① The implementation of the high-voltage direct current (HVDC) solution has simultaneously increased the operational complexity of AI data centers; ② Industry insiders note that HVDC, liquid cooling, digital twin technology, and intelligent sensing will together form the core infrastructure of next-generation AI computing facilities.
According to The Science and Technology Innovation Board Daily on June 17 (Editor: Song Ziqiao), the surging power consumption of AI data centers is accelerating the commercial deployment of 800V high-voltage direct current (HVDC) power architectures.
As reported by Taiwan’s Industrial and Commercial Times, NVIDIA’s Vera Rubin platform and Google’s next-generation AI data centers will be among the first to adopt this solution, potentially triggering a new wave of AI infrastructure upgrades. The supply chain anticipates small-volume shipments of related products to commence in the third quarter, with companies such as Delta Electronics, PTK Precision Technology, and Songchuan Precision expected to benefit first, leading to an early release of market opportunities across the HVDC ecosystem.
The rollout of the HVDC solution has also heightened the operational challenges of AI data centers. With 800V HVDC power delivery, high-density GPU racks, and liquid cooling systems operating in tandem, traditional server monitoring approaches relying solely on Baseboard Management Controllers (BMCs) to track internal server status are no longer sufficient to meet operational demands.
In response, NVIDIA’s Vera Rubin DSX AI computing facility reference architecture incorporates digital twin technology to enable end-to-end simulation for planning, construction, and operations of large-scale AI data centers, with collaboration from partners including Schneider Electric, Eaton, Siemens, and Vertiv.
Industry analysts observe that competition among AI data centers has expanded beyond GPUs and server hardware to encompass overall energy efficiency and operational management capabilities. HVDC, liquid cooling, digital twins, and intelligent sensing will collectively constitute the foundational infrastructure of next-generation AI computing facilities. As the 800V HVDC power solution gains broader adoption, the entire supporting ecosystem—including power equipment, energy management, and intelligent operations—will see expanded market potential, driving a new cycle of AI infrastructure upgrades.
Currently, AI server power consumption continues to rise sharply, with single-rack power densities rapidly escalating from tens of kilowatts to hundreds of kilowatts—and even megawatt levels—pushing traditional alternating current (AC) power architectures toward their efficiency limits.
Guosheng Securities recently noted in a research report that 800V HVDC—based on the principle of ‘higher voltage, lower current’—has become inevitable: it offers approximately 5% higher end-to-end efficiency and around 70% lower maintenance costs compared to 54V systems, and provides greater scalability than 400V intermediate platforms. Coupled with NVIDIA’s 800V supplier alliance and established ecosystem players like Vertiv and Schneider Electric, mass adoption is expected around 2027.
The firm estimates that global AI data centers (AIDCs) will add approximately 15/30/40/50/60 GW of new power capacity annually from 2026 to 2030. Assuming HVDC penetration increases from 20% to 70% over this period, the corresponding annual新增 HVDC-covered capacity would be roughly 3.0/12.0/22.0/32.5/42.0 GW.