Nvidia's 800V Power Revolution: Opportunity or Threat for Korean Chipmakers?

A voltage revolution is quietly reshaping the data centre industry. Just as carmakers migrated from 400V to 800V electrical systems to speed up electric-vehicle charging, artificial intelligence infrastructure is undergoing an analogous shift—one with sweeping implications for the global supply chain.

On 25th June 2026, the market research firm TrendForce published a report forecasting that Nvidia will introduce 800V high-voltage power racks as an optional configuration in its next-generation AI server platform, Vera Rubin, before making them the de facto standard in the subsequent generation, Rubin Ultra.

Why raise the voltage?

A power rack is, in essence, the electrical distribution hub for a dense array of AI servers. Most data centres today operate on 48V or low-hundreds-of-volts systems. But as the GPUs (graphics processing units) that drive AI workloads have grown more powerful, their energy appetite has ballooned. A single cutting-edge Nvidia AI chip can consume several hundred watts; pack dozens of them into one rack and total rack-level power demand easily exceeds 100kW.

At lower voltages, delivering that much power requires thicker, more numerous cables, and a greater share of energy is lost as heat along the way. Raising the voltage solves both problems simultaneously: thinner cables suffice, transmission losses fall, and overall efficiency improves. That is the central logic of the 800V transition.

A platform shift, not merely a wiring upgrade

Vera Rubin is Nvidia's forthcoming AI computing platform, integrating GPUs and CPUs in a unified architecture. Rubin Ultra is the generation that follows. TrendForce's analysis suggests the 800V standard will arrive in two distinct waves—optional in Vera Rubin, mandatory in Rubin Ultra—implying a gradual but irreversible transition.

The consequences extend well beyond the power rack itself. Power supply units, transformers, cabling materials, cooling systems, and connectors will all require redesign. Once Nvidia sets this benchmark, the hyperscalers that build and operate the world's largest data centres—Google, Microsoft, Amazon—will have little choice but to adopt the same specifications in new and retrofitted facilities. In effect, the global standard for AI infrastructure is being rewritten.

What this means for South Korean industry

*Power semiconductors: a significant opening.* Widespread adoption of 800V systems will drive demand for power semiconductors capable of handling high-voltage, high-current environments—particularly those based on silicon carbide (SiC) and gallium nitride (GaN). At present, this market is dominated by foreign incumbents: Infineon of Germany, onsemi of the United States, and STMicroelectronics of Europe. Should Samsung Electronics' foundry division or DB Hitek build credible contract-manufacturing capabilities in SiC and GaN power devices, a meaningful opportunity to enter this market could open up.

*Memory: an indirect tailwind.* The 800V transition does not directly affect the high-bandwidth memory (HBM) businesses of SK Hynix and Samsung. However, greater power efficiency at the data centre level should accelerate capacity expansion, which in turn increases the number of AI accelerators deployed—and the HBM chips stacked atop them. The underlying demand driver for HBM thus grows stronger.

*PCBs, connectors, and cables: act now or risk exclusion.* High-voltage environments render many existing low-voltage components unusable. Substrates, connectors, and cables with enhanced dielectric strength will be essential. Korean printed circuit board and components makers—among them Daeduck Electronics and Samsung Electro-Mechanics—must identify the new specifications early and pursue the necessary certifications. Suppliers that fail to qualify before standards harden risk being designed out of the supply chain entirely.

*Thermal management: a niche opportunity.* Denser, higher-voltage power systems generate more heat. Demand for advanced cooling technologies—including liquid immersion cooling—is expected to rise in tandem, offering Korean thermal materials and equipment firms a potential foothold.

The window is open, but not indefinitely

History suggests that standard transitions are precisely the moments when new industry leaders emerge. The shift to smartphones displaced entrenched mobile-phone component suppliers and elevated entirely new ones. The move to 800V AI power architecture offers South Korean firms a comparable chance to close the gap on established leaders.

Yet the opportunity is conditional on preparation. Taking TrendForce's roadmap at face value—optional in Vera Rubin, standard in Rubin Ultra—meaningful volume demand looks set to materialise from 2027 to 2028 at the earliest. For Korean companies to secure technology certifications and embed themselves in the relevant supply chains by that point, the work must begin immediately.