Overview of the Delta and NVIDIA Collaboration

Delta Electronics highlighted its strategic collaboration with NVIDIA at the OCP Global Summit 2026, focusing on energy-efficient 800 VDC power and cooling solutions. Announced alongside the NVIDIA AI Factory MGX Ecosystem Ceremony on October 14, the partnership addresses the escalating power and thermal demands of increasingly dense AI factories. This collaboration underscores a critical industry transition toward higher voltage direct current architectures to optimize power delivery for next-generation AI racks.

Technical Deep Dive: 800 VDC Power Delivery Architecture

Delta showcased advanced power delivery mechanisms tailored for the NVIDIA MGX platform, simplifying grid-to-chip power conversion and reducing energy losses. Key innovations include:

  • Solid State Transformer (SST): Converts medium-voltage AC directly to 800 VDC without intermediate conversion steps.
  • HVDC Power Distribution Board (PDB): Converts 800 VDC or ±400 VDC to 54 VDC and 12 VDC, achieving up to 98% conversion efficiency at a power density of 1,000 W/in³ with built-in hot-swap capability.
  • 800 VDC 660kW In-Row Power Rack: Features 480kW backup power for the NVIDIA Vera Rubin NVL72, utilizing a 90 kW DC/DC Power Shelf for 800 VDC to 50 V conversion at 98.5% efficiency.

Advanced Thermal Management and Liquid Cooling

To manage the immense heat generated by high-density compute, Delta integrated robust liquid cooling solutions. The company unveiled a 5,000A liquid-cooled busbar for NVIDIA 3rd generation MGX racks, merging high-current power distribution with liquid cooling. The thermal portfolio also features a new 2.5 MW Liquid-to-Liquid Cooling Distribution Unit (L2L CDU). This CDU utilizes three hot-swappable pumps delivering a combined flow rate of up to 3,850 LPM and 60 psi of pressure, ensuring zero-downtime targets. Additionally, a micro-channel cold plate designed for the Vera Rubin NVL72 employs multi-impinging flow design, passive flow control, and advanced metal bonding processes to enhance heat transfer.

Market Context: The Shift to Higher Voltage DC

The transition to 800 VDC architectures represents a pivotal shift in data center design. As AI workloads scale, traditional power infrastructures face severe limitations in efficiency and physical footprint. The physical limitations of copper busbars and the thermal throttling associated with air cooling at these densities necessitate a fundamental rethinking of facility-level power distribution. By adopting 800 VDC, data centers can significantly reduce energy losses. NVIDIA MGX, a modular reference architecture for accelerated computing, facilitates this transition by enabling system builders to design future-compatible systems faster, thereby reducing costs, accelerating time to market, and improving return on investment.

Industry Impact on AI Data Center Infrastructure

The integration of Delta’s power and cooling solutions directly impacts the scalability of AI infrastructure. Ted Shyy, Executive Vice President of Power Electronics Business at Delta Electronics, emphasized that scaling AI requires infrastructure that evolves alongside computing platforms. By supporting Intelligent Power Smoothing (IPS) with up to 98% efficiency and offering flexible configurations, data center operators can achieve improved efficiency. Furthermore, the modular nature of the MGX ecosystem means that these power and cooling upgrades can be seamlessly integrated into existing data center footprints. This creates a viable upgrade path for future AI infrastructure without requiring complete facility overhauls, thereby maximizing capital expenditure efficiency for hyperscalers and enterprise operators alike.

Future Outlook for Next-Generation AI Factories

Looking ahead, the knowledge sharing initiated at the summit will guide future deployments. During an executive session on October 12, Ralf Pieper from Delta and Mike Tu from NVIDIA explored 800 VDC power architectures and rack-level liquid cooling. The ongoing dialogue between component manufacturers and silicon architects indicates a unified roadmap. Future iterations of AI accelerators will likely demand even tighter integration of power delivery and thermal extraction at the chip level, making the foundational architectures showcased by Delta and NVIDIA critical enablers for the next decade of artificial intelligence advancement.