Compal is expanding its AI infrastructure strategy beyond servers, showcasing 800VDC power, liquid cooling and rack-scale systems designed to support increasingly dense NVIDIA-accelerated AI factories at OCP Global Summit 2026.
AI Factory Design Is Moving Beyond the GPU
The next generation of AI data centers is forcing infrastructure providers to rethink a problem that goes well beyond installing more GPUs. As accelerated computing becomes denser, power delivery, thermal management and rack integration increasingly determine how much computing capacity a data center can deploy.
Compal Electronics is using OCP Global Summit 2026 to showcase an infrastructure strategy built around that challenge, combining NVIDIA accelerated computing with high-voltage power architectures, advanced liquid cooling and rack-scale engineering.
The company says its next-generation systems are designed for AI factories moving toward megawatt-scale power requirements and substantially higher rack densities. Rather than treating servers, power and cooling as independent components, Compal is positioning them as interconnected parts of a single AI infrastructure architecture.
That shift reflects a broader change in data-center engineering. Traditional enterprise workloads could often be supported by incremental improvements to server density and cooling. AI workloads, particularly large-scale model training and inference, are creating much more concentrated demands on both electrical and thermal infrastructure.
800VDC Targets Rising AI Rack Power
Power delivery is one of the areas Compal is targeting with its next-generation architecture.
The company is advancing 800VDC power architecture and megawatt-class power infrastructure, designed to support future high-density AI data centers. Higher-voltage DC distribution can reduce some of the conversion stages involved in moving electricity through a data center, potentially improving efficiency and simplifying power delivery as system requirements increase.
The significance is closely tied to rack power density. As accelerators become more capable, the electrical infrastructure feeding them has to scale accordingly. At very high densities, power distribution becomes a fundamental constraint on how data-center operators design racks, rows and entire facilities.
Compal’s approach is therefore less about a single power component and more about creating an architecture that can accommodate future generations of accelerated computing.
The company has not provided detailed efficiency figures or deployment specifications for the 800VDC architecture in its OCP announcement, so the practical gains will depend on implementation, workload and data-center design.
Liquid Cooling Becomes Essential Infrastructure
Power density creates a corresponding thermal challenge.
Compal is also showcasing advanced liquid-cooling technologies and megawatt-scale cooling capabilities intended to manage the heat generated by increasingly dense AI systems.
Air cooling remains common across data centers, but its ability to economically remove heat becomes more constrained as rack power increases. Liquid cooling can transfer heat more efficiently and is increasingly being incorporated into high-performance computing and AI infrastructure.
For AI factories, cooling cannot be designed independently of compute and power. The amount of electricity delivered to accelerators ultimately becomes heat that must be removed, making the electrical and thermal architectures closely connected.
Compal says it is approaching the problem at the rack and data-center levels, integrating compute, power and cooling rather than optimizing each subsystem separately.
That model resembles the broader movement toward rack-scale AI infrastructure, where servers, networking, power and cooling are increasingly engineered as a coordinated system.
From Server Manufacturer to AI Infrastructure Provider
The strategy also represents a broader expansion of Compal’s role in the AI hardware ecosystem.
Instead of focusing solely on individual computing platforms, the company is positioning its engineering and manufacturing capabilities around infrastructure spanning the electrical grid to the GPU. The approach includes accelerated-computing platforms, power systems, cooling technology and rack integration.
The underlying market is expanding rapidly. The International Energy Agency estimates that global data-center electricity consumption could reach roughly 945 terawatt-hours by 2030, more than doubling from 2024 levels, with AI identified as a major driver of demand.
The power challenge is particularly acute for AI-focused facilities. NVIDIA’s latest accelerated-computing platforms are driving higher performance per rack, while hyperscalers and infrastructure providers are developing increasingly specialized power and cooling architectures around them.
Companies including NVIDIA, Dell Technologies, Hewlett Packard Enterprise, Supermicro, Schneider Electric and Vertiv are all participating in different parts of this infrastructure transition, spanning accelerators, servers, power distribution and thermal management.
Compal’s strategy sits across several of those layers.
The AI Factory Becomes an Infrastructure Problem
The concept of an AI factory increasingly describes a data center designed specifically to turn electricity and computing capacity into AI outputs at industrial scale. That makes infrastructure efficiency as important as accelerator performance.
An AI factory can have powerful GPUs but still face deployment constraints if electrical distribution cannot deliver sufficient power, cooling cannot remove the resulting heat, or rack infrastructure cannot accommodate the density.
Compal’s OCP showcase highlights this interdependency.
Its 800VDC architecture addresses power delivery. Liquid cooling addresses thermal density. Rack-scale integration connects those systems to accelerated computing. Together, they point toward a model in which AI infrastructure is designed as an integrated platform rather than assembled from isolated components.
The company says it will demonstrate these technologies at Booth E65 during OCP Global Summit 2026.
The larger question for data-center operators is how quickly this integrated approach can move from demonstrations and infrastructure road maps into standardized, economically deployable AI factories.
As AI compute requirements continue rising, the bottleneck may increasingly be found outside the GPU itself. Power availability, cooling capacity and rack-level engineering are becoming core parts of the architecture required to scale AI.
Market Landscape
AI data centers are entering an infrastructure cycle defined by higher rack power, liquid cooling and increasingly integrated rack-scale designs. The IEA expects global data-center electricity consumption to reach about 945 TWh by 2030, more than double 2024 levels, with AI a significant contributor.
The resulting market spans accelerators, servers, networking, power distribution, cooling and data-center engineering. NVIDIA supplies the accelerated-computing foundation, while infrastructure companies such as Schneider Electric and Vertiv are developing power and thermal systems for increasingly dense deployments.
Top Insights
- Compal is combining NVIDIA accelerated computing with 800VDC power, liquid cooling and rack-scale integration for next-generation AI factories.
- Its megawatt-class infrastructure strategy addresses the electrical and thermal constraints created by increasingly dense AI computing environments.
- Liquid cooling is becoming increasingly important as AI accelerator power densities exceed the practical limits of conventional air-cooling approaches.
- Compal’s approach treats compute, power and cooling as interconnected infrastructure rather than separate systems optimized independently.
- Rising data-center electricity demand is making power availability and energy efficiency increasingly important constraints on large-scale AI deployment.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
