Overview: The Escalating Power Challenge in AI Supercomputers

The race to build more powerful AI supercomputers is forcing a fundamental redesign of electrical infrastructure. At the Open Compute Project (OCP) Global Summit 2026 on October 15, 2026, Delta Electronics showcased a comprehensive portfolio of high-voltage direct-current power delivery and liquid-cooling systems. These systems are designed for the next generation of NVIDIA-based AI infrastructure. As accelerator systems scale toward megawatt-class racks, delivering electricity efficiently has transitioned from a peripheral engineering task to a central constraint on computing capacity, footprint, and cooling budgets.

Technical Details: Delta’s 800 VDC Architecture and Hardware

Delta’s newly highlighted portfolio spans multiple stages of infrastructure between the electrical supply and computing hardware. Key components include an 800 VDC in-row power rack rated for 660 kilowatts of output, incorporating 480 kW of battery backup capacity, alongside a compact 800 kW variant without integrated battery backup. Additional hardware features a 90 kW DC/DC power shelf, a solid-state transformer converting medium-voltage AC directly to 800 VDC, and a 5,000-ampere liquid-cooled busbar designed for third-generation NVIDIA MGX racks. Furthermore, Delta presented a liquid-to-liquid coolant distribution unit rated for up to 2.5 megawatts of cooling capacity and cold-plate technology specifically for NVIDIA Vera Rubin NVL72 platforms.

Physics of Power Delivery: Why 800 VDC Matters

The underlying challenge is electrical distribution at extreme power densities. Traditional architectures struggle as racks approach a megawatt. The physics of direct current dictates that power equals voltage multiplied by current (P=VI). Delivering 1 megawatt at 54 volts requires approximately 18,519 amperes. At 800 volts, the same power requires only about 1,250 amperes. Because resistive losses scale with the square of the current, increasing voltage dramatically reduces distribution losses. This higher-voltage distribution reduces conductor cross-sections, simplifies buswork, and frees valuable rack space for accelerators and networking. However, higher voltage shifts engineering complexity toward insulation, protection, connectors, and fault isolation.

Market Context: Megawatt Cooling and System Integration

Electrical distribution is only half the challenge; nearly all consumed power becomes heat. Delta’s 2.5 MW liquid-to-liquid coolant distribution unit places it in the same capacity class as competing summit systems. Liquid cooling captures heat close to the processors, transferring it to a circulating loop. At these power levels, operators must account for flow distribution, coolant chemistry, filtration, pressure drop, and leak detection. Power and cooling must be engineered together. A rack capable of receiving hundreds of kilowatts electrically is useless if its thermal infrastructure cannot continuously reject the corresponding heat. Delta’s systems-level approach attempts to coordinate the power path rather than optimizing products independently.

Implications for AI Data Centers: Efficiency and Maintenance

While Delta’s announcement is primarily a showcase of its existing product portfolio rather than evidence of a new confirmed NVIDIA contract, the architectural shift carries significant operational implications. By consolidating power conversion and backup functions, and by utilizing 800 VDC to minimize current and resistive losses, data centers can drastically reduce their physical equipment footprint. According to the strategic summary of this initiative, this planned 800V VDC architecture targets NVIDIA's next-generation systems and promises up to 70% lower maintenance requirements. This reduction in maintenance is critical as operators seek to maximize uptime and reduce the total cost of ownership for increasingly complex, densely packed AI facilities.

Future Outlook: Standardization and Next-Generation Deployments

The transition to 800 VDC is not yet a single, settled industry standard. Nuances remain, such as distinguishing between 50 V and 54 V outputs in intermediate conversion stages, which cannot be treated as interchangeable without confirming specific system designs. The practical benefits of higher voltage depend on the complete electrical design, including the solid-state transformer's actual performance under varying operating conditions. As the industry moves forward, the integration of grid-to-rack power delivery with megawatt-class liquid cooling will define the viability of next-generation AI data centers. Delta’s comprehensive portfolio demonstrates a clear engineering direction, even as commercial deployments and independently verified efficiency gains continue to evolve.