Compal Showcases NVIDIA Rubin AI Racks at SC26

Compal Showcases NVIDIA Rubin AI Infrastructure at SC26 Compal Showcases NVIDIA Rubin AI Infrastructure at SC26

Compal Electronics is showcasing next-generation AI infrastructure built around NVIDIA Rubin GPUs and Vera CPUs at Supercomputing 2026 (SC26) in Chicago, highlighting liquid-cooled rack systems designed for generative AI, agentic AI and high-performance computing. The company’s demonstrations include an NVIDIA HGX Rubin NVL8-based architecture supporting up to 72 Rubin GPUs per rack and a separate Vera CPU rack configuration accommodating up to 256 CPUs. The announcement reflects a shift in AI infrastructure engineering toward integrated compute, power delivery, networking and liquid cooling as increasingly complex AI workloads place greater demands on data center architecture.

Compal Showcases NVIDIA Rubin and Vera AI Infrastructure at SC26

AI infrastructure is entering a new phase in which raw GPU performance is only one part of the data center equation. As generative AI applications evolve toward complex reasoning, reinforcement learning and autonomous agent workflows, infrastructure providers must coordinate GPU acceleration, CPU processing, networking, power distribution and thermal management.

Compal Electronics is addressing this shift with two rack-scale infrastructure solutions showcased at Supercomputing 2026 (SC26), running November 17–19 at McCormick Place in Chicago.

At Booth #5008, the Taiwanese electronics manufacturer is demonstrating a liquid-cooled system built around NVIDIA HGX Rubin NVL8 alongside a separate NVIDIA Vera CPU rack solution. The demonstrations highlight Compal’s role in integrating NVIDIA computing platforms into larger infrastructure systems intended for enterprise AI and high-performance computing environments.

HGX Rubin NVL8 Brings 72 GPUs Into a Single Rack

The first solution centers on Compal’s OG231-2-L1 direct liquid-cooled server, built around the NVIDIA HGX Rubin NVL8 platform.

The architecture supports up to nine servers in one rack, with eight NVIDIA Rubin GPUs per server. At maximum configuration, this translates to 72 GPUs within a single rack.

Each eight-GPU system incorporates NVIDIA’s sixth-generation NVLink technology, which provides high-bandwidth communication between GPUs. This is important for workloads that distribute large models and computational tasks across multiple accelerators.

According to NVIDIA’s published HGX Rubin NVL8 specifications, an eight-GPU configuration delivers up to 400 petaflops of NVFP4 inference performance and 24 TB/s of aggregate NVLink 6 Switch bandwidth. These are vendor-specified peak performance figures, rather than guaranteed application-level results.

NVIDIA also projects up to 10 times higher token throughput compared with HGX B200 for specific agentic AI inference workloads. Actual performance will depend on model characteristics, software configuration, precision and workload conditions.

The infrastructure is designed for large-scale model inference, AI training and other compute-intensive applications.

Liquid Cooling Addresses High-Density AI Infrastructure

Increasing GPU density creates thermal and power-management challenges for data center operators.

Compal’s rack architecture integrates direct liquid cooling (DLC), rack-level power infrastructure, centralized system management and leak protection mechanisms. Instead of treating cooling as an independent facility requirement, the design coordinates thermal management with compute and power infrastructure.

Direct liquid cooling transfers heat away from high-performance components through liquid-based cooling systems. Compared with conventional air-cooling approaches, it can support higher thermal densities and reduce dependence on moving large volumes of chilled air through server enclosures.

For AI infrastructure operators, the benefits extend beyond processor temperatures. Cooling architecture affects rack deployment flexibility, energy consumption, maintenance requirements and the amount of computing capacity that can be installed within existing data center constraints.

Compal’s approach reflects the growing importance of system-level engineering as AI servers become more demanding.

NVIDIA Vera CPU Rack Targets Agentic AI Workloads

Beyond GPU-accelerated computing, Compal is showcasing a separate rack-scale solution built around NVIDIA Vera CPUs.

The company’s OX130-8N-L1 Vera CPU rack can integrate up to 32 NVIDIA Vera CPU Compute Trays, supporting a maximum of 256 Vera CPUs. Each tray contains eight Vera CPUs and one NVIDIA BlueField-4 Data Processing Unit (DPU).

The architecture targets workloads that require substantial CPU resources, including AI agent execution, reinforcement learning, code processing, tool invocation and data management.

This distinction is important as agentic AI systems increasingly involve more than generating model responses. Autonomous agents may need to execute code, call external tools, retrieve information, manage intermediate tasks and coordinate multiple operations. These activities can create substantial CPU-side processing requirements alongside GPU-intensive inference.

NVIDIA positions Vera as a processor designed specifically for agentic AI and simulation. Its published specifications describe a rack supporting more than 22,500 concurrent CPU execution environments, allowing organizations to run large numbers of isolated workloads simultaneously.

NVIDIA also reports up to 1.8 times faster agentic sandbox performance compared with its designated x86 CPU baseline. Compal’s announcement cites up to four times better energy efficiency under specified comparison conditions. These are NVIDIA performance claims based on particular workloads and comparison configurations, not universal improvements across all CPU applications.

The inclusion of BlueField-4 DPUs adds infrastructure acceleration capabilities for networking and data movement, helping separate infrastructure processing from general-purpose CPU workloads.

Why CPU and GPU Integration Matters for AI Factories

The two Compal demonstrations illustrate an important architectural development: next-generation AI infrastructure requires coordinated CPU and GPU resources.

GPU systems handle computationally intensive operations such as model training and inference. CPUs support orchestration, application execution, data preparation and the processing activities surrounding AI workloads.

In agentic AI environments, these responsibilities become more interconnected. A system may need to execute multiple agents simultaneously, coordinate tool calls, process information and return results while GPU accelerators handle model inference.

NVIDIA’s Vera and Rubin platforms address different parts of this infrastructure requirement. Compal’s rack-scale integration approach brings these capabilities into deployment-oriented server and rack architectures.

The broader competitive landscape includes manufacturers and infrastructure providers such as Dell Technologies, HPE, Supermicro and Lenovo, which are also developing high-density AI systems, liquid-cooled servers and integrated data center solutions.

Compal’s opportunity lies in its engineering and manufacturing capabilities, particularly its ability to translate reference computing platforms into complete rack-scale systems that combine power, cooling and centralized management.

AI Infrastructure Investment Is Increasing

The demand for these systems is supported by the rapid expansion of AI-related infrastructure spending.

According to Gartner’s September 2026 forecast, worldwide AI spending is expected to reach approximately $2.67 trillion in 2026, representing 49.5% annual growth. AI infrastructure is a major component of this expanding investment market. Gartner

The International Energy Agency has also highlighted the growing electricity requirements of data centers as AI adoption accelerates. Its 2025 Energy and AI analysis projected global data center electricity consumption could more than double to approximately 945 TWh by 2030, with AI identified as a major driver of growth.

These developments increase the importance of energy-efficient processors, liquid cooling, power infrastructure and system-level optimization.

Compal’s Position in the NVIDIA Infrastructure Ecosystem

Compal’s SC26 demonstrations are part of a broader strategy to expand its presence in AI servers and advanced computing infrastructure.

Its value proposition extends beyond supplying individual compute modules. The company is emphasizing engineering capabilities across server platforms, rack integration, cooling systems and power management.

For cloud providers, enterprise data centers and research institutions, this approach could simplify the process of assembling infrastructure for increasingly demanding AI workloads.

However, the announcement describes showcased architectures and engineering capabilities rather than confirmed large-scale customer deployments. Compal has not disclosed customer orders, commercial shipment volumes or deployment performance benchmarks for these specific rack configurations.

The significance of the showcase is therefore architectural: AI infrastructure is moving toward integrated systems designed around the interaction of compute, networking, power and cooling.

As agentic AI and large-scale inference expand, the ability to deploy and operate these systems efficiently could become as important as the performance of the underlying processors.

Market Landscape

The AI infrastructure market is shifting from standalone accelerator servers toward integrated rack-scale systems designed for increasingly complex AI workloads.

Four developments are shaping the sector:

  • GPU infrastructure: NVIDIA Rubin and competing accelerator architectures are targeting larger AI training and inference workloads.
  • CPU-intensive agentic AI: Autonomous agents, reinforcement learning and simulation are increasing demand for high-performance CPU infrastructure.
  • Liquid cooling: Direct liquid cooling is becoming increasingly important as GPU density and rack-level power requirements rise.
  • Integrated AI factories: Data center operators are prioritizing coordinated compute, networking, power delivery and cooling architectures.

The International Energy Agency estimates that global data center electricity consumption could reach approximately 945 TWh by 2030, more than double its 2024 level. This reinforces the importance of infrastructure efficiency and thermal management.

Compal’s NVIDIA-based systems position the company within this expanding market, where server manufacturers compete on system integration, scalability, thermal engineering and deployment readiness.

Top Insights

  • Compal’s HGX Rubin NVL8 rack supports up to 72 NVIDIA Rubin GPUs across nine direct liquid-cooled servers.
  • Each eight-GPU HGX Rubin NVL8 system delivers up to 400 PFLOPS NVFP4 inference performance and 24 TB/s NVLink bandwidth.
  • Compal’s NVIDIA Vera CPU rack supports up to 256 CPUs across 32 compute trays.
  • NVIDIA Vera targets agentic AI, reinforcement learning and CPU-intensive workloads requiring large-scale concurrent execution.
  • Direct liquid cooling, rack-level power infrastructure and centralized management address growing AI data center density requirements.

Power Tomorrow’s Intelligence — Build It with TechEdgeAI

Grow Your
Brand Visibility

Looking to publish a press release, guest article, interview or podcast? Connect with us.

GET FEATURED
Subscribe

Sign up today for exclusive insights and updates.

Newsletter Signup