Compal Electronics will showcase a portfolio of AI computing and networking systems built around NVIDIA technologies at Supercomputing 2026 (SC26), taking place November 17–19 in Chicago. The portfolio spans NVIDIA HGX-based servers, Rubin accelerated computing, CPUs, data processing units, high-speed network interfaces and Ethernet switching. The focus is on a growing infrastructure challenge: scaling AI workloads across multiple servers requires not only powerful processors, but also fast interconnects, coordinated data movement, and rack-level integration of power and cooling.
Compal Connects NVIDIA Compute and Networking for AI Clusters
As artificial intelligence workloads become larger and more distributed, the performance of an AI infrastructure system increasingly depends on how efficiently its components communicate. Powerful GPUs remain central to model training and inference, but scaling across multiple servers also requires high-bandwidth networking, coordinated CPU-GPU operations, fast data movement and infrastructure capable of managing demanding power and cooling requirements.
Taiwanese electronics manufacturer Compal Electronics plans to showcase its approach to this challenge at Supercomputing 2026 (SC26), where it will present AI infrastructure solutions built around NVIDIA’s latest computing and networking technologies. The event takes place November 17–19 at McCormick Place in Chicago, with Compal exhibiting at Booth #5008.
The portfolio spans air-cooled and direct liquid-cooled GPU servers, NVIDIA Rubin-based accelerated computing, modular server architecture, data processing units and high-speed Ethernet networking. The broader objective is to move beyond individual server performance toward integrated AI clusters in which computing, networking, rack design and cooling operate as a coordinated system.
AI Infrastructure Shifts Toward System-Level Design
The growth of agentic AI, large-scale inference and distributed computing is changing the requirements for data centre infrastructure. AI agents can generate complex sequences of tasks and interact with multiple models and tools, increasing demand for computing resources and efficient communication between processing nodes.
In large AI clusters, delays in transferring data between GPUs and servers can undermine the benefits of faster processors. Network congestion, inefficient resource coordination and infrastructure overhead can also affect how effectively expensive computing resources are used.
NVIDIA’s Vera Rubin platform reflects this architectural shift through an approach that brings CPUs, GPUs, networking and infrastructure processors into a more tightly integrated design. Its components include the Vera CPU, Rubin GPU, NVLink 6 switch, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet switch.
Compal’s SC26 showcase focuses on integrating these technologies into server platforms and larger infrastructure configurations.
Compal Expands Its NVIDIA Server Portfolio
The company will highlight several NVIDIA-based systems designed for different performance and cooling requirements.
Its SGX30-2 is an air-cooled server powered by NVIDIA HGX B300, while the OG430-2-L1 uses direct liquid cooling with the same accelerated computing platform. Compal will also showcase the OG231-2-L1, an accelerated computing server featuring NVIDIA HGX Rubin NVL8.
These systems give customers different deployment options depending on computing density, thermal requirements and the intended workload. Air-cooled platforms can suit configurations where conventional data centre cooling is sufficient, while direct liquid cooling is increasingly relevant for high-density AI deployments that generate substantial heat.
Compal will also demonstrate NVIDIA MGX modular architecture and NVIDIA RTX PRO acceleration technologies, extending its portfolio toward enterprise AI and visual computing applications.
Another component of the showcase is the NVIDIA Vera CPU Compute Tray. Its inclusion highlights the importance of CPU resources in AI infrastructure, particularly for coordinating workloads, managing data movement and supporting operations around GPU-intensive computing.
The company is positioning its engineering capabilities across individual compute modules, server platforms and rack-scale infrastructure rather than treating GPU servers as standalone products.
Networking Becomes a Critical AI Performance Layer
The networking portfolio is central to Compal’s message at SC26.
For multi-node AI clusters, high-speed connections are necessary to move data between computing resources without creating bottlenecks. Compal will highlight three NVIDIA technologies designed to address different parts of that challenge.
- NVIDIA BlueField-4 DPU: Offloads selected networking, storage and infrastructure management tasks from host CPUs, helping reduce the processing burden on general-purpose computing resources.
- NVIDIA ConnectX-9 SuperNIC: Provides high-performance, low-latency connectivity for data exchange between compute nodes.
- NVIDIA Spectrum-6 Ethernet switches: Provide high-speed cluster networking intended to improve data transfer efficiency across distributed AI infrastructure.
These technologies are designed to complement one another. The DPU handles selected infrastructure functions, the SuperNIC provides high-performance network connectivity at the server level, and the Ethernet switching layer connects systems across the wider cluster.
NVIDIA’s architecture places particular emphasis on coordinated networking between endpoints and the network fabric. Its technical documentation describes ConnectX-9 and Spectrum-6 as components of the Vera Rubin platform’s scale-out networking architecture.
For infrastructure providers such as Compal, integrating these components into complete server and rack configurations is important because networking performance depends on more than the specifications of an individual switch or network adapter.
From Servers to Integrated AI Network Fabrics
Compal’s system-level strategy brings together CPU and GPU resources, network interfaces, inter-node connectivity, rack integration, power delivery and cooling.
This is increasingly important as AI infrastructure moves toward larger clusters designed to support distributed training, inference and scientific computing. In such environments, the performance of the overall system depends on how effectively its individual components operate together.
The approach also reflects a broader shift in data centre design. Instead of optimising servers independently, manufacturers and infrastructure providers are increasingly engineering complete systems around specific AI workloads.
NVIDIA’s Vera Rubin platform is an example of this direction. The company describes it as an extreme-codesign architecture in which compute and networking technologies are developed to operate together as an integrated AI supercomputer.
Compal’s role is to translate that architecture into deployable systems, combining hardware integration with manufacturing and rack-level engineering.
Manufacturing Capacity Supports Larger Deployments
The ability to deliver AI infrastructure at scale also depends on manufacturing capacity and supply-chain flexibility.
Compal says its Daxi AI Server Manufacturing Center in Taiwan, opened in August 2026, has expanded its server manufacturing and rack integration capabilities. Investments in its Texas operations and Vietnam production capacity are intended to strengthen regional manufacturing flexibility and supply-chain resilience.
These investments matter as demand for high-density AI systems increases. Customers deploying large clusters need more than access to processors and networking components; they also require suppliers capable of integrating, testing and delivering complex systems at scale.
Compal’s manufacturing strategy therefore complements its portfolio expansion. Its business opportunity lies in supporting customers across server adoption, system integration and larger infrastructure deployments.
The Competitive Focus Moves Beyond GPU Performance
The AI infrastructure market is becoming more competitive as technology companies and hardware manufacturers work to address bottlenecks across the full computing stack.
NVIDIA continues to expand its platform across GPUs, CPUs, networking and data processing. Server manufacturers and system integrators are competing to turn those technologies into complete systems that can be deployed in enterprise data centres, cloud environments and high-performance computing facilities.
For Compal, the SC26 showcase highlights a strategy built around integration rather than individual component performance. Its ability to combine compute, networking, cooling and manufacturing capabilities could be important for customers building larger AI clusters.
The next phase of AI infrastructure will be shaped not only by how quickly processors can execute workloads, but also by how efficiently entire systems coordinate data and resources. Compal’s SC26 portfolio reflects that transition, with an emphasis on connecting NVIDIA’s computing and networking technologies into scalable infrastructure for distributed AI and high-performance computing.
4. Market Landscape
AI infrastructure is evolving from GPU-centric server deployments toward integrated computing systems that combine processors, networking, storage, power and cooling.
Four developments are shaping the market:
- Distributed AI computing: Large-scale training and inference increasingly depend on efficient communication between servers and computing nodes.
- High-speed AI networking: SuperNICs, DPUs and advanced Ethernet switches are becoming important to reducing data-transfer bottlenecks and improving cluster utilisation.
- Rack-scale system engineering: Infrastructure providers are designing complete systems rather than optimising individual servers in isolation.
- Manufacturing and supply-chain expansion: AI infrastructure suppliers are investing in regional production and integration capacity to support increasingly complex deployments.
NVIDIA’s Vera Rubin architecture illustrates the shift toward extreme codesign across compute and networking. Its platform combines CPUs, GPUs, switches, SuperNICs and DPUs as components of a coordinated AI infrastructure system.
Compal’s position is at the system integration and manufacturing layer. Its opportunity lies in translating advanced NVIDIA components into complete server and rack configurations that customers can deploy for AI, scientific computing and HPC workloads.
Top Insights
- Compal will demonstrate NVIDIA-based AI infrastructure at SC26, including HGX B300 servers, Rubin NVL8 systems and rack-level integration.
- Its portfolio combines air-cooled and direct liquid-cooled servers to address different computing density and thermal requirements.
- BlueField-4, ConnectX-9 and Spectrum-6 provide networking and infrastructure capabilities designed for distributed AI clusters.
- Compal is expanding beyond server manufacturing toward integrated AI network fabrics, combining compute, connectivity, power and cooling.
- Manufacturing investments in Taiwan, Texas and Vietnam are intended to support larger AI infrastructure deployments and supply-chain flexibility.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
