5/28/2026
Delta Bolsters 800 VDC Architecture for Next-gen AI Factories with Cutting-Edge Power, Cooling and Microgrid Solutions
[Estimated reading time: 4 minutes]
As we’re entering a new AI industrial revolution where data centers are transforming into AI factories that can produce intelligence at massive scale, AI reasoning and agentic workflows now dominate. Workloads have shifted from humans talking to AI to AI talking to AI, requiring systems to process, reason, and act across extremely long token sequences.
Four scaling laws are reshaping system requirements: pretraining, post-training, test-time scaling, and agentic scaling. State-of-the-art AI is moving to mixture-of-experts (MoE)-based reasoning models where tokens are dynamically routed across GPUs, racks, and data centers with strict latency and cost sensitivity. To keep pace with the rapid evolution of accelerated computing, system builders need architectures that shorten time to market, scale efficiently, and deliver consistent performance from single nodes to full AI factories.
NVIDIA MGX™ Architecture: A Modular Foundation for AI Infrastructure
NVIDIA MGX™ is an open, modular reference architecture for accelerated computing and AI factories ranging from single-node servers to full AI data center infrastructures. It enables original design manufacturers (ODMs) and original equipment manufacturers (OEMs) to create a wide variety of server configurations tailored to enterprise needs.
Building on the NVIDIA MGX™ architecture, the NVIDIA Vera Rubin platform introduces five specialized rack-scale systems designed for advanced AI workloads. These systems support high-throughput and ultra-low-latency AI inference, dense CPU sandboxing, and large context memory storage, all critical for agentic AI applications. Together, these racks form an integrated system designed to deliver exceptional energy efficiency and cost-effectiveness, powering the next generation of data centers optimized for AI performance.
NVIDIA 800 VDC Architecture
As AI models continue to grow in size and complexity, data centers must deliver dramatically more compute performance within the same physical, power, and cooling constraints. Traditional power architectures weren’t designed for the density required by next-generation AI factories, where more power must be delivered efficiently to increasingly compute-dense racks.
To support next-generation AI factories, NVIDIA’s 800 VDC architecture rethinks how power flows from the grid to accelerated computing systems, helping support advanced AI computing platforms such as systems built on NVIDIA MGX™.
Delta’s Integrated Solutions Bolster 800 VDC Architecture for Future AI Data Centers
To meet the power and cooling demands of next-generation AI factories, Delta works closely with NVIDIA and offers an integrated ecosystem of advanced power, cooling, and microgrid for AI data center solutions tailored for NVIDIA’s cutting-edge 800 VDC power architecture helping both new and existing AI data centers adopt higher-density AI infrastructure while protecting current investments.
- Power Solutions – New 660kW Power Rack Systems with BBU Function
Developed for systems based on NVIDIA MGX architecture, Delta’s power solutions highlight the new 800 VDC In-Row 660kW Power Racks. It features six 110kW Power Shelves with embedded 80kW BBUs for each shelf (480kW in total). These power racks are supported by the newly developed 18.5kW AC/DC Power Supply Units (PSU), which offer energy efficiency as high as 98% and innovative energy storage aluminum capacitors to optimize GPU workloads against high-frequency dynamic distortion.
Delta also provides 1RU 90kW DC/DC Power Shelves, 1RU Power Capacitance Shelves, and management switches that support system scale up.
- Advanced Cooling Solutions – New 2.4MW/3MW liquid-to-liquid CDU and Latest Micro Channel Cold Plates for Next-Gen AI Servers
To support the cooling of NVIDIA MGXTM platform, Delta’s thermal management portfolio includes in-row and in-rack liquid cooling solutions to address the increasing heat densities of AI systems. Designed for AI factories adopting 800 VDC power architectures, the new 2.4MW liquid-to-liquid (L2L) CDU solution offers self-contained 800 VDC electrical pumps with N+1 redundant pump design and unique cooling channel designs to achieve 2,400kW of cooling capacity and approach temperature as low as 4oC with a small footprint of 1,500mm width * 1,200mm depth * 2,286mm height.
In addition, Delta’s new 3MW liquid-to-liquid CDU offers up to 3,000kW of cooling capacity with 1.5 LPM/kW and 50 psi hydraulic performance. The system is scalable up to eight units operating as a group to enable optimal liquid cooling in next-generation AI factories.
As to the rack-level cooling, Delta’s 4RU In-Rack 140kW liquid-to-liquid CDU for NVIDIA GB300 NVL72 leverages plate-type heat exchangers to transfer heat from the secondary liquid loop (server side) to the primary facility water system of the data center.
- Microgrid for Data Centers Solution
To support customers requiring multiple, flexible, sustainable, and resilient energy sources for their 800 VDC-powered AI factories and edge computing data centers, Delta has leveraged its superior innovation prowess and system integration capabilities to develop energy-efficient and scalable Microgrid for Data Center Solutions. The key highlight of this solution is Delta’s newly developed Solid State Transformer, which boasts high-frequency power electronics to deliver leading efficiency as high as 98.5% while converting medium voltage AC to 800 VDC.
Data centers are under pressure to address both ‘time-to-power’ and sustainability challenges. Delta’s SOFC provides resilient onsite power generation to AI data centers with quiet operations and lower CO2 emissions. Demonstrated as a greener alternative to traditional gas turbines, Delta’s SOFC can also accelerate deployment efficiency—reducing 'time-to-power' from years to just months.
The unveiled Microgrid for Data Centers Solution also integrates Delta’s All-in-One Energy Storage System and Energy Management System, offering AI factory ecosystems with power plant-scale modeling and simulation, real-time control with millisecond-scale response time while maintaining stable system voltage within required parameters.
Together, these solutions illustrate Delta’s system level readiness to support AI data centers—providing the foundational AI infrastructure that hyperscalers and system integrators need.
Delta showcased the NVIDIA MGX architecture along with the 800 VDC In-Row Power Rack and at GTC 2026 in March.
Delta’s 2.4MW and 3MW L2L CDUs are ready to support the cooling of NVIDIA MGX platform.