Delta Electronics is showcasing high-voltage DC power delivery and liquid-cooling technologies designed for next-generation AI data centers at the 2026 OCP Global Summit. Developed for NVIDIA’s 800 VDC architecture and MGX infrastructure ecosystem, the portfolio includes a 660 kW in-row power rack with 480 kW of backup capacity and a liquid-to-liquid cooling distribution unit rated for up to 2.5 MW. The systems target the growing power density and thermal-management demands of accelerated computing platforms, including NVIDIA Vera Rubin NVL72. Delta says its approach can streamline power conversion and integrate power delivery with cooling as AI infrastructure scales.
Delta Advances 800 VDC Power and Cooling for AI Data Centers
As artificial intelligence workloads push data center power requirements higher, infrastructure providers are rethinking how electricity reaches AI servers and how heat is removed from increasingly dense computing systems. Delta Electronics is addressing both challenges with a portfolio of high-voltage direct-current (HVDC) power delivery and liquid-cooling technologies designed around NVIDIA’s 800 VDC architecture.
At the 2026 Open Compute Project (OCP) Global Summit, Delta is demonstrating systems for NVIDIA MGX-compatible infrastructure, including a 660 kW in-row power rack with integrated battery backup and a liquid-to-liquid cooling distribution unit (CDU) with up to 2.5 MW of cooling capacity. The company is positioning the technologies as building blocks for AI factories using next-generation platforms such as NVIDIA Vera Rubin NVL72.
The announcement reflects a broader infrastructure shift: as AI racks consume more power, data center operators need to coordinate electrical conversion, backup power, distribution, and thermal management rather than optimize each subsystem independently.
Simplifying power delivery from grid to server
Delta’s proposed 800 VDC architecture begins with a solid-state transformer (SST), designed to convert medium-voltage alternating current directly to 800 VDC. A high-voltage power distribution board then converts 800 VDC or ±400 VDC into lower-voltage outputs, including 54 VDC and 12 VDC, for AI servers.
Delta reports that its distribution board reaches up to 98% conversion efficiency at a power density of 1,000 watts per cubic inch. It also includes hot-swap capability, allowing supported components to be replaced without requiring the same maintenance approach as a fully fixed system.
The objective is to reduce conversion stages and associated energy losses as power demand grows. However, the actual efficiency advantage will depend on the complete electrical architecture, operating loads, and the equipment being replaced. High-voltage DC systems also require appropriate protection, isolation, monitoring, and maintenance procedures.
A 660 kW rack-level power system
For rack-level delivery, Delta is demonstrating an 800 VDC in-row power rack intended for NVIDIA Vera Rubin NVL72 and other MGX configurations. Six 7RU modules combine 110 kW power shelves with 80 kW battery backup units, delivering a stated total output of 660 kW and 480 kW of backup capacity.
A separate 90 kW DC/DC power shelf converts 800 VDC to 50 V, with Delta reporting conversion efficiency of up to 98.5%. The portfolio also includes an 800 kW in-row power configuration without integrated battery backup, providing an alternative for deployments with different power and redundancy requirements.
Delta is also highlighting a 5,000-amp liquid-cooled busbar for third-generation NVIDIA MGX racks. Combining high-current power distribution with liquid cooling is intended to address heat generated by the electrical infrastructure itself, not just by processors and accelerators.
The company additionally describes Intelligent Power Smoothing, with reported efficiency of up to 98%, as part of its approach to managing changes in power demand. The supplied announcement does not provide independent validation or full test conditions for these performance figures.
Liquid cooling targets megawatt-scale thermal loads
Electrical capacity is only one part of the challenge. High-density AI systems require cooling infrastructure capable of transferring heat away from servers while maintaining consistent coolant flow and pressure.
Delta’s new liquid-to-liquid CDU is rated for up to 2.5 MW of cooling capacity. It uses three hot-swappable pumps in a 3+0 configuration, with a combined stated flow rate of up to 3,850 liters per minute and a pressure of 60 psi at the specified operating point. The design is intended to support maintenance flexibility and cooling-system availability.
Delta is also presenting a cold plate for NVIDIA Vera Rubin NVL72. Its microchannel structure and multi-impinging flow design are intended to improve heat transfer while reducing flow resistance. The company says passive flow control and advanced metal-bonding processes support cooling performance and reliability.
These specifications describe different components within an integrated system; they should not be interpreted as proof that every deployment will achieve the maximum stated power and cooling figures simultaneously.
Integration becomes central to AI infrastructure
Delta’s participation in the NVIDIA MGX ecosystem and its joint technical session with NVIDIA at OCP highlight the importance of coordination between computing-platform designers and infrastructure suppliers. MGX provides a modular reference architecture for accelerated computing, while suppliers develop compatible power and thermal-management components.
For data center operators, the practical question is whether these systems can deliver dependable power, lower conversion losses, manageable maintenance, and effective cooling at scale. Deployment economics will also depend on facility compatibility, redundancy design, and the total cost of upgrading existing infrastructure.
Delta’s 800 VDC portfolio illustrates how power delivery and cooling are evolving together as AI facilities grow. Independent performance data and real-world deployment results will be needed to quantify the gains, but the underlying priority is clear: future AI infrastructure requires electrical and thermal systems designed for higher density from the outset.
Market Landscape
AI infrastructure is increasing demand for higher-capacity power distribution, efficient conversion, rack-level backup, and liquid cooling. As conventional rack architectures face growing power and thermal loads, 800 VDC designs are emerging as one approach to supporting future high-density computing systems.
Delta’s portfolio spans facility-to-rack power conversion, DC distribution, battery backup, liquid-cooled busbars, cold plates, and cooling distribution units. This integrated approach addresses a key infrastructure challenge: electrical systems generate heat too, so power and thermal designs must work together.
The market opportunity depends on more than peak specifications. Data center operators will evaluate conversion efficiency under real workloads, system redundancy, serviceability, cooling performance, and compatibility with existing facilities. Delta’s announcement provides product specifications and company-reported efficiency figures, but independent validation and deployment data remain important for assessing real-world benefits.
Top Insights
- Delta is showcasing 800 VDC power and liquid-cooling technologies for NVIDIA MGX-compatible AI infrastructure at the 2026 OCP Global Summit.
- The 660 kW in-row power rack includes 480 kW of stated backup capacity, targeting high-density AI systems and demanding power requirements.
- Delta’s liquid-to-liquid CDU offers up to 2.5 MW of cooling capacity, with three hot-swappable pumps in a 3+0 configuration.
- The company reports up to 98% efficiency for its HVDC power distribution board and up to 98.5% for its 90 kW DC/DC power shelf.
- Integrated electrical and thermal engineering is becoming increasingly important as AI data centers scale toward higher rack power densities.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
