At SC26, Delta introduced a from-grid-to-chip portfolio for AI infrastructure that supports emerging 800 VDC and ±400 VDC power architectures. The lineup includes the Project Deschutes in-row CDU, which delivers up to 2 MW of cooling at a 3°C approach temperature, and the Google OCP Project Brazos rack-mounted liquid-to-air CDU.
Overview
At Supercomputing 2026 in Chicago, Delta Electronics presented a “From-Grid-to-Chip” cooling portfolio at booth #1443, aimed at AI infrastructure and emerging 800 VDC and ±400 VDC power architectures. The show runs November 17–19, 2026 at McCormick Place. Delta’s message is that cooling can no longer be a facility-level afterthought; it must be engineered from chip package through board, rack, row, and facility. The portfolio spans liquid and air cooling and supports the higher-voltage architectures now being discussed for AI data centers.
The From-Grid-to-Chip Framework
Delta frames the portfolio as spanning chip, board, rack, row, and facility applications. That matters because AI infrastructure creates thermal hotspots at multiple scales: chip-level heat density demands new interface materials and lid structures; board-level power, memory, and networking components need cold plates beyond traditional compute cooling; and rack- and row-level coolant distribution must scale to megawatt-class loads while remaining serviceable. Martin Tsai, GM of the fans and thermal management business group at Delta Electronics, said the company’s strength comes from deep integration of core power and thermal management technologies. He added that full-stack chip-to-system liquid cooling is expanding beyond compute into networking, memory/storage, and power cooling.
Chip and Board-Level Thermal Innovations
At chip level, Delta is advancing thermal management with next-generation cooling technologies featuring advanced thermal interface materials and micro-channel lid structures. These changes matter because as compute components push higher power, moving heat from die into the cooling loop becomes a primary constraint. At board level, Delta extends liquid cooling beyond traditional computing to power, memory, and networking. The new Power Shelf Embedded Cold Plate debuts with a customizable, ultra-thin design, flexible manifold layouts, and low-resistance multi-channel flow paths for targeted hotspot cooling in space-constrained power systems.
Delta will also showcase its Vertical Power Delivery (VPD) Double-Sided Cooling Solution for Open Adapter Card (OAC) architectures, supporting up to 3,000 W of computing components and 500 W of power components within a slim 40 mm cold plate profile, with heat flux exceeding 170 W/cm². For memory, it features a hot-swappable SSD cold plate with a modular design for stacked SSDs, enhancing serviceability, scalability, and thermal performance.
Rack and Row Coolant Distribution
At rack level, Delta showcases its latest 4U In-Rack Coolant Distribution Unit (CDU), delivering up to 200 kW. At row level, the portfolio includes the Project Deschutes Open In-Row CDU, delivering up to 2 MW at a 3°C approach temperature, and the Google OCP Project Brazos rack-mounted liquid-to-air CDU, delivering up to 60 kW per rack through a self-contained, closed-loop design. These address different deployment models: in-rack CDUs for localized distribution, open in-row CDUs for higher-capacity row architectures, and liquid-to-air CDUs where a liquid loop must interface with air cooling. The 3°C approach temperature on Project Deschutes is notable because it indicates how closely the CDU can approach coolant supply temperature, a key efficiency metric in liquid-cooled AI halls.
Power Architecture Shift: 800 VDC and ±400 VDC
Delta dedicates exhibition space to cooling for 800 VDC and ±400 VDC infrastructure. The shift toward higher-voltage distribution changes thermal requirements because power conversion and delivery systems generate heat in constrained spaces. A headline product is the 800 VDC 2.5 MW In-Row Liquid-to-Liquid Coolant Distribution Unit (L2L CDU), equipped with three 800 VDC 25 kW hot-swappable pumps in a 3+0 configuration designed for zero-downtime operation. Delta specifies flow up to 3,850 LPM and 60 psi at the operating point. Hot-swappable pumps are an availability feature: coolant distribution downtime can be as disruptive as compute downtime.
Air Cooling and Facility Integration
Although liquid cooling dominates the AI thermal narrative, Delta also introduces a next-generation fan wall for air cooling. It features 450 mm EC blowers with high-efficiency brushless DC motors, biomimetic impellers, and waterproof protection, designed to support the advanced power architectures. This matters because not every heat load will be liquid-cooled, and facility-level air movement remains part of the from-grid-to-chip chain. The fan wall also shows Delta addressing both liquid and air paths, which can simplify integration for data center operators.
Industry Impact and Implications for AI Data Centers
The broader implication is that AI infrastructure is forcing convergence of power delivery and thermal management. Higher-voltage architectures such as 800 VDC and ±400 VDC may reduce distribution losses, but they concentrate heat in power shelves and conversion stages. Delta’s Power Shelf Embedded Cold Plate and VPD double-sided cooling are direct responses. At row level, megawatt-class CDUs and liquid-to-air options let operators stage deployments: start with rack-level liquid cooling and scale to row-level distribution as density rises. Hot-swappable SSDs and pumps reflect a practical concern beyond peak performance: serviceability and uptime.
Future Outlook
Delta’s SC26 exhibit will test whether the industry accepts a fully integrated from-grid-to-chip model rather than piecemeal cooling procurement. The company positions its portfolio around the idea that chip, board, rack, row, and facility cooling must be designed together, especially as 800 VDC and ±400 VDC architectures emerge. Visitors to booth #1443 at McCormick Place from November 17–19, 2026 will see liquid cooling extended to power, memory, networking, and storage, alongside air-cooling products. If AI rack densities continue to rise, co-engineering power and thermal subsystems may become a defining requirement for infrastructure vendors.