PG

From data to tokens

ASUS AI Factory delivers intelligence at scale

Isometric illustration of the ASUS AI Infrastructure data center infographic featuring three warter towers, two rows of GPU servers, a central control hub with a large display screen, icons for cloud services and storage devices, and groups of people collaborating and interacting with technology—conveying a high-tech and dynamic environment.

  • The ASUS AI Factory brings innovation to life by transforming data into actionable intelligence and scaling ideas into real-world AI solutions. ASUS delivers comprehensive AI solutions, driving continuous innovation and maintaining a leading position. As an AI company, ASUS is committed to building an intelligent and sustainable future through design-thinking to weave AI into every aspect of our business. Our cross-product portfolio is helping lead the way to make AI accessible everywhere for everyone and make sure that everybody can benefit from the possibilities guaranteed by AI.

    Ubiquitous AI. Incredible possibilities.

    ASUS showcased a holistic AI ecosystem that extends beyond the PC, empowering users across diverse segments. This ecosystem is supported by robust AI infrastructure, encompassing scalable cloud and powerful on-premise solutions that seamlessly integrate hardware, software, and services, supporting the development and deployment of AI for various enterprise AI workloads. Our diverse portfolio includes intelligent solutions, strategically tailored for key segments such as developers, AI Researchers, data scientists, biotech, mission-critical applications, manufacturing, 5G and networking, and could-service providers.

  • ASUS co-engineering with NVIDIA to deliver full-stack solutions for token-generating AI Factory

    The comprehensive ASUS AI Factory approach seamlessly integrates hardware, software platforms, and services to support our clients’ success in AI deployment. From edge AI to AI supercomputing, our scalable infrastructure is designed to meet the demands of diverse workloads. Critically, we help organizations accelerate AI adoption while optimizing total cost of ownership (TCO).

    With a global network of over 2,200 service centers, ASUS delivers SLA-driven, tailored solutions that span the entire product lifecycle — from design and manufacturing to ongoing support. Our robust AI and data governance framework ensures data quality, regulatory compliance, and trust, safeguarding the integrity and accuracy of AI outcomes.

    Diagram of ASUS AI solutions, highlighting consulting, software services, and infrastructure. Features NVIDIA tech, ASUS platforms, and deployment orchestration.
    A digital illustration showing a data scientist analyzing graphs and servers on the left, connected by arrows to a large server stack on the right.

    Discover how the ASUS AI Factory is revolutionizing applications across industries.

  • Infrastructure architecture for all workloads

    Driven by our All-in-AI server strategy, we work closely with strategic partners to advance the AI ecosystem. Inspired by industrial manufacturing principles, the ASUS AI Factory approach treats AI development as a streamlined, repeatable process — turning data into valuable AI-driven insights and actions efficiently.

    The AI Factory with ASUS AI POD, validated by NVIDIA, provides guidance for developing, deploying, and managing agentic AI, physical AI, and HPC workloads on the NVIDIA Blackwell platform on-premises. Designed for enterprise IT, it provides accelerated computing, networking, storage, and software to help deliver faster time-to-value AI factory deployments while mitigating deployment risks.

    Below are the reference architecture designs that help clients use approved practices, acting as a knowledge repository and a standardized framework for diverse applications.

    Large and massive-scale computing

    The advanced ASUS AI POD, accelerated by NVIDIA GB200/GB300 NVL72 racks and incorporating NVIDIA Quantum InfiniBand or NVIDIA Spectrum-X Ethernet networking platforms, features liquid cooling to enable a non-blocking 576-GPU cluster across eight racks, or an air-cooled solution to support one rack with 72 GPUs. This ultra-dense, ultra-efficient architecture redefines AI reasoning computing performance and efficiency.

    Hyperscale GPU computing for AI training and inference

    NVIDIA HGX™ reference architecture optimized by ASUS delivers unmatched efficiency, thermal management, and GPU density for accelerated AI fine-tuning, LLM inference, and training. Built on the ASUS XA NB3I-E12 with NVIDIA HGX™ B300 or ESC NB8-E11 embedded with NVIDIA HGX™ B200, this centralized rack solution offers unmatched manufacturing capacity for liquid-cooled or air-cooled rack systems, ensuring timely delivery, reduced total cost of ownership (TCO), and consistent performance.

    • ASUS AI POD with NVIDIA HGX™ B300

      Reference architecture of ASUS AI POD featuring NVIDIA HGX B300 with liquid cooled solutions
      B300LC Quantum Spectrum
      GPU Support B300 NVL8 B300 NVL8
      Form Factor 5RU​ 5RU
      Rack/Node 8x Node per Rack / 9x rack 8x Node per Rack / 4x rack
      1 SU 9x rack / 72x node 4x rack / 32x Node
      Switch (N/S) 8x SN5610 (Leaf) / 4x SN5610 (Spine) 2x SN5610 (Leaf)
      Switch (E/W) 8x Q3400-RA (Leaf) / 4x Q3400-RA (Spine) 8x SN5610 (Leaf) / 4x SN5610 (Spine)
      Server XA NB3I-E12L XA NB3I-E12L
      Storage RS501A-E12-RS12U RS501A-E12-RS12U
      MGMT RS700-E12-RS4U RS700-E12-RS4U​

    • ASUS AI POD with NVIDIA HGX™ B300 AC

      Reference architecture of ASUS AI POD featuring NVIDIA HGX B300 with air cooled solutions
      B300AC Quantum Spectrum
      GPU Support B300 NVL8 B300 NVL8
      Form Factor 9RU 9RU
      Rack/Node 4x Node per Rack / 18x rack 4x Node per Rack / 8x rack
      1 SU 18x rack / 72x node 8x rack / 32x Node
      Switch (N/S) 8x SN5610 (Leaf) / 4x SN5610 (Spine) 2x SN5610 (Leaf)
      Switch (E/W) 8x Q3400-RA (Leaf) / 4x Q3400-RA (Spine) 8x SN5610 (Leaf) / 4x SN5610 (Spine)
      Server XA NB3I-E12 XA NB3I-E12
      Storage RS501A-E12-RS12U RS501A-E12-RS12U
      MGMT RS700-E12-RS4U RS700-E12-RS4U

    • ASUS AI POD with NVIDIA HGX™ B200

      Reference architecture of ASUS AI POD featuring NVIDIA HGX B200 with air cooled solutions
      B200AC Quantum Spectrum
      GPU Support B200 NVL8 B200 NVL8
      Form Factor 10RU 10RU
      Rack/Node 4x Node per Rack / 8x rack 4x Node per Rack / 8x rack
      1 SU 8x rack / 32x node 8x rack / 32x Node
      Switch (N/S) 2x SN5610 (Leaf) 2x SN5610 (Leaf)
      Switch (E/W) 8x QM9700 (Leaf) / 4x QM9700 (Spine) 4x SN5610 (Leaf) / 2x SN5610 (Spine)
      Server ESC N8B-E11 ESC N8B-E11
      Storage RS501A-E12-RS12U RS501A-E12-RS12U
      MGMT RS700-E12-RS4U RS700-E12-RS4U

    General purpose for scalable computing
    Scalable power for LLMs and immersive workloads

    ASUS presents NVIDIA MGX-compliant rack designs with the ASUS ESC8000 series, featuring dual Intel® Xeon® 6 processors and RTX PRO™ 6000 Blackwell Server Edition with the latest NVIDIA ConnectX-8 SuperNIC – supporting speeds of up to 800Gb/s or other scalable configurations – delivering exceptional expandability and performance for state-of-the-art AI workloads. Integration with the NVIDIA AI Enterprise software platform provides highly-scalable, full-stack server solutions that meet the demanding requirements of modern computing.

    ASUS AI POD, featuring MGX 4U Platform

    Reference architecture of ASUS AI POD featuring MGX 4U platform
    MGX CX8 Quantum Spectrum
    GPU Support H200 NVL, L40S, RTX Pro 6000 BSE H200 NVL, L40S, RTX Pro 6000 BSE
    Form Factor 4RU 4RU
    Rack/Node 8x Node per Rack / 4x rack in total
    or
    4x Node per Rack / 8x rack in total    		 8x Node per Rack / 4x rack in total
    or
    4x Node per Rack / 8x rack in total
    1 SU 8x rack / 4x or 8x rack / 32x Node 8x rack / 4x or 8x rack / 32x Node
    Switch (N/S) 2x SN5610 (Leaf) 2x SN5610 (Leaf)
    Switch (E/W) 4x QM9700 (Leaf) / 2x QM9700 (Spine) 2x SN5610 (Leaf)
    Server ESC8000-E12P
    ESC8000A-E13P    		 ESC8000-E12P
    ESC8000A-E13P
    Storage RS501A-E12-RS12U RS501A-E12-RS12U
    MGMT RS700-E12-RS4U RS700-E12-RS4U

  • Al Service and Platform Development

    ASUS is committed to not only enabling rapid deployment of your data center but also ensuring its smooth and reliable operation. That’s why we offer a comprehensive suite of software tools and platforms, allowing for one-click deployment and customizable interfaces and services—such as billing—tailored to your specific business needs.

    • Deployment and Management Platform

      • AIDC

        Automated large-scale deployment and provisioning

        Screenshot shows data center monitoring status and provisioning.

      • Log & Monitoring

        Full-stack visibility and proactive alerts

        Screenshot shows data center logs and alerts

      • ACC

        Centralize server management

        Screenshot shows the centralized control dashboard for data center

      • UFM & NMX

        Smart fabric monitoring and network optimization

        Screenshot shows the fabric and network diagram in the data center

    • Service Platform

      • AI Platform

        Scalable AI training and inference platform

        Screenshot shows ai training and inference platform

      • HPC Scheduler

        Efficient compute resource and job orchestration

        Screenshot shows the compute resources arrangement.
    ASUS total infrastructure solution from computing implementation to cooling design and professional services

  • Driving AI transformation across industries with ASUS AI Factory

    More Success Stories 〉

    • NVIDIA-Powered Precision: ASUS Servers Drive AI Transformation

      An automated factory scene with 3 robotic arms.
      ASUS partners with manufacturing clients to realize visionary concepts using NVIDIA Omniverse and large-scale AI. Powered by ASUS AI servers on NVIDIA HGX and AGX platforms, intelligent agents operate across real-world environments, collecting data, processing insights at the edge, and syncing with Omniverse simulations in real time. ASUS-built datacenters handle core compute while clients’ robotic systems run on NVIDIA AGX, seamlessly integrating sensor feeds, edge inferencing, and cloud coordination to deliver reliable AI performance.
      • ESC8000A-E13P

        ESC8000A-E13P

        • AMD Turin CPU
        • NVIDIA RTX PRO 6000 x 8
        • DIMM 64GB x 8
        • NVIDIA CX7 x 3
        • 1TB OS drive x 2
      • VSD320D-RS12

        VS320D-RS12

        • Intel Xeon CPU
        • Auto-tiering
        • DIMM 32GB x 4
        • NIC 25G Ethernet 4
        • SAS HDD 20TB x 12

    • ASUS Powers National Biomedical AI Supercomputer, Helping Revolutionizing Life Sciences

      Two researchers are discussing a medical image on a screen.
      ASUS partnered with the National Health Research Institutes to build a biomedical AI supercomputer based on its robust AIHPC platform and NVIDIA SDK (Software Development Kit). This partnership drives breakthroughs in biomedical research and precision medicine, enabling the medical institutions to advance life science research with cutting-edge technology. ASUS handles model development, infrastructure maintenance, and optimization to ensure seamless AI compute performance. The supercomputer delivers significant speedups: genomic data analysis runs 87 times faster, molecular dynamics simulations 9 times faster, and AI image annotation and natural language processing are accelerated by up to 180–250 times compared to traditional CPUs.
      • ESC NB8-E11

        ESC NB8-E11

        • 4th/5th Gen Intel® Xeon® Scalable Processors x2
        • NVIDIA Blackwell HGX™ B200 x 8
        • DIMM x 32, DDR5 5600 RDIMM/ 3DS RDIMM
        • Direct GPU-to-GPU interconnect via NVIDIA NVLink™
      • VSD320D-RS12

        VS320D-RS12

        • Intel Xeon CPU
        • Auto-tiering
        • DIMM 32GB x 4
        • NIC 25G Ethernet 4
        • SAS HDD 20TB x 12

    • Transforming Public Safety and Governance with the Smart City AI Platform

      The road detection display shows numerous cars highlighted with yellow boxes.
      ASUS provides the essential AI infrastructure and computing power to help drive Taiwan’s first Smart City Dynamic Image AI platform. Leveraging advanced ASUS GPU/CPU servers, storage solutions, and service management, the platform delivers unified data aggregation and AI-driven insights for the local city government. Essential information is all presented on the seamless Smart City dashboard, including traffic flow analytics, heat maps, and social distancing facemask detection. With a KVM-based service platform and CNCF-certified Kubernetes, ASUS ensures efficient training, inferencing, and container services. This collaboration enhances public safety, optimizes traffic, promotes data-driven governance, and fosters innovation while improving cost efficiency.   
      • ESC8000A-E13P

        ESC8000A-E13P

        • AMD Turin CPU
        • NVIDIA RTX PRO 6000 x 8
        • DIMM 64GB x 8
        • NVIDIA CX7 x 3
        • 1TB OS drive x 2
      • VSD320D-RS26

        VSD320D-RS26

        • Intel Xeon CPU
        • Auto-tiering
        • DIMM 32GB x 4
        • NIC 25G Ethernet 4
        • SAS SSD 15.36TB x 26

  • ASUS advantages and services

    As your trusted NVIDIA® Omniverse integration provider, ASUS offer a complete solution from strategic consulting to deployment support to help you unlock the full potential of industrial digitalization.

    Integration consulting
    Customized digital transformation blueprints
    Technical education
    Professional Omniverse skills training
    Deployment support
    Seamless system integration and implementation
    Custom solutions
    Application development for specific needs
    An engineer stands in the data center, holding a handheld computer. and Two engineers stand in the data center's control center, discussing a real-time monitoring display. and Four buttons float on a virtual screen. and Rows of computing clusters in the data center. An engineer stands in the data center, holding a handheld computer. and Two engineers stand in the data center's control center, discussing a real-time monitoring display. and Four buttons float on a virtual screen. and Rows of computing clusters in the data center.

  • FAQ

  • GB200 NVL72 Quantum
    Quantum Quantum Spectrum
    GPU Support GB200 GB300 GB300
    Form Factor 48RU 48RU 48RU
    Rack/Node 18x Compute Trays / 9x Switch Trays 18x Compute Trays / 9x Switch Trays 18x Compute Trays / 9x Switch Trays
    1 SU 16x rack / 288x node 16x rack / 288x node 16x rack / 288x node
    Switch (N/S) 8x SN5610 (Leaf) / 4x SN5610 (Spine) / 4x SN5610 (Core) 8x SN5610 (Leaf) / 4x SN5610 (Spine) 8x SN5610 (Leaf) / 4x SN5610 (Spine)
    Switch (E/W) 16x Q9700 (Leaf) / 18x Q9700 (Spine) 16x Q3400-RA (Leaf) / 9x Q3400-RA (Spine) 16x Q3400-RA (Leaf) / 9x Q3400-RA (Spine)
    Server ESC NM2N721-E1 XA GB721-E2 XA GB721-E2
    Storage RS501A-E12-RS12U RS501A-E12-RS12U RS501A-E12-RS12U
    MGMT RS700-E12-RS4U RS700-E12-RS4U RS700-E12-RS4U

  • ASUS AI WEKA Storage Reference

    Storage server with layers of drives

    Management node spec (WMS)

    • Server or VM must meet the following requirements :
    • Boot drivers : A system with two identical boot drives has the OS installed on mirrored partitions (LVM) with minimum 141GB usable capacity.
    • Cores and RAM : Minimum 4 cores and 16GB RAM
    • Network interface: Minimum 1 Gbps
    GB Series B Series MGX/HGX Series
    Server Model ASUS RS501A-E12-RS12U
    CPU & Memory AMD 64C ( 9005 series ) & 768G RAM
    Network 400G(200x2) IB NIC x 2
    Storage Boot disk : 2 x 960G NVMe M.2
    Storage disk : 16 x 30.72 TLC NVMe U.2
    Server nodes 27 19 8
    Usable Capacity 10PB 7PB 2.2PB
    Read Speed 2284GB 1607GB 676GB
    Write Speed 691GB 486GB 164GB
    Read IOPS 49.5 Million 34.8 Million 14.6 Million
    Write IOPS 12.4 Million 8.7 Million 3.6 Million
    Stripe Width 16+2+1 16+2+1 5+2+1

    Certified for WEKA Data Platform

  • Networking Topology

    Efficient and high-bandwidth network connectivity is crucial for NVIDIA GB200 NVL72 performance. Below is an outline of the typical external network architecture and technologies used to connect multiple NVL72 systems and integrate them into a larger network infrastructure. 

    Switch for data center

    NVIDIA Quantum-2 QM9700 Series InfiniBand Switch

    ODs 1 1 2 4
    Racks 1 2 4 8 6 32 64
    Computing Tray 18 36 72 144 288 576 1152
    Blackwell GPU 72 144 288 576 1152 2304 4608
    Bluefield-3 (DPU) 32 72 144 288 576 1152 2304
    Switch
    # of Spine Switches 0 2 4 8 16 32
    # of Leaf Switches 2 4 8 16 32 64

  • Liquid Cooling Solutions

    ASUS AI POD with NVIDIA GB200 NVL72

    ASUS, in collaboration with partners, offers comprehensive cabinet-level liquid-cooling solutions. These include CPU/GPU cold plates, cooling distribution units and cooling towers, all designed to minimize power consumption and optimize power-usage effectiveness (PUE) in data centers.

    Liquid-to-air solutions

    Ideal for small-scale data centers with compact facilities.
    Designed to meet the needs of existing air-cooled data centers and easily integrate with current infrastructure.
    Perfect for enterprises seeking immediate implementation and deployment

    Liquid-to-liquid solutions

    Ideal for large-scale, extensive infrastructure with high workloads.
    Provides long-term, low PUE with sustained energy efficiency over time.
    Reduces TCO for maximum value and cost-effective operations.

    Asus Air-cooled solution. There are two CDUs next to one rack.
    ASUS liquid-cooled solution featuring two rows of GB200 NVL72 units with two CDUs and an integrated pipeline system beneath.
PG