "standard," from direct-to-chip cooling to immersion cooling, and from liquid-cooled busbars to intelligent cooling distribution units (CDUs) - the boundaries of cooling systems continue to expand. At the building automation level, digital twins and AI-driven approaches are reshaping operations and maintenance. When the cooling system is optimized end-to-end from chip to cooling tower and deeply integrated with building management systems (BMS), data center power usage effectiveness (PUE) can be expected to drop below 1.1 - this is the "green dividend" that the 800VDC architecture bestows upon AI factories.

The other side of the AI factory is a fundamental transformation in operating models. AI robots are pushing data centers from "manned" toward "unmanned autonomous operations." Autonomous mobile robots (AMRs), equipped with LiDAR, depth cameras, and SLAM algorithms, perform 24/7 equipment inspections and environmental monitoring; four-finger or five-finger robotic hands, combined with force, tactile, and visual sensors, can carry out fine tasks such as button operations, module replacement, and cable management, freeing personnel from high-voltage hazard zones. These robots are deeply integrated with building automation systems through real-time industrial networks such as EtherCAT and TSN, forming a complete closed loop from perception to decision-making to execution. AI robots are becoming the "hands" and "eyes" connecting the physical factory to the digital world.

From SST to AI robots, the six technology pillars together form the complete vision of the new-generation AI factory. 800VDC is not merely a voltage increase but a comprehensive restructuring of the entire power ecosystem - from grid access to chip power delivery, from thermal management to intelligent operations, every link is undergoing profound change. The Arrow Electronics booth at PCIM Asia 2026 is the best platform to witness this transformation.
 
Please click on the exhibition zone topics you are interested in above to learn more about the detailed technical concepts and solution introductions.