INNOSCIENCE (02577) announced that NVIDIA will support the 800 VDC power architecture. The 800 VDC rack power architecture brings breakthrough progress to artificial intelligence data centers, enabling higher efficiency and power density while reducing energy consumption and carbon dioxide emissions. Similar to the electric vehicle industry's upgrade from 400V to 800V, increasing rack voltage from 48V to 800V can reduce current by 16 times, significantly reducing I²R losses and minimizing copper material requirements. INNOSCIENCE is collaborating with NVIDIA to jointly support the 800 VDC power architecture, providing assurance for the expansion of next-generation GPU roadmaps.
Traditional AI systems based on 48V voltage are facing severe challenges—low efficiency, excessive copper consumption, with over 45% of total power consumption spent on cooling. Future AI clusters (such as racks equipped with more than 500 GPUs) will have no space for computing units if they continue using outdated PSU power designs. The 800 VDC architecture is the solution to support system upgrades from kilowatt-scale to megawatt-scale.
Beyond transitioning to 800V rack power, this architecture also requires achieving ultra-high power density and ultra-high efficiency in voltage conversion from 800V to 1V. Only gallium nitride (GaN) power devices can simultaneously meet these stringent requirements.
To meet the power density requirements of 800 VDC, power switching frequency must be increased to nearly 1MHz to reduce the size of magnetic components and capacitors. Existing rack power supplies typically have maximum switching frequencies up to 300kHz, and increasing to 1MHz can reduce magnetic core size by approximately 50%.
INNOSCIENCE's third-generation gallium nitride technology has decisive advantages: • On the 800V input side, INNOSCIENCE's GaN reduces driving losses by 80% and switching losses by 50% per switching half-cycle compared to silicon carbide (SiC), achieving overall power consumption reduction of 10%. • At the 54V output end, only 16 INNOSCIENCE GaN devices are needed to achieve the same conduction losses as 32 silicon MOSFETs, not only doubling power density but also reducing driving losses by 90%. • Compared to silicon MOSFETs in existing rack architectures, using GaN materials in the low-voltage power conversion stage of 800 VDC can reduce switching losses by 70% and achieve 40% power output improvement in the same volume, significantly enhancing power density. • GaN-based low-voltage power stages can be scaled to support higher-power GPU models with improved dynamic response while reducing capacitor costs on circuit boards.
As the industry's only full-stack GaN supplier and leading GaN IDM company, INNOSCIENCE is the only company achieving mass production of GaN from 1200V to 15V, providing full-chain solutions from 800V to 1V. This makes INNOSCIENCE the only supplier capable of providing full GaN power solutions for all conversion stages, confidently addressing future architectural evolution to meet higher power requirements.
INNOSCIENCE's GaN also leads in reliability. Its third-generation devices have passed rigorous accelerated stress testing, including extended 2000-hour dynamic HTOL testing, high-temperature (175°C) verification, and large-sample failure verification. Self-developed online dynamic resistance monitoring and long-term board-level stress testing ensure data center-grade products have high-performance operating lifespans exceeding 20 years.
As a globally leading GaN IDM company, INNOSCIENCE's third-generation GaN devices feature excellent fast switching characteristics, high efficiency, high power density, and superior reliability. By integrating 800 VDC power architecture with INNOSCIENCE's GaN technology, AI data centers will achieve the leap from kilowatt-scale racks to megawatt-scale racks, opening a new era of more efficient, higher-performance, more reliable, and more environmentally friendly AI accelerated computing.