Delta Electronics has introduced a new 800VDC modular data center architecture designed specifically for NVIDIA's upcoming Vera Rubin AI platforms. This high-voltage direct current system aims to significantly improve power efficiency, reduce conversion losses, and support the extreme power density required by next-generation AI data centers.
Overview of Delta Electronics' DCW Asia 2026 Unveiling
At Data Center World (DCW) Asia 2026, held on September 29 and 30 at the Marina Bay Sands Convention Centre in Singapore, Delta Electronics introduced a comprehensive suite of infrastructure solutions designed to meet the escalating demands of artificial intelligence. The centerpiece of the unveiling was the new AI Modular Data Center (AI MDC) and Microgrid Solutions. Engineered specifically to support the upcoming NVIDIA Vera Rubin architecture, these systems represent a significant leap in power density, efficiency, and thermal management. By integrating next-generation 800VDC In-Row Power with advanced liquid cooling, Delta Electronics is positioning its infrastructure to serve as the foundational backbone for the AI factories of the future.
Technical Details of the 800VDC Power Architecture
The transition to higher voltage direct current is a critical evolution in data center power distribution, minimizing conversion losses and maximizing efficiency. Delta’s newly showcased 800VDC In-Row Power architecture is a prime example of this paradigm shift. The system delivers a robust 660kW of output power while achieving an impressive AC-DC efficiency of up to 98%. To ensure uninterrupted operations during peak computational loads, the architecture incorporates battery backup units (BBU) offering 480kW of capacity. Furthermore, the integration of aluminum capacitors is specifically utilized to optimize GPU workloads, ensuring that the extreme power transients characteristic of next-generation AI data centers are managed seamlessly and safely.
Advanced Thermal Management and Liquid Cooling
As AI compute densities scale exponentially, traditional air cooling becomes insufficient. Addressing this, Delta Electronics has deployed a high-voltage direct current (HVDC) 3.6MW in-row coolant distribution unit (CDU) designed to handle the intense thermal demands of high-density AI workloads. At the component level, the company has developed a dedicated cold plate specifically for the NVIDIA Vera Rubin NVL72. This cold plate incorporates a sophisticated microchannel and multi-impinging flow design, providing highly targeted thermal management directly at the heat source. Through the integration of an advanced Liquid-to-Air CDU system, the overall Power Usage Effectiveness (PUE) of the facility is driven down to as low as 1.19, approaching the theoretical ideal of 1.0 and drastically reducing cooling overhead.
Containerized Solutions and Spatial Efficiency
Speed of deployment and spatial efficiency are paramount for modern AI infrastructure. Delta’s prefabricated AI Containerized Data Center (CDC) Solutions address these requirements by delivering up to 60kW per rack of AI compute power within the physical footprint of a single standard parking space. This modular approach allows for rapid scalability and flexible deployment, enabling organizations to expand their compute capabilities without the prolonged construction timelines associated with traditional data center builds. The containerized design seamlessly integrates the high-density power and liquid cooling systems, ensuring that spatial constraints do not bottleneck computational expansion.
Microgrid Solutions and Solid State Transformers
Beyond the data hall, Delta Electronics presented a holistic Microgrid Solution that combines on-site energy generation, energy storage, power conversion, and Solid State Transformers (SST) within a prefabricated DC-coupled architecture. A standout feature of this microgrid is the use of SSTs, which are capable of directly converting medium voltage power to 800VDC. This direct conversion significantly reduces the number of required power conversion stages, thereby minimizing energy loss and allowing the system to respond dynamically to fluctuating AI loads. Additionally, the integration of Solid Oxide Fuel Cell (SOFC) technology provides reliable, around-the-clock on-site power generation. Real-time coordination across these multiple energy sources ensures continuous uptime and pristine power quality, even during grid connection delays or during off-grid operations.
Implications for the Future of AI Data Centers
The infrastructure unveiled by Delta Electronics highlights a critical inflection point in the design of AI data centers. The sheer power density required by platforms like the NVIDIA Vera Rubin necessitates a fundamental rethinking of both power delivery and thermal management. By standardizing on an 800VDC architecture and leveraging solid-state transformers, the industry can bypass the inefficiencies of legacy alternating current distribution. Furthermore, the integration of SOFC and advanced liquid cooling demonstrates that future AI factories will need to operate as self-sustaining microgrids. Ultimately, Delta’s comprehensive approach at DCW Asia 2026 provides a scalable, highly efficient blueprint for the next generation of AI infrastructure, ensuring that power and thermal constraints do not impede the rapid advancement of artificial intelligence.