> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/dsx/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/dsx/_mcp/server.

# DSX Facilities Infrastructure Reference Design Overview

> Interactive walkthrough of the components and technologies in the NVIDIA DSX Facilities Infrastructure Reference Design.

#### inside-ai-factory

<h2 id="site-plan-components">
  Site Plan Component Details
</h2>

<p>
  These details mirror the interactive pin popouts so the page remains readable in generated Markdown and static exports.
</p>

<p>
  Source: 

  <em>NVIDIA DSX Facilities Infrastructure Design Guide v2.0</em>

   · August 19, 2026. Supplemental design references are available through 

  <a href="https://partners.nvidia.com/" target="_blank" rel="noopener noreferrer">NVOnline</a>

  .
</p>

<details open>
  <summary>
    250 MW / 96 SU Example Site Figures
  </summary>

  <p>
    This table presents example site figures for a 250 MW-class IT load design with 96 Scalable Units. It is a sizing reference independent of the rendered campus image, not a description of the pictured site. Actual site layouts and capacities vary by project.
  </p>

  <table>
    <thead>
      <tr><th scope="col">Example feature</th><th scope="col">Value</th></tr>
    </thead>

    <tbody>
      <tr>
        <th scope="row">Site footprint</th>

        <td>157 acres</td>
      </tr>

      <tr>
        <th scope="row">Building</th>

        <td>972,500 SF</td>
      </tr>

      <tr>
        <th scope="row">Compute halls</th>

        <td>4</td>
      </tr>

      <tr>
        <th scope="row">GPU racks</th>

        <td>1,536</td>
      </tr>

      <tr>
        <th scope="row">GPUs</th>

        <td>110,592</td>
      </tr>

      <tr>
        <th scope="row">Compute IT and mechanical load</th>

        <td>240 MW</td>
      </tr>

      <tr>
        <th scope="row">Core IT and mechanical load</th>

        <td>22.8 MW</td>
      </tr>

      <tr>
        <th scope="row">Stormwater drainage</th>

        <td>5 acres</td>
      </tr>
    </tbody>
  </table>

  <p>
    GPU rack count assumes a maximum of eight racks per compute row.
  </p>
</details>

<details open>
  <summary>
    Grid Substation & Utility Interconnect
  </summary>

  <p>
    <strong>DSX design value:</strong>

     Creates a scalable power backbone that aligns campus expansion with available utility capacity, bringing grid requirements, BESS functions, and long-lead equipment into the design before they become schedule or operating risks.
  </p>

  <p>
    Two buses in the same 34.5 kV air-insulated switchgear (AIS) or gas-insulated switchgear (GIS) distribute power across the site. The Service Substation lands utility lines above 

    <strong>100 kV</strong>

     and transforms them through HV-to-34.5 kV transformers into the campus medium-voltage backbone.
  </p>

  <p>
    This reference pattern separates Network Core and GPU Compute paths for different availability objectives. On-premises generation, if used, and BESS share the same 34.5 kV bus, making the substation the integration point for flexible power and DSX Flex grid services.
  </p>

  <ul>
    <li>
      Utility in: 100 kV+ transmission.
    </li>

    <li>
      Switchgear: 34.5 kV, 2,500 A, 40 kA.
    </li>

    <li>
      Includes HV breaker scheme plus HV/34.5 kV transformers, two buses for Core and GPU Compute, integration for utility, on-premises generation, and BESS, plus utility grid operator coordination through DSX Flex.
    </li>
  </ul>
</details>

<details>
  <summary>
    Stormwater Detention Basins
  </summary>

  <p>
    Twin engineered basins on the east and west of the campus capture, slow, and treat rainwater runoff.
  </p>

  <p>
    The two detention basins are positioned just outside the building perimeter on the east and west edges of the site. Their job is to 

    <strong>buffer storm flow</strong>

     from an impervious-heavy campus, detain that water, allow solids to settle, and discharge slowly at a metered rate that approximates pre-development runoff.
  </p>

  <p>
    See the 

    <a href="?open=reference-site">250 MW / 96 SU Example Site Figures</a>

     for the example stormwater figure.
  </p>

  <ul>
    <li>
      Quantity: two basins.
    </li>

    <li>
      Permitting: EPA NPDES plus state stormwater.
    </li>

    <li>
      Cooling-tower blowdown: none in this design.
    </li>
  </ul>
</details>

<details>
  <summary>
    Dry Coolers - Outdoor Heat Rejection
  </summary>

  <p>
    <strong>DSX design value:</strong>

     NVIDIA's 45°C liquid-cooling design point expands the operating window for rejecting campus heat without full mechanical chilling, leaving more of the facility power budget for AI compute.
  </p>

  <p>
    Dry coolers are positioned at the far east and west of the site, directly north of each Central Utility Building. Warm facility water from the chillers flows through finned-tube heat exchangers; fans pull ambient air across the fins, dumping heat to the air with 

    <strong>no evaporative water use</strong>

    .
  </p>

  <p>
    The CUB facility-water loop serves 

    <a href="#mechanical-gallery">mechanical gallery</a>

     

    <a href="#cdu">CDUs</a>

     and 

    <a href="#crah">CRAHs</a>

    , while the paired dry coolers reject its heat to the atmosphere.
  </p>

  <p>
    Dry coolers run on 

    <strong>480 V</strong>

     alongside the CUB pumps and controls. On the GPU side, the cooling plant is not UPS-backed; during utility outages it relies on chilled-water loop thermal mass until backup generation comes online and the system auto-restarts.
  </p>

  <ul>
    <li>
      Function: air-side heat rejection.
    </li>

    <li>
      Evaporative water use: none.
    </li>

    <li>
      Bus: 480 V.
    </li>

    <li>
      UPS: Core CUB only.
    </li>
  </ul>
</details>

<details>
  <summary>
    Battery Energy Storage System (BESS) & Backup Generation
  </summary>

  <p>
    <strong>DSX design value:</strong>

     Adds fast-response power flexibility around dynamic AI loads: BESS stabilizes load behavior and bridges grid events, while standby generation supports continuity for critical Core systems.
  </p>

  <p>
    AI loads are dynamic by nature, while the grid and generators are built to serve steady loads. In this reference campus, BESS sits on the 34.5 kV switchgear bus as a controllable buffer between them.
  </p>

  <p>
    Applicability and sizing are evaluated case-by-case with OEMs, grid operators, and regulators based on the site power architecture, operating objectives, grid requirements, and any product-specific requirements.
  </p>

  <p>
    Central standby generators provide backup power through an 

    <a href="#n-plus">N+2 configuration</a>

     at the 34.5 kV switchgear level, supporting five-nines (99.999%) reliability for critical Core network and mechanical loads.
  </p>

  <ul>
    <li>
      BESS functions: facility-level buffer, abnormal-event response, bridge power, and grid dispatchability.
    </li>

    <li>
      BESS applicability and sizing: deployment-specific and evaluated case-by-case.
    </li>

    <li>
      Shared connection: 34.5 kV MSG.
    </li>

    <li>
      Backup generation: N+2.
    </li>

    <li>
      Reliability supported: five-nines (99.999%).
    </li>

    <li>
      Core DSX Technologies: DSX Flex and DSX MaxLPS.
    </li>
  </ul>
</details>

<details>
  <summary>
    Central Utility Building (CUB)
  </summary>

  <p>
    <strong>DSX design value:</strong>

     Centralizes dual-temperature cooling for liquid- and air-cooled systems in a shared plant, which can reduce equipment and cooling overhead while improving site acoustics and maintenance conditions.
  </p>

  <p>
    Two CUBs at the east and west ends of the building house chillers, pumps, and facility-water distribution serving 

    <a href="#cdu">CDUs</a>

     and 

    <a href="#crah">CRAHs</a>

     in the 

    <a href="#mechanical-gallery">mechanical galleries</a>

    . Paired outdoor 

    <strong>dry coolers</strong>

     reject heat at both ends.
  </p>

  <p>
    The CUB is fed from dedicated MV loops. Chillers run on a decoupled 

    <strong>4.16 kV</strong>

     bus while the rest of the CUB, including pumps, coolers, and controls, runs on 

    <strong>480 V</strong>

    . UPS is provided only for the Core portion of the CUB.
  </p>

  <p>
    See the 

    <a href="?open=reference-site">250 MW / 96 SU Example Site Figures</a>

     for reference-scale Compute and campus load figures.
  </p>

  <ul>
    <li>
      Chiller bus: 4.16 kV.
    </li>

    <li>
      CUB / coolers bus: 480 V.
    </li>

    <li>
      UPS: Core CUB only.
    </li>
  </ul>
</details>

<details>
  <summary>
    CDUs - Cooling Distribution Units in the Mechanical Gallery
  </summary>

  <p>
    <strong>DSX design value:</strong>

     Places redundant CDU groups in mechanical galleries to reduce equipment count and support concurrent maintenance, while rack-level control and isolation limit cooling-fault impact and support tighter data-hall access control.
  </p>

  <p>
    This representative pin marks cooling equipment inside the 

    <a href="#mechanical-gallery">mechanical gallery</a>

    , not exterior site equipment. Larger shared piping arrangements reduce the number of 

    <strong>liquid-to-liquid</strong>

     CDUs while N+1 group operation supports concurrent maintenance.
  </p>

  <p>
    CDUs receive high-temperature facility water from the CUB and use it to cool a separate, technical-grade 

    <a href="#tcs">secondary loop</a>

     serving GPU cold plates.
  </p>

  <p>
    <a href="#crah">CRAHs</a>

     receive medium-temperature facility water from the CUB and provide air cooling for the data halls and other occupied spaces.
  </p>

  <p>
    The validated CDU partner list is dynamic. See the live NVIDIA Enterprise Marketplace catalog for current MP-Ready and Sample-Ready products.
  </p>

  <ul>
    <li>
      DSX form factor: liquid-to-liquid.
    </li>

    <li>
      TCS design flow: at least 1.5 LPM/kW.
    </li>

    <li>
      Validation: CDU self-qualification suite.
    </li>

    <li>
      CDU redundancy: N+1.
    </li>
  </ul>
</details>

<details>
  <summary>
    Compute Data Hall
  </summary>

  <p>
    <strong>DSX design value:</strong>

     Standardizes power, cooling, networking, and rack arrangements as scalable AI-capacity building blocks that can evolve across campus phases and GPU generations, simplifying design, procurement, and construction.
  </p>

  <p>
    A Scalable Unit is the smallest repeatable AI factory building block: 

    <strong>1 Compute <a href="#hac">Hot Aisle Containment (HAC)</a> plus 1 Support HAC</strong>

    , with a consistent power budget per SU.
  </p>

  <p>
    In the reference pattern, each data hall contains 18 Scalable Units, or 24 in the 

    <a href="#maxlps">NVIDIA DSX MaxLPS</a>

     design.
  </p>

  <p>
    Cabinet TDP scales from MGX Gen 1.1 at 198 kW to Vera Rubin NVL72 at 330 kW.
  </p>

  <p>
    See the 

    <a href="?open=reference-site">250 MW / 96 SU Example Site Figures</a>

     for the example compute-hall, GPU-rack, and GPU counts.
  </p>

  <ul>
    <li>
      Cabinet TDP: 198 kW to 330 kW.
    </li>

    <li>
      Air cooling for remaining equipment is provided via CRAHs in the Mechanical Gallery.
    </li>
  </ul>
</details>

<details>
  <summary>
    Network Core & Shared Services
  </summary>

  <p>
    <strong>DSX design value:</strong>

     Centralizes shared network services to keep fiber paths within design limits, simplify backup-power distribution, and support phased construction without disturbing installed cabling.
  </p>

  <p>
    The Network Core and Shared Services areas sit between the Compute Data Halls on both sides of the CIN Spine. They connect into the 

    <strong>CIN Spine</strong>

     and house the 

    <strong>Tenant Access Network</strong>

    , 

    <strong>Secure Management Network (SMN)</strong>

    , 

    <strong>High-Speed Storage</strong>

    , 

    <a href="#control-nodes">Control Nodes</a>

    , Meet-Me Rooms, and Fiber Entrance Rooms.
  </p>

  <p>
    See the 

    <a href="?open=reference-site">250 MW / 96 SU Example Site Figures</a>

     for reference-scale Core and campus load figures.
  </p>

  <ul>
    <li>
      Network relationship: CIN Spine, TAN, SMN, HSS.
    </li>
  </ul>
</details>

<details>
  <summary>
    CIN Spine & Campus Fiber
  </summary>

  <p>
    <strong>DSX design value:</strong>

     Centralizes the CIN spine so separate data halls can operate as one compute system within optical limits, with direct fiber paths preserving reach and supporting phased cluster growth.
  </p>

  <p>
    The Cluster Interconnect Network spine is the campus's east-west GPU fabric, connecting every Compute Hall into one cluster. Its central location and direct fiber routes keep connections within the 500 m CIN distance limit.
  </p>

  <p>
    The base single-story form factor is optimized for speed and cost-efficient development. Where CIN fiber runs would exceed 500 m, the modular building components can be arranged in an H-shaped form to keep links within the distance limit.
  </p>

  <p>
    Diverse-path carrier fiber lands at the MPOE in the Network Core. Inter-DSX cross-connects can link this campus to other DSX deployments over the same fiber spine.
  </p>

  <ul>
    <li>
      Function: east-west CIN spine.
    </li>

    <li>
      Scale steps: 10 / 16 / 21 / 32 / 42 / 64 / 128 SU.
    </li>

    <li>
      Centralized, modular CIN design supports phased cluster growth while keeping fabric connections within design distance limits.
    </li>

    <li>
      East-West CIN scales linearly; smaller configurations are available. For requirements between listed steps, use the next larger CIN configuration.
    </li>

    <li>
      TAN and SMN sizing varies by deployment.
    </li>

    <li>
      External fiber: diverse-path carrier.
    </li>

    <li>
      Cross-connects: Inter-DSX site links.
    </li>
  </ul>
</details>

<details>
  <summary>
    Operations Center & Site Entry
  </summary>

  <p>
    <strong>DSX design value:</strong>

     Organizes stable core services, adaptable perimeter space, and a full-width back-of-house corridor so operations areas can scale or change while preserving access and layout flexibility.
  </p>

  <p>
    The south side of the campus brings together the Primary Site Entrance, Security Guardhouse, parking lot, and Entry and Offices block. This is where badging happens, vendor tickets are triaged, and the 

    <a href="#noc">NOC</a>

    , 

    <a href="#soc">SOC</a>

    , and 

    <a href="#foc">FOC</a>

     monitor the campus 24/7.
  </p>

  <ul>
    <li>
      Entries: primary south plus secondary north.
    </li>

    <li>
      Operations: 24/7 NOC/SOC/FOC.
    </li>

    <li>
      Full-width back-of-house corridor and regular structural grid support layout flexibility.
    </li>
  </ul>
</details>

#### core-dsx-technologies

<p>
  NVIDIA DSX platform encompasses a combination of core technologies, software components, and reference designs. The following sections introduce key technologies used in the Facilities Infrastructure Reference Design and show how they relate to their parent NVIDIA DSX components.
</p>

<h2>
  NVIDIA DSX MaxLPS
</h2>

<p>
  NVIDIA DSX MaxLPS is a suite of technologies that brings together facilities and site design, dynamic power management software, and advanced performance-per-watt techniques. Within a fixed AI factory power envelope, it helps improve energy efficiency and utilization while maximizing AI factory performance per watt. Within the Facilities Infrastructure Reference Design, MaxLPS connects power and cooling system planning to AI-factory performance per watt within the fixed power envelope.
</p>

<a href="https://blogs.nvidia.com/blog/liquid-cooling-ai-factories/" target="_blank" rel="noopener noreferrer">
  Cooling · DSX MaxLPS

  45°C Liquid Cooling for AI Factories

  <p>Learn how higher-temperature liquid cooling and dry coolers can reduce cooling overhead and water use in suitable climates.</p>

  Read the cooling overview
</a>

Partner Context · Phaidra

Cooling, Power, and Workload Management

<p>Explore a partner example of using AI agents to coordinate cooling, power, and workloads around facilities constraints in liquid-cooled AI factories.</p>

<h2>
  NVIDIA DSX Flex
</h2>

<p>
  NVIDIA DSX Flex provides secure, dynamic power management between the grid and the AI factory. It receives grid signals, such as load shedding, demand response, and pricing events, and dynamically adapts AI workloads while orchestrating utility, on-site renewable, and energy-storage resources.
</p>

<a href="https://nvidianews.nvidia.com/news/nvidia-and-emerald-ai-join-leading-energy-companies-to-pioneer-flexible-ai-factories-as-grid-assets" target="_blank" rel="noopener noreferrer">
  Partner Context · Emerald AI

  Flexible AI Factories as Grid Assets

  <p>Explore NVIDIA’s public collaboration with Emerald AI on coordinating AI workloads with grid signals, utility power, and on-site energy resources in an AI factory.</p>

  Read the collaboration
</a>

<h2>
  NVIDIA DSX Exchange
</h2>

<p>
  NVIDIA DSX Exchange is the AI factory IT/OT communication hub. In the Facilities Infrastructure Reference Design, it enables building management systems and facilities components to share power, cooling, and operational signals with IT and AI factory operations systems.
</p>

IT/OT · DSX Exchange

DSX Exchange Architecture

<p>See how DSX Exchange enables building management systems and facilities components to share real-time power, cooling, and operational signals with IT services and AI factory operations systems.</p>

Explore DSX Exchange architecture

#### faq

<h2>
  Frequently Asked Questions
</h2>

<p>
  After reviewing the walkthrough and Core DSX Technologies, these questions help prospective NCPs consider their next decisions. Explore the public resources in 

  <a href="#learn-more">Learn More</a>

   for next steps.
</p>

<h3>Why do facilities matter to AI factory economics?</h3><p>Cooling, power conversion, and operational constraints affect how much available energy can reach productive AI workloads. At AI-factory scale, small inefficiencies in those systems compound across the facility and directly affect token efficiency and operating economics. DSX helps teams consider those constraints alongside compute, networking, and operations rather than as separate projects.</p>

<h3>When is NVIDIA DSX the right starting point for an AI factory project?</h3><p>NVIDIA DSX is an AI factory platform and portfolio that brings together generation-specific reference designs, core technologies, simulation, operations, and ecosystem technologies. Partners, AI factory operators, and software providers can adopt the elements that fit their use cases and requirements incrementally and independently; DSX does not require a fully NVIDIA-supplied facility.</p>

<h3>Do partners need to adopt the full DSX architecture?</h3><p>No. Partners can apply the DSX reference elements, published integration patterns, and ecosystem technologies that fit their architecture and deployment objectives; DSX does not require a fully NVIDIA-supplied facility. NVIDIA intends DSX to support a healthy ecosystem in which partners can build with DSX, extend it with their own automation and technology, and retain the differentiated parts of their stack.</p>

<h3>How can NVIDIA DSX Sim help before and during operations?</h3><p>DSX Sim helps teams model and validate facility and IT decisions before deployment, then revisit those decisions as the factory evolves in operation. Digital twins, simulation-ready assets, and infrastructure simulation can support configuration validation, change testing, and operational planning before changes reach production.</p>

<h3>When should teams plan for NVIDIA DSX Exchange?</h3><p>Plan for DSX Exchange when building management systems and facilities components need to share power, cooling, and operational signals with IT and AI factory operations systems. As the AI factory IT/OT communication hub, it enables scalable and secure integration across compute, network, energy, power, and cooling plant signals, while the BMS retains responsibility for equipment control.</p>

<h3>How do NVIDIA DSX MaxLPS, NVIDIA DSX Flex, and BESS affect power planning?</h3><p>Start by mapping the facility power budget, workload priorities, utility and grid conditions, on-site generation or storage, and the operational outcomes the deployment needs to support. Within a fixed AI factory power envelope, NVIDIA DSX MaxLPS brings together facilities and site design, dynamic power-management software, and advanced performance-per-watt techniques to make more site power available for AI workloads. Assess NVIDIA DSX Flex for grid-aware power orchestration and BESS for site-specific buffering, demand-response, and ride-through objectives. Engage NVIDIA and relevant energy, utility, and equipment partners for deployment-specific planning and qualification.</p>

<h3>What changes as AI infrastructure scales?</h3><p>As an AI factory grows, repeatable infrastructure decisions become more important. The DSX reference-design approach starts with scalable compute units, groups them with dedicated power and cooling into deployment units, and repeats those patterns across data halls and network infrastructure. This gives teams a way to plan compute, cooling, power, network capacity, and operations as the facility expands, while adapting the design to the product generation and project conditions.</p>

#### glossary-acronyms

<h2>
  Glossary & Acronyms
</h2>

<p>
  Plain-language definitions for terminology used across the DSX Facilities Infrastructure Reference Design Overview.
</p>

<article id="bess">
  <h3>BESS</h3><p>Battery Energy Storage System</p>Power

  <p>In this DSX facilities reference pattern, BESS ties into the 34.5 kV main switchgear and supports four functions: facility-level buffering, abnormal-event response, bridge power for source transfer and black start, and grid dispatchability—the ability to respond to grid operating requirements. BESS is a supported option in DSX AI Factory designs; applicability and sizing depend on the site power architecture, operating objectives, grid requirements, and any product-specific requirements.</p>
</article>

<article id="ais-gis">
  <h3>AIS / GIS</h3><p>Air-Insulated Switchgear / Gas-Insulated Switchgear</p>Power

  <p>Substation switchgear configurations that use air or insulating gas around energized equipment. The selected arrangement depends on site, utility, and equipment requirements.</p>
</article>

<article id="bms">
  <h3>BMS</h3><p>Building Management System</p>Site & Ops

  <p>Facilities control system that monitors and manages building equipment such as cooling, electrical, alarms, and environmental systems. DSX Exchange integration lets BMS and operational technology telemetry participate in AI Factory operations.</p>
</article>

<article id="it-ot">
  <h3>IT/OT</h3><p>Information Technology / Operational Technology</p>Site & Ops

  <p>Information technology includes the compute, network, and software systems that support AI factory workloads. Operational technology includes the facilities systems that monitor and control power, cooling, and other physical equipment. DSX Exchange provides an integration path between these domains.</p>
</article>

<article id="cdu">
  <h3>CDU</h3><p>Coolant Distribution Unit</p>Cooling

  <p>The bridge between the facility cooling loop and the rack. CDUs isolate fluids, regulate pressure, and deliver coolant directly to GPU cold plates.</p>
</article>

<article id="cin">
  <h3>CIN</h3><p>Cluster Interconnect Network</p>Compute & Network

  <p>The east-west spine that fuses data halls into a single training cluster. Its optical-link budget drives building geometry.</p>
</article>

<article id="control-nodes">
  <h3>Control Nodes</h3><p>Server Hardware Configuration</p>Compute & Network

  <p>A Control Node is a defined server hardware configuration (SKU) that provides non-GPU compute capacity for cluster operations and customer workloads. Control nodes are fungible: they can be provisioned for IaaS management functions or customer roles. The total control-node count includes management-role nodes, customer-role nodes, and warm spares.</p>
</article>

<article id="crah">
  <h3>CRAH</h3><p>Computer Room Air Handler</p>Cooling

  <p>An air-handling unit in the <a href="#mechanical-gallery">mechanical gallery</a> that uses medium-temperature facility water to cool remaining air-cooled equipment and occupied spaces.</p>
</article>

<article id="cub">
  <h3>CUB</h3><p>Central Utility Building</p>Cooling

  <p>The mechanical plant: chillers, pumps, and heat-rejection equipment that produce and circulate facility water. Two CUBs flank the AI Factory building.</p>
</article>

<article id="dry-coolers">
  <h3>Dry Coolers</h3><p>Dry Coolers</p>Cooling

  <p>Outdoor finned-tube heat exchangers that reject heat to ambient air with no evaporative water use.</p>
</article>

<article id="du">
  <h3>DU</h3><p>Deployment Unit</p>Compute & Network

  <p>A repeatable cluster pattern containing GPU racks, support racks, and CDUs.</p>
</article>

<article id="foc">
  <h3>FOC</h3><p>Facilities Operations Center</p>Site & Ops

  <p>Command center for monitoring and coordinating power, cooling, and other facilities operations.</p>
</article>

<article id="gpu">
  <h3>GPU</h3><p>Graphics Processing Unit</p>Compute & Network

  <p>The compute engine of the AI Factory. This facilities overview covers MGX Gen 1.1 and Vera Rubin NVL72 configurations, with cabinet TDP from 198 kW to 330 kW.</p>
</article>

<article id="hac">
  <h3>HAC</h3><p>Hot Aisle Containment</p>Cooling

  <p>A data-hall cooling arrangement that contains hot exhaust air and keeps it separate from supply air.</p>
</article>

<article id="hss">
  <h3>HSS</h3><p>High-Speed Storage</p>Compute & Network

  <p>Storage tier in the Network Core that feeds training data and checkpoints to GPU halls.</p>
</article>

<article id="it">
  <h3>IT</h3><p>Information Technology</p>Site & Ops

  <p>In this overview, IT load is the electrical demand of compute, network, and storage equipment, distinguished from mechanical, cooling, and other facility loads.</p>
</article>

<article id="idf">
  <h3>IDF</h3><p>Intermediate Distribution Frame</p>Compute & Network

  <p>Local distribution point that connects facility cabling and devices back to the MDF or Network Core.</p>
</article>

<article id="mechanical-gallery">
  <h3>Mechanical Gallery</h3><p>Mechanical Gallery</p>Cooling

  <p>A dedicated service area within the building that houses CDUs, CRAHs, piping, and other cooling-distribution equipment serving the data halls and occupied spaces.</p>
</article>

<article id="maxlps">
  <h3>NVIDIA DSX MaxLPS</h3><p>NVIDIA DSX MaxLPS</p>Power & Operations

  <p>NVIDIA DSX MaxLPS is an AI factory design and operating framework that coordinates facilities and site design, NVIDIA Dynamic Power Software (DPS), and advanced performance-per-watt techniques to help AI factory operators maximize performance per watt within a fixed power budget.</p>
</article>

<article id="mdf">
  <h3>MDF</h3><p>Main Distribution Frame</p>Compute & Network

  <p>Primary facility point for terminating and cross-connecting network services.</p>
</article>

<article id="mmr">
  <h3>MMR</h3><p>Meet Me Room</p>Compute & Network

  <p>The main point of entry where external carrier fiber lands on the campus and cross-connects to tenant networks.</p>
</article>

<article id="msg">
  <h3>MSG</h3><p>Main Switchgear</p>Power

  <p>The 34.5 kV main switchgear stage immediately downstream of the 230/34.5 kV transformers. Four MSGs operate in N+1 configuration.</p>
</article>

<article id="mv">
  <h3>MV / HV / LV</h3><p>Medium / High / Low Voltage</p>Power

  <p>MV = 34.5 kV class. HV = 230 kV utility transmission. LV = 480 V switchboards and below. The DSX backbone is MV; ride-through and cooling support live on LV.</p>
</article>

<article id="mvs">
  <h3>MVS</h3><p>Medium Voltage Switchboard</p>Power

  <p>The next stage down from the MSGs, distributing 34.5 kV into each building block. Connected in a loop with one end Normally Closed and the other Normally Open.</p>
</article>

<article id="n-plus">
  <h3>N+1 / N+2</h3><p>Redundancy levels</p>Power

  <p>N is the number of units required to support the designated load. N+1 provides one additional unit; N+2 provides two additional units.</p>
</article>

<article id="noc">
  <h3>NOC</h3><p>Network Operations Center</p>Site & Ops

  <p>Command center for monitoring campus network connectivity and IT infrastructure.</p>
</article>

<article id="ot">
  <h3>OT</h3><p>Operational Technology</p>Site & Ops

  <p>Plant-control side of the campus, including BMS, power telemetry, and generator controls.</p>
</article>

<article id="soc">
  <h3>SOC</h3><p>Security Operations Center</p>Site & Ops

  <p>Command center for monitoring and responding to physical and cybersecurity security events across the campus.</p>
</article>

<article id="smn">
  <h3>SMN</h3><p>Secure Management Network</p>Compute & Network

  <p>Network segment used for secure infrastructure management and operations traffic.</p>
</article>

<article id="su">
  <h3>SU</h3><p>Scalable Unit</p>Compute & Network

  <p>The repeatable DSX deployment building block. In this overview, each Scalable Unit combines compute and support capacity that drives data hall, power, cooling, and network sizing.</p>
</article>

<article id="tan">
  <h3>TAN</h3><p>Tenant Access Network</p>Compute & Network

  <p>The tenant-facing north-south network in the Network Core building.</p>
</article>

<article id="tcs">
  <h3>TCS</h3><p>Technology Cooling System</p>Cooling

  <p>Rack-level cooling loop connected through CDUs to the facility cooling loop.</p>
</article>

<article id="ups">
  <h3>UPS</h3><p>Uninterruptible Power Supply</p>Power

  <p>415 V UPS systems support Core network racks and indoor mechanical equipment for ride-through until backup generation is online. Not provided for primary GPU compute loads.</p>
</article>

<article id="xfmr">
  <h3>XFMR</h3><p>Transformer</p>Power

  <p>Power transformer. Used throughout: 230/34.5 kV at the substation, 34.5/0.48 kV at the building, and 4.16 kV step-downs for chillers.</p>
</article>

<article id="4m3">
  <h3>4M3</h3><p>Four-to-make-three</p>Power

  <p>Four parallel units sized so any three can carry the full load. Used at the rack-power-shelf level on GPU buildings.</p>
</article>

#### learn-more

<h2>
  Resource Path
</h2>

<ol>
  <li>
    1

    <strong>Design</strong>
  </li>

  <li>
    2

    <strong>Simulate</strong>
  </li>

  <li>
    3

    <strong>Build</strong>
  </li>

  <li>
    4

    <strong>Operate</strong>
  </li>
</ol>

<p>
  This is a high-level path, not a linear handoff. Teams typically revisit each step as an AI factory evolves, with DSX Sim supporting both design validation and ongoing facility operations.
</p>

<h2>
  1

  Design with NVIDIA DSX Reference Designs
</h2>

<p>
  Use this public overview to align internally on the architecture decisions that shape an AI factory. When a project needs deployment-specific reference-design guidance, engage NVIDIA. The resources below include selected reference designs and supporting design guidance from the broader portfolio of generation-specific, validated NVIDIA DSX AI factory architectures.
</p>

<p>
  {FACILITIES_LOCKED_RESOURCE_NOTE}
</p>

<a href="https://partners.nvidia.com/DocumentDetails?DocID=1151654" target="_blank" rel="noopener noreferrer">


  Reference design · NVOnline access

  NVIDIA Vera Rubin NVL72 Reference Design

  <p>(NVOnline #1151654)</p>

  Access reference design
</a>

<a href="https://partners.nvidia.com/DocumentDetails?DocID=1145739" target="_blank" rel="noopener noreferrer">


  Reference design · NVOnline access

  NVIDIA DSX - Vera Rubin Facilities Infrastructure Reference Design

  <p>(NVOnline #1145739)</p>

  Access reference design
</a>

<a href="https://partners.nvidia.com/DocumentDetails?DocID=1152370" target="_blank" rel="noopener noreferrer">


  Reference design · NVOnline access

  NVIDIA DSX Facilities Infrastructure Design Guide

  <p>(NVOnline #1152370)</p>

  Access reference design
</a>

<h2>
  2

  Simulate with NVIDIA DSX Sim
</h2>

<p>
  Use NVIDIA DSX Sim throughout the resource path: model and validate design choices before deployment, test configurations and changes, and revisit the digital twin as the facility operates. Choose the blueprint, SimReady assets, or DSX Air according to whether the work centers on facility and digital-twin context, asset preparation, or IT and network scenarios.
</p>

<a href="https://docs.omniverse.nvidia.com/dsx/latest/index.html" target="_blank" rel="noopener noreferrer">
  Simulation · DSX Sim

  NVIDIA Omniverse DSX Blueprint

  <p>Explore digital-twin workflows that connect facilities, hardware, and operations before and after deployment.</p>

  Explore the blueprint
</a>

<a href="https://nvidia-omniverse.github.io/aif-pipeline-samples/" target="_blank" rel="noopener noreferrer">
  Simulation · SimReady Assets

  AI Factory Digital Twin Pipeline Samples

  <p>Sample scripts and presets for creating DSX SimReady USD assets, covering CAD ingestion, optimization, validation, and metadata workflows for digital twin and AI factory applications.</p>

  Explore SimReady assets
</a>

<a href="https://docs.nvidia.com/networking-ethernet-software/nvidia-air/" target="_blank" rel="noopener noreferrer">
  Simulation · DSX Air

  NVIDIA DSX Air

  <p>Use cloud-hosted data center simulation to create IT infrastructure digital twins, validate configurations, and test network architectures.</p>

  Explore DSX Air
</a>

<h2>
  3

  Build with Validated Ecosystem Infrastructure
</h2>

<p>
  Use the NVIDIA DSX AI Factory Marketplace to identify infrastructure products evaluated against NVIDIA functional requirements for AI factory applications. For energy-storage solutions, review the BESS Self-Qualification Guidelines. Treat these resources as inputs to deployment-specific component selection, not as a prescribed bill of materials.
</p>

<a href="https://marketplace.nvidia.com/en-us/enterprise/dsx-infrastructure/" target="_blank" rel="noopener noreferrer">
  Infrastructure · Marketplace

  NVIDIA DSX AI Factory Marketplace

  <p>Explore products validated to meet NVIDIA functional requirements for AI factory applications.</p>

  Explore validated components
</a>

<a href="/dsx/facilities-infra/bess/overview">
  Energy storage · DSX Flex

  BESS Self-Qualification Guidelines

  <p>Review the partner-run qualification process for BESS supporting AI load buffering, demand response, and ride-through.</p>

  Read the guidelines
</a>

<h2>
  4

  Operate with Power and Operations Integration
</h2>

<p>
  Plan operational integration as the facility moves from design assumptions to live signals: connect BMS data with IT-side services where coordinated response is needed. The resource below illustrates a facility-to-IT coordination path, not a complete operating model; other telemetry, control, and automation decisions depend on the deployment. Reach out to NVIDIA and/or the appropriate facilities, energy, and technology partners where it makes sense to co-design these integration patterns.
</p>

IT/OT · DSX ExchangeBMS Integration Companion Guide<p>See the data contract and implementation context for connecting BMS data to DSX Exchange.</p>Read the integration guide

<aside>
  <h3>Public ecosystem examples</h3>

  <p>DSX is designed to bring together NVIDIA capabilities and ecosystem technologies in a shared AI factory context. See the public <a href="https://nvidianews.nvidia.com/news/nvidia-releases-vera-rubin-dsx-ai-factory-reference-design-and-omniverse-dsx-digital-twin-blueprint-with-broad-industry-support" target="_blank" rel="noopener noreferrer">DSX reference-design and digital-twin announcement</a> for examples of how NVIDIA and partners are applying shared design, simulation, power, cooling, and operations patterns.</p>
</aside>