Power Semiconductor Value to Nearly Double with Nvidia's Vera Rubin
The value of power semiconductors in Nvidia's next-generation AI server platform 'Vera Rubin' (VR200) is projected to be nearly double that of its predecessor (GB300). Power semiconductors are components that reliably supply power to servers. According to an 'AI Infrastructure Market Trends' report released by market research firm TrendForce on the 24th, the sharp rise in power semiconductor value stems from the platform's architecture. Vera Rubin is a platform that bundles 72 Rubin GPUs and 36 Vera central processing units (CPUs), along with next-generation networking equipment, into a single rack-scale system, developed for agentic AI and large-scale inference.
As AI server compute performance has increased, the importance of components that stably supply and convert power within servers has grown. TrendForce analyzed that expanding AI server demand has lengthened power semiconductor lead times once again and sent prices back on an upward trajectory. As customers secure production capacity in advance through long-term supply agreements, suppliers' pricing leverage is also strengthening.
Global players are responding in kind. ON Semiconductor is addressing rising demand by boosting utilization of existing facilities rather than building new large-scale fabs, while pursuing product price increases. STMicroelectronics is strengthening its product lineup to address the high-voltage power architectures of AI data centers. U.S.-based Vicor, facing growing production capacity burdens, is reportedly pursuing the development of two new production facilities simultaneously. TrendForce found that lead times for some power semiconductors have stretched to as long as 52 weeks, and Infineon has reportedly raised prices this year on some power semiconductor products for AI servers and automotive applications.
As power consumption rises, cooling systems are also becoming more advanced. With air cooling alone increasingly unable to handle the heat generated by high-performance GPUs, liquid cooling—which dissipates heat directly using coolant—has established itself as a core piece of AI server infrastructure. According to TrendForce, liquid cooling equipment maker Auras is rapidly expanding production capacity while supplying cold plates and manifolds to major cloud operators. Components manufacturer Lotus is also ramping up shipments of quick-connect and quick-release cooling connectors.
Nvidia itself is revamping its power architecture. TrendForce expects Nvidia to support an 800-volt (V) high-voltage direct current power architecture starting with the launch of the Vera Rubin VR200, with full-scale deployment beginning from Rubin Ultra. Higher voltage reduces the current needed to deliver the same power, lowering power losses and copper usage.
Alongside power semiconductors, the role of passive components is also growing. Multilayer ceramic capacitors (MLCCs) stably regulate current in electronic circuits and reduce signal interference; since AI servers operate for long hours under high voltage and temperature, they demand higher capacity and reliability than ordinary servers. TrendForce analyzed in May that roughly 6,500 MLCCs are used in a single Nvidia GB200 board, a figure expected to rise to around 12,000 in the next-generation Rubin architecture.
As AI server power consumption and circuit complexity increase, a clear structural shift is emerging: not only power semiconductors but also MLCCs, power supply units, printed circuit boards (PCBs), and liquid cooling components are seeing greater content per server and higher performance requirements. An industry official said, "As AI server compute performance rises, how reliably power is delivered and how efficiently heat is managed determine system performance," adding, "The benefits of expanding AI investment are no longer confined to GPUs and HBM but are spreading across the entire server supply chain, including power semiconductors, MLCCs, power supply units, and liquid cooling."
