Delta Electronics unveiled its From-Grid-to-Chip liquid cooling and power portfolio at Supercomputing 2026 in Chicago, headlined by an 800 VDC-powered 2.5 MW in-row coolant distribution unit and 3,000 W cold plates. As university supercomputing centers and national laboratories deploy dense AI clusters exceeding 100 kW per rack, the move to 800 VDC architectures resolves critical facility power distribution losses and thermal barriers.
What Happened at SC26 in Chicago
At Supercomputing 2026 (SC26), held at McCormick Place in Chicago, power and thermal vendor Delta Electronics unveiled an expanded high-voltage direct current (HVDC) power and liquid-cooling portfolio for hyperscale artificial intelligence and high-performance computing (HPC) data centers. Announced ahead of the November 17–19 exhibition, the launch tackles the severe power delivery and cooling bottlenecks encountered as accelerator cluster footprints scale.
‘Delta’s unique strength in cooling comes from the deep integration of our core power and thermal management technologies, enabling us to address the increasingly complex demands of AI infrastructure,’ said Martin Tsai, General Manager of the Fans and Thermal Management Business Group at Delta Electronics. ‘Particularly, our full stack chip-to-system liquid cooling solutions are expanding beyond compute platforms to next-generation networking, memory/storage, and power cooling applications.’
The portfolio establishes a continuous thermal chain spanning silicon, compute boards, server racks, facility rows, and primary electrical feeds. By unifying power conversion with high-capacity liquid extraction, the architecture targets installations where traditional air cooling and standard 54 VDC rack busbars can no longer sustain operational thermal thresholds.
Key Details and Engineering Specifications
The flagship hardware debut is Delta’s 800 VDC 2.5 MW In-Row Liquid-to-Liquid (L2L) Coolant Distribution Unit (CDU). Designed to run directly from facility 800 VDC and ±400 VDC architectures, the unit integrates three 800 VDC 25 kW hot-swappable pumps in a 3+0 configuration to deliver uninterrupted zero-downtime operation. At nominal load, the unit circulates primary liquid at flow rates up to 3,850 liters per minute (LPM) under 60 psi of system pressure.
Alongside the 2.5 MW system, Delta highlighted complementary coolant distribution units designed for Open Compute Project (OCP) standards. The Project Deschutes Open In-Row CDU, built to Google open specifications, provides up to 2 MW of cooling at an approach temperature of 3°C between primary facility water and secondary rack loops. For facilities lacking hydronic infrastructure, Delta displayed the Google OCP Project Brazos rack-mounted liquid-to-air CDU, rejecting up to 60 kW per rack through a closed-loop internal circuit, as well as a 4U in-rack CDU providing 200 kW of localized heat exchange.
At the board and silicon level, Delta introduced cold plates engineered for extreme heat flux. Its Power Shelf Embedded Cold Plate uses multi-channel fluid passages to remove up to 3,000 W from high-density power shelves. The Vertical Power Delivery (VPD) double-sided cooling system for Open Adapter Card (OAC) form factors cools 3,000 W of compute silicon and 500 W of voltage regulation components within a 40 mm height, sustaining heat fluxes above 170 W/cm². For storage nodes, Delta added hot-swappable modular cold plates for stacked enterprise NVMe SSDs to prevent thermal throttling under continuous I/O.
| System | Tier | Capacity | Input Power | Key Specification |
|---|---|---|---|---|
| Delta In-Row L2L CDU | Row | 2.5 MW | 800 VDC / ±400 VDC | Three 25 kW pumps (3+0), 3,850 LPM, 60 psi |
| Project Deschutes CDU | Row | 2.0 MW | Facility Water | 3°C approach temp, Google OCP design |
| Delta 4U In-Rack CDU | Rack | 200 kW | Chassis Loop | Compact liquid-to-liquid exchange |
| Project Brazos CDU | Rack | 60 kW | Air Rejection | Closed-loop, no external hydronic feed |
| Power Shelf Cold Plate | Board | 3,000 W | Direct Liquid | Low-profile embedded power cooling |
| VPD Double-Sided Cooler | Module | 3,500 W | Direct Liquid | 40 mm height, heat flux >170 W/cm² |
Background and the 800 VDC Architectural Shift
The transition toward 800 VDC power topologies addresses basic physical constraints in high-density accelerator racks. Historically, data center distribution stepped 480 VAC down to 48 VDC or 54 VDC at the rack busbar. While adequate for 15 kW to 30 kW server racks, dense AI systems consume between 120 kW and 250 kW per rack. Supplying 120 kW at 48 VDC requires 2,500 amperes, necessitating massive copper busbars that add weight, complicate airflow, and produce severe resistive I²R heat losses.
By stepping voltage up to 800 VDC (or ±400 VDC referenced to ground), the required current drops by a factor of sixteen. This reduces resistive transmission losses by approximately 96%, allowing data centers to replace thick busbars with lighter cabling. Through engineering collaborations with NVIDIA, Delta demonstrated 800 VDC power shelves, solid-state transformers, and battery backup units capable of achieving 98% grid-to-chip AC-DC power conversion efficiency.
However, elevating rack voltage concentrates thermal dissipation within the power delivery equipment itself. Delivering hundreds of kilowatts in confined power shelves generates heat loads that exceed the limits of traditional air cooling blowers. Embedding liquid cold plates directly into power shelves resolves this bottleneck, integrating power electronics into the facility hydronic loop.
Reaction and Ecosystem Alignment
Industry observers and infrastructure specialists at SC26 underscored that power distribution and liquid cooling can no longer be architected in isolation. Financial analysis on Nasdaq and Webull, alongside technical coverage from Data Centre Solutions, highlighted that scaling AI supercomputing clusters requires co-designing power electronics and mechanical fluid delivery from the utility substation down to the silicon die.
Previous national laboratory supercomputing deployments—such as the exascale El Capitan supercomputer at Lawrence Livermore National Laboratory—demonstrated facility-wide direct liquid cooling for hundreds of thousands of processors. The SC26 announcements indicate that the next generation of leadership-class systems will pair direct liquid cooling with high-voltage DC distribution at the row level to control substation-level energy consumption.
The engineering trajectory aligns directly with community objectives for sustainable HPC and energy proportionality in computing. By leveraging high-efficiency 800 VDC variable-speed pumps and aggressive approach temperatures, data centers can operate secondary cooling loops with warm water up to 32°C, avoiding mechanical refrigeration chillers and lowering Power Usage Effectiveness (PUE) figures toward 1.05.
What It Means for Researchers and University Data Centers
For universities, research institutes, and facility managers, the arrival of commercial 800 VDC power and 2.5 MW CDUs impacts research computing operations across several administrative and technical dimensions:
- Brownfield Facility Constraints: Most academic data centers were engineered for 10 kW to 20 kW air-cooled racks. Accommodating megawatt-scale AI hardware requires either extensive mechanical retrofits—installing secondary hydronic loops, external heat exchangers, and chemical water treatment—or adopting self-contained liquid-to-air systems such as Project Brazos as an interim measure.
- Research Grant Budgeting: Principal investigators applying for institutional equipment grants or federal awards through research infrastructure funding programs must account for site-preparation expenses. Proposals can no longer calculate total cost of ownership solely from compute nodes; budgets must incorporate 800 VDC switchgear, CDU plumbing, and power-shelf infrastructure.
- Electrical Safety Protocols: Managing 800 VDC and ±400 VDC architectures requires compliance with updated National Electrical Code (NEC) guidelines and specialized arc-flash mitigation. Campus facility teams trained on alternating current must establish updated maintenance procedures for high-voltage DC busbars and battery systems.
- Storage Bottleneck Prevention: AI training and high-throughput data analysis generate heavy I/O workloads. Liquid-cooled NVMe storage enclosures ensure that high-speed storage tiers avoid thermal throttling, ensuring expensive GPU clusters maintain high utilization during large-scale data processing runs.
What Happens Next
Delta is exhibiting its live hardware portfolio at booth #1443 at McCormick Place throughout the SC26 conference from November 17 to 19, 2026. Supercomputing researchers, facility engineers, and system integrators can inspect operating demonstrations of the 2.5 MW L2L CDU, embedded power shelf cold plates, and 450 mm EC blower fan wall assemblies.
Commercial shipments of the 800 VDC in-row cooling systems and power shelves are slated for early 2027, aligning with the broader rollout of next-generation OCP Open Rack Version 3 (ORV3) specifications. Meanwhile, Delta is contributing to ongoing OCP technical committees to standardize universal quick-disconnect fittings, coolant purity standards, and high-voltage DC safety interfaces for multivendor scientific data centers.
Sources
- PR Newswire — ‘Delta Unveils From-Grid-to-Chip Cooling Innovations for AI Infrastructure and for New 800 VDC, ±400 VDC Architectures at SC26’ (2026-11-11).
- Delta Electronics (Americas) — ‘Supercomputing 2026: From-Grid-to-Chip Cooling Solutions Portfolio’ (2026-11-15).
- Nasdaq Corporate Solutions — ‘Delta Highlights Collaboration with NVIDIA in Advancing Energy-efficient 800 VDC Power and Cooling Solutions for AI Data Centers’ (2026-11-11).
- Webull Financial News — ‘Delta Showcases Next-Generation AI Liquid Cooling and 800 VDC Architecture at SC26’ (2026-11-11).
- Data Centre Solutions — ‘Delta Showcases AI Data Center Cooling and Power Solutions’ (2026-11-12).







