INNOSCIENCE provides a full GaN power solution for 800 VDC power architecture, empowering the next generation of AI Factories

According to the announcement from INNOSCIENCE (02577), NVIDIA will support the 800 VDC power architecture. The 800 VDC rack power architecture brings breakthrough advancements to artificial intelligence data centers, enabling higher efficiency, higher power density, while reducing energy consumption and carbon dioxide emissions. Similar to the upgrade from 400V to 800V in the electric vehicle industry, increasing the rack voltage from 48V to 800V can reduce the current by 16 times, significantly decreasing I²R losses and minimizing the demand for copper. INNOSCIENCE is collaborating with NVIDIA to jointly support the 800 VDC power architecture, ensuring the expansion of the next-generation GPU roadmap. Traditional artificial intelligence systems based on 48V voltage are facing severe challenges—inefficiency and excessive copper consumption, with over 45% of total power consumption spent on heat dissipation. Future AI clusters (such as racks equipped with more than 500 GPUs) will have no space to accommodate computing units if they continue to use outdated PSU power designs. The 800 VDC architecture is the solution to support the system's leap from kilowatt to megawatt levels. In addition to transitioning to 800V rack power, this architecture also requires achieving ultra-high power density and ultra-high efficiency in the voltage conversion from 800V to 1V. Only Gallium Nitride (GaN) power devices can meet these stringent requirements simultaneously. To meet the power density requirements of 800 VDC, the power switch frequency must be increased to nearly 1MHz to reduce the size of magnetic components and capacitors. The typical switching frequency of existing rack power supplies can reach up to 300kHz, and increasing it to 1MHz can reduce the core size by about 50%. INNOSCIENCE's third-generation Gallium Nitride technology has decisive advantages: • On the 800V input side, INNOSCIENCE's GaN can reduce driving losses by 80% and switching losses by 50% per switching half-cycle compared to Silicon Carbide (SiC), achieving an overall power consumption reduction of 10%. • At the 54V output, only 16 INNOSCIENCE GaN devices are needed to achieve the same conduction losses as 32 Silicon MOSFETs, not only doubling the power density but also reducing driving losses by 90%. • Compared to Silicon MOSFETs in existing rack architectures, the low-voltage power conversion stage of 800 VDC using GaN material can reduce switching losses by 70% and achieve a 40% increase in power output within the same volume, significantly enhancing power density. • The low-voltage power stage based on GaN can be scaled to support higher power GPU models, with improved dynamic response while reducing capacitor costs on the circuit board. As the only full-stack GaN supplier in the industry and a leading GaN IDM company, INNOSCIENCE is the only company to achieve mass production of GaN from 1200V to 15V, providing a full-link solution from 800V to 1V This makes INNOSCIENCE the only supplier capable of providing full GaN power solutions for all conversion stages, confidently addressing the evolution of future architectures to meet higher power demands. INNOSCIENCE's gallium nitride also leads in reliability. Its third-generation devices have passed rigorous accelerated stress tests, including an extended 2000-hour dynamic HTOL test, high-temperature (175°C) validation, and large-sample failure validation. The independently developed online dynamic resistance monitoring and long-term board-level stress testing ensure that its data center-grade products have a high-performance operational lifespan exceeding 20 years. As a global leader in gallium nitride IDM, INNOSCIENCE's third-generation gallium nitride devices feature excellent fast-switching characteristics, high efficiency, high power density, and outstanding reliability. By integrating 800 VDC power architecture with INNOSCIENCE's gallium nitride technology, artificial intelligence data centers will achieve a leap from kilowatt-level racks to megawatt-level racks, ushering in a new era of more efficient, higher performance, more reliable, and more environmentally friendly artificial intelligence accelerated computing