Turning Up the Voltage: Taiwan Drives the Global Shift to 800V Data Center Power
A wave of announcements at this year’s Open Compute Project (OCP) Global Summit has underscored the pivotal role Taiwan’s major technology providers, led by Foxconn, Delta, Lite‑On, and Wiwynn, are playing in driving the global transition from traditional 54‑volt rack architectures towards 800-volt high‑voltage direct current (HVDC) systems for data centers.
For decades, most data centers relied on 48–54 V DC power distribution within racks –a design that worked well in the early days of web infrastructure when rack power rarely exceeded 10–20 kW. But as AI training models balloon in size and computing systems approach megawatt levels per cluster, that legacy design has hit its physical and economic limits. The issue is simple physics: at lower voltages, delivering the same power requires exponentially higher current, resulting in severe resistive losses, heat buildup, and excessive copper usage. The jump from 54 V to 800 V HVDC cuts current by a factor of nearly 15, dramatically reducing power loss, cable bulk, and material costs.
According to research presented at OCP 2025, this change is already under way. GPU power consumption now exceeds 1 kW per accelerator, and next‑generation chips like Nvidia’s Blackwell and AMD’s Turin require entire racks drawing 600 kW or more. Each incremental node increase in AI complexity compounds the need for cleaner, higher‑efficiency power delivery. The 800 V DC architecture addresses these challenges directly by transmitting more energy at lower current, minimizing conversion stages, and improving total system efficiency by 3–5% across the rack. Lower cable resistance also means less heat, reducing cooling load and enabling more compact designs.
At OCP 2025, Foxconn (formally known as Hon Hai Precision Industry) and its subsidiaries made a series of announcements highlighting this shift. Foxconn unveiled a collaboration with Nvidia to implement a full 800 V DC architecture at its Kaohsiung K-1 AI Factory, which it called the blueprint for next-generation AI factories. The system integrates HVDC power with modular rack layouts and direct liquid-cooling systems, cutting cable losses and copper usage while enhancing computational density. By pairing 800 V DC delivery with advanced coolant distribution units (CDUs) and immersion systems, Foxconn expects to achieve unprecedented energy efficiency for AI workloads.
Foxconn’s affiliate, Ingrasys Technology, reinforced this direction at the same event, presenting Nvidia GB300 NVL72 and HGX B300 AI platforms configured for high-voltage environments. These systems include rack-mounted cooling modules and busbars specifically rated for 400–800 V DC operation, components that will become commonplace as AI data centers grow. Foxconn’s thermal‑management partners, such as Cooler Master and Teco, are now collaborating on electromechanical integration, combining liquid cooling, power delivery, and cable management inside unified frames.
Other Taiwan firms are aligning quickly. Delta Electronics showcased its “grid-to-chip” 800 V DC ecosystem co-designed with Nvidia, capable of delivering 1.1-MW rack clusters with over 99% conversion efficiency. Lite‑On Technology demonstrated modular 72 kW power shelves expandable to 1 MW configurations by 2026, while AcBel Polytech unveiled Ruby‑standard server power supplies with ultra‑high density and energy‑storage readiness. Delta’s reference design, a separate HVDC side‑power rack, houses power shelves, batteries, and conversion units that are liquid-cooled and built for hot-swappable deployment, emphasizing redundancy and safety at scale.
Taiwan’s suppliers enjoy a unique advantage in this shift. The island’s decades-long experience in semiconductors, power electronics, and server manufacturing gives it a complete stack, from power supplies and connectors to liquid cooling hardware and switching gear, all within a tightly knit ecosystem. With Foxconn’s Kaohsiung K-1 serving as the region’s testbed and Delta’s engineering partnerships across Nvidia’s supply chain, Taiwan is becoming the gravitational center of the HVDC transition.
The benefits extend far beyond efficiency. 800 V DC systems enable finer control of power flow and easier integration of on‑site solar, fuel‑cell, or battery‑storage systems. They simplify backup design through direct DC coupling and cut conversion losses inherent in alternating‑current (AC) systems. Importantly, the architecture complements liquid and immersion cooling by lowering heat at the electrical level, contributing directly to lower Power Usage Effectiveness (PUE) ratings, which leading facilities now target at 1.05 or below.
Suppliers such as Song Chuan Precision are even optimizing small components for this high-voltage world. The company is manufacturing 400/800-V relays, DC contactors, and fluid‑cooled safety switches for AI servers, EV charging, and energy‑storage systems — essential equipment for reliable high‑current operation. Its products, some rated at 1,500 V DC, represent the granular innovation needed to make HVDC practical and safe at scale.
The migration to 800 V DC won’t happen overnight, with industry analysts expecting broad adoption in 2027, but the direction is clear. Between rising GPU power demands, and the quest for sustainable AI infrastructure, HVDC has become inevitable. The announcements in San Jose made that official: the age of low‑voltage data centers is ending, and Taiwan’s technology ecosystem is stepping forward to power what comes next.
