Innoscience has announced its collaboration with NVIDIA to support the implementation of the 800 VDC rack power architecture, a major breakthrough in power distribution for AI data centers. Similar to the transition from 400 V to 800 V systems in electric vehicles, this new architecture dramatically increases efficiency, power density, and sustainability by reducing current flow sixteenfold compared to traditional 48 V systems. The result is significantly lower resistive losses, reduced copper consumption, and lower overall CO₂ emissions.

Current 48 V data center systems are approaching physical and thermal limits, with high copper demand and nearly half of total power lost to heat dissipation. The 800 VDC system overcomes these constraints, enabling scalability from kilowatt to megawatt levels for future AI clusters with hundreds of GPUs. However, achieving such high power density and efficiency—especially converting from 800 V down to 1 V—requires advanced semiconductor technologies such as gallium nitride (GaN).

Innoscience’s third-generation GaN devices are designed to meet the demanding requirements of this new power paradigm. Operating at switching frequencies near 1 MHz, these devices significantly reduce the size of magnetic and capacitive components while improving overall system efficiency.

Key advantages of Innoscience’s third-generation GaN include:

  • Up to 80% reduction in driver losses and 50% lower switching losses compared to SiC devices, yielding a 10% decrease in total system power loss.
  • Twice the power density on the 54 V output side, requiring only 16 GaN devices instead of 32 Si MOSFETs for equivalent conduction performance.
  • 70% reduction in switching losses and a 40% increase in power output compared to silicon-based architectures.
  • Scalability to support next-generation GPU platforms with improved dynamic response and reduced capacitor costs.

As the world’s only integrated GaN manufacturer (IDM) producing devices from 1200 V to 15 V, Innoscience offers a complete GaN-based conversion chain from 800 V to 1 V, delivering an all-GaN power solution across every stage of the data center power architecture.

Innoscience’s GaN devices are also proven for long-term reliability, passing extended high-temperature and stress tests, including 2000-hour dynamic HTOL and 175°C endurance validation. These tests confirm datacenter-grade durability with expected lifetimes exceeding 20 years.

With its third-generation GaN technology and collaboration with NVIDIA, Innoscience is helping drive the shift from kilowatt to megawatt AI racks—ushering in a new era of ultra-efficient, high-performance, and sustainable AI computing infrastructure.

Original – Innoscience Technology