--- title: "What Is a Modular AI Data Center? | Aguasto" description: "Distinguish a facility module, equipment assembly and GPU rack, then define the interfaces that connect an enclosure to usable AI compute." canonical: "https://www.aguasto.com/learn/what-is-a-modular-ai-data-center" last-updated: "2026-10-13" --- # What Is a Modular AI Data Center? | Aguasto Canonical HTML: [What Is a Modular AI Data Center?](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center) > Distinguish a facility module, equipment assembly and GPU rack, then define the interfaces that connect an enclosure to usable AI compute. [Aguasto](https://www.aguasto.com/) [learn](https://www.aguasto.com/learn) What Is a Modular AI Data Center? [Resources](https://www.aguasto.com/resources) [GPU Library](https://www.aguasto.com/gpus) [Model Index](https://www.aguasto.com/models) [Inference Stack](https://www.aguasto.com/inference-stack) [Learn](https://www.aguasto.com/learn) [Insights](https://www.aguasto.com/insights) [Research](https://www.aguasto.com/research) [Engineering tools](https://www.aguasto.com/tools) Aguasto / learn ## What Is a Modular AI Data Center? Distinguish a facility module, equipment assembly and GPU rack, then define the interfaces that connect an enclosure to usable AI compute. Original Aguasto conceptual illustration. Not manufacturer geometry, an operating Aguasto asset, or a measured benchmark. By Aguasto Research · Published October 13, 2026 · Revision 1 Aguasto Research documents third-party hardware, models and software. References do not represent Aguasto inventory, evaluated deployments or available inference services. Illustration: Conceptual modular facility. Original isometric illustration for the module assembly development stage. A proposed enclosure shows equipment spaces and a central service aisle as systems are assembled. The enclosure is cut away to reveal its interior. Components are shown in their illustrative assembled positions. All systems are shown. No geographic intelligence layers are selected.No real property, equipment inventory, capacity, service right or operating Aguasto facility is represented. Physically installed electrical equipment does not establish usable capacity or utility authorization. Original generic module study. Racks, electrical assemblies and cooling are different system boundaries. Concept geometry; no equipment specification or operating Aguasto asset. ### Define the unit before the benefit Prefabrication describes assembly or integration performed away from the final installation site. Modularity describes a system divided into bounded, repeatable units. Containerized describes one possible enclosure. Schneider Electric’s taxonomy distinguishes IT-space, power and cooling modules; the terms should not be treated as interchangeable. A prefabricated electrical room and a container holding GPU racks solve different portions of a facility problem. For a buyer, the useful first question is what the supplier’s boundary includes. Ask for an inclusion schedule covering enclosure, electrical distribution, cooling equipment, controls and installation interfaces. Mark what remains with the site developer, utility, rack supplier and operator. A product called turnkey still needs a defined handover boundary. Our analysis treats the module as an engineering package whose value depends on those connections, rather than using its exterior shape as a readiness signal. [Types of Prefabricated Modular Data Centers, revision2 — Schneider Electric (external source)](https://www.se.com/ww/en/download/document/SPD_WTOL-97GLP9_EN/) · [Prefabricated modular data centers: terminology — Schneider Electric (external source)](https://blog.se.com/datacenter/2013/12/02/prefabricated-modular-data-centers-sorting-terminology-confusion/) ### Keep facility modules and GPU systems on different levels An accelerator, a rack-scale GPU system and a facility module are different reference units. NVIDIA’s GB200 SuperPOD architecture describes a particular rack deployment with power, cooling and network dependencies. It provides an example of requirements that a facility must accommodate; it does not establish a universal configuration for every AI enclosure. OCP’s cooling work likewise makes integration boundaries central to high-density systems. Build a requirements record from the proposed workload upward: checkpoint and runtime, intended concurrency, supported GPU configuration, rack interfaces and facility services. Keep the identity of each system visible as the record crosses these levels. Comparing a rack’s published memory with an individual accelerator’s memory can conceal an invalid comparison. Similarly, counting containers says little about usable compute until their equipment and acceptance scope are known. [DGX SuperPOD Architecture for DGXGB200 — NVIDIA (external source)](https://docs.nvidia.com/dgx-superpod/reference-architecture-scalable-infrastructure-gb200/latest/dgx-superpod-architecture.html) · [Cooling Environments — Open Compute Project (external source)](https://www.opencompute.org/projects/cooling-environments/) ### Draw an interface contract OCP’s integration guidance distinguishes facility and technology cooling circuits, including the coolant distribution unit interface. Prefabricated facility suppliers describe modular arrangements spanning power, cooling and IT space. Together these references suggest a useful coordination question: which team supplies each side of a connection, and what evidence demonstrates that the two sides work together? Our proposed interface worksheet has five columns: boundary, upstream requirement, downstream requirement, responsible reviewer and acceptance evidence. Example rows are facility supply to electrical assembly, cooling plant to distribution unit, distribution unit to rack, rack to network and cluster to customer service. Leave values unfilled until project engineering validates them. The worksheet is a way to identify missing decisions, not an Aguasto procurement specification or a substitute for qualified design. [Liquid Cooling Integration and Logistics White Paper, revision1.0 — Open Compute Project (external source)](https://www.opencompute.org/documents/ocp-liquid-cooling-integration-and-logistics-white-paper-revision-1-0-1-pdf) · [Prefabricated Data Center brochure — Vertiv (external source)](https://www.vertiv.com/4a7618/globalassets/shared/vertiv-prefabricated-data-center-brochure.pdf) ### Factory and site work can overlap Schneider Electric describes prefabrication enabling factory integration alongside site preparation. This changes where selected work occurs and allows some workstreams to overlap. It does not make every task independent: final equipment requirements affect interfaces, and the site join still requires completed external services. A credible sequence therefore shows two branches meeting at integration, followed by acceptance and operating handover. Use a dependency diagram before attaching calendar estimates. Identify which decisions must be frozen before an assembly can leave the factory, what site information is required, and how a change is approved after fabrication begins. If the interface is unstable, earlier factory work can become earlier rework. Buyers should request the assumptions behind a proposed schedule and the responsibility for late changes. This is more informative than adopting a generic speed percentage from a different project. [Prefabricated Data Center brochure — Vertiv (external source)](https://www.vertiv.com/4a7618/globalassets/shared/vertiv-prefabricated-data-center-brochure.pdf) · [Network edge enabled by prefabricated modular data centers — Schneider Electric (external source)](https://blog.se.com/datacenter/2022/05/17/supporting-network-edge-prefabricated-modular-data-centers/) ### Repeatable does not automatically mean movable OCP’s logistics guidance treats transport, handling and site integration as engineering concerns. A previously assembled cooling system may require preparation and checks around shipping and installation. Schneider Electric’s discussion of modular benefits should be read with the particular configuration and site conditions in view. Modularity alone does not guarantee interchangeability, portability or unlimited expansion. An expansion review should ask whether a later unit can be connected under the existing operating conditions, who controls the connection process and what additional infrastructure the new phase needs. A relocation review should define transport and reinstatement scope rather than assuming the enclosure can simply be moved. In both reviews, preserve the original acceptance record and document changes. Repeatability is strongest when the design baseline, configuration history and handover evidence remain coherent. [Liquid Cooling Integration and Logistics White Paper, revision1.0 — Open Compute Project (external source)](https://www.opencompute.org/documents/ocp-liquid-cooling-integration-and-logistics-white-paper-revision-1-0-1-pdf) · [Benefits and Drawbacks of Prefabricated Modules, revision4 — Schneider Electric (external source)](https://www.se.com/us/en/download/document/SPD_WTOL-7NGRBS_EN/) ### Evaluate the compute service at the end of the chain Our buyer worksheet closes with six questions: what exactly is supplied, which interfaces remain unresolved, what approvals are required, what factory evidence exists, what site acceptance remains, and which customer workload will establish useful service? The answers should connect the enclosure to an operating responsibility boundary. A GPU rental customer and a managed inference customer will require different service evidence even if their workloads use the same physical facility. Use the modular facility concept as the beginning of an investigation. Read the commissioning guide for the progression from factory checks to site acceptance, inspect the GPU Library for manufacturer boundaries, and use the inference benchmark methodology to define future workload measurements. Aguasto is developing this integrated approach. This publication explains a planning method; it does not represent an installed module, operating GPU inventory or a proven deployment duration. ### Scope & limitations - Original generic diagrams explain boundaries; they are not certified construction drawings or exact OEM equipment. - No Aguasto installation or deployment-time measurement is represented. - Actual interface values, approvals and acceptance procedures require project-specific engineering. ### Continue the research [How Modular AI Data Centers Are Built and Commissioned](https://www.aguasto.com/learn/modular-data-center-factory-to-commissioning) [Why tokens per second is not enough to benchmark inference](https://www.aguasto.com/research/inference-benchmark-methodology) [NVIDIA GB200 NVL72](https://www.aguasto.com/gpus/gb200-nvl72) [The Aguasto modular thesis](https://www.aguasto.com/why-modular) [Potential OEM and strategic collaboration](https://www.aguasto.com/partners) ### Sources & verification The register preserves each source's publication and verification date. Manufacturer claims and documentation are attributed; Aguasto has not run the cited products in production. - [Types of Prefabricated Modular Data Centers, revision2 — Schneider Electric ↗](https://www.se.com/ww/en/download/document/SPD_WTOL-97GLP9_EN/) Prefabricated systems/subsystems terminology and form-factor distinctions. Publication: 2017-09-29. Verified: 2026-10-13. - [Prefabricated modular data centers: terminology — Schneider Electric ↗](https://blog.se.com/datacenter/2013/12/02/prefabricated-modular-data-centers-sorting-terminology-confusion/) Function/form-factor/configuration taxonomy. Publication: 2013-12-02. Verified: 2026-10-13. - [DGX SuperPOD Architecture for DGXGB200 — NVIDIA ↗](https://docs.nvidia.com/dgx-superpod/reference-architecture-scalable-infrastructure-gb200/latest/dgx-superpod-architecture.html) DGXGB200 uses both direct-liquid and air cooling; Reference architecture ties systems, networks, management and storage. Last modified: 2025-11-19. Verified: 2026-10-13. - [Cooling Environments — Open Compute Project ↗](https://www.opencompute.org/projects/cooling-environments/) Cold-plate/TCS scope includes tubing/manifolds/couplings/CDUs; Different cooling architectures and standardization work. Publication: date not stated in the reviewed source. Verified: 2026-10-13. - [Liquid Cooling Integration and Logistics White Paper, revision1.0 — Open Compute Project ↗](https://www.opencompute.org/documents/ocp-liquid-cooling-integration-and-logistics-white-paper-revision-1-0-1-pdf) FWS/TCS/CDU terminology and liquid-to-liquid isolation; Manufacturing, transport, inspection and commissioning handoffs; Wetted materials and project-specific tests. Publication: date not stated in the reviewed source. Verified: 2026-10-13. - [Prefabricated Data Center brochure — Vertiv ↗](https://www.vertiv.com/4a7618/globalassets/shared/vertiv-prefabricated-data-center-brochure.pdf) Factory/site/service integration phases; Engineering, acceptance plans, installation and commissioning roles. Publication: date not stated in the reviewed source. Verified: 2026-10-13. - [Network edge enabled by prefabricated modular data centers — Schneider Electric ↗](https://blog.se.com/datacenter/2022/05/17/supporting-network-edge-prefabricated-modular-data-centers/) Factory assembly and site work may proceed in parallel. Publication: 2022-05-17. Verified: 2026-10-13. - [Benefits and Drawbacks of Prefabricated Modules, revision4 — Schneider Electric ↗](https://www.se.com/us/en/download/document/SPD_WTOL-7NGRBS_EN/) Standardized factory integration shifts work toward site integration; Advantages are compared with drawbacks. Publication: 2026-02-03. Verified: 2026-10-13. ### In this publication [Define the unit before the benefit](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center#section-1) [Keep facility modules and GPU systems on different levels](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center#section-2) [Draw an interface contract](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center#section-3) [Factory and site work can overlap](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center#section-4) [Repeatable does not automatically mean movable](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center#section-5) [Evaluate the compute service at the end of the chain](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center#section-6) [More learn ↗](https://www.aguasto.com/learn) [Source register ↓](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center#sources) ## Company navigation - [Home](https://www.aguasto.com/) - [Infrastructure](https://www.aguasto.com/infrastructure) - [GPU Cloud](https://www.aguasto.com/gpu-cloud) - [AI Inference](https://www.aguasto.com/ai-inference) - [Development & Investment](https://www.aguasto.com/development) - [Company](https://www.aguasto.com/company) - [Contact](https://www.aguasto.com/contact) - [Privacy](https://www.aguasto.com/privacy) - [Resources](https://www.aguasto.com/resources) - [GPU Library](https://www.aguasto.com/gpus) - [Model Index](https://www.aguasto.com/models) - [Inference Stack](https://www.aguasto.com/inference-stack) - [Learn](https://www.aguasto.com/learn) - [Insights](https://www.aguasto.com/insights) - [Research](https://www.aguasto.com/research) - [Industry cases](https://www.aguasto.com/research/cases) - [Why Modular](https://www.aguasto.com/why-modular) - [Partners](https://www.aguasto.com/partners) - [GPU comparison](https://www.aguasto.com/gpus/compare) - [Model comparison](https://www.aguasto.com/models/compare) - [Engineering tools](https://www.aguasto.com/tools) - [Development friction](https://www.aguasto.com/research/hyperscale-development-friction) - [NVIDIA H100 SXM](https://www.aguasto.com/gpus/h100) - [NVIDIA H200 SXM](https://www.aguasto.com/gpus/h200) - [NVIDIA B200 SXM](https://www.aguasto.com/gpus/b200) - [NVIDIA B300](https://www.aguasto.com/gpus/b300) - [NVIDIA GB200 NVL72](https://www.aguasto.com/gpus/gb200-nvl72) - [NVIDIA GB300 NVL72](https://www.aguasto.com/gpus/gb300-nvl72) - [NVIDIA Vera Rubin NVL72](https://www.aguasto.com/gpus/vera-rubin-nvl72) - [AMD Instinct MI300X](https://www.aguasto.com/gpus/mi300x) - [AMD Instinct MI325X](https://www.aguasto.com/gpus/mi325x) - [AMD Instinct MI355X](https://www.aguasto.com/gpus/mi355x) - [AMD Instinct MI455X](https://www.aguasto.com/gpus/mi455x) - [AMD Helios](https://www.aguasto.com/gpus/helios) - [GLM-4.5](https://www.aguasto.com/models/glm-4-5) - [Qwen3-235B-A22B](https://www.aguasto.com/models/qwen3-235b-a22b) - [DeepSeek-R1](https://www.aguasto.com/models/deepseek-r1) - [Llama 3.3 70B Instruct](https://www.aguasto.com/models/llama-3-3-70b-instruct) - [Gemma 3 27B IT](https://www.aguasto.com/models/gemma-3-27b-it) - [gpt-oss-120b](https://www.aguasto.com/models/gpt-oss-120b) - [Kimi-K2-Instruct](https://www.aguasto.com/models/kimi-k2-instruct) - [GLM-5.3](https://www.aguasto.com/models/glm-5-3) - [Qwen3.8-27B](https://www.aguasto.com/models/qwen3-8-27b) - [DeepSeek-V4.1-Flash](https://www.aguasto.com/models/deepseek-v4-1-flash) - [Llama 4 Scout 17B-16E Instruct](https://www.aguasto.com/models/llama-4-scout-17b-16e-instruct) - [Gemma 4 31B IT](https://www.aguasto.com/models/gemma-4-31b-it) - [Kimi K3](https://www.aguasto.com/models/kimi-k3) - [vLLM](https://www.aguasto.com/inference-stack/vllm) - [SGLang](https://www.aguasto.com/inference-stack/sglang) - [TensorRT-LLM](https://www.aguasto.com/inference-stack/tensorrt-llm) - [NVIDIA Dynamo](https://www.aguasto.com/inference-stack/dynamo) - [NVIDIA Triton Inference Server](https://www.aguasto.com/inference-stack/triton) - [AMD ROCm](https://www.aguasto.com/inference-stack/rocm) - [Ray Serve](https://www.aguasto.com/inference-stack/ray-serve) - [Kubernetes GPU orchestration — NVIDIA device plugin](https://www.aguasto.com/inference-stack/kubernetes) - [Model weights are only the beginning of the memory budget](https://www.aguasto.com/learn/model-weights-and-runtime-memory) - [Why tokens per second is not enough to benchmark inference](https://www.aguasto.com/research/inference-benchmark-methodology) - [Prefill, decode and the cost of separating them](https://www.aguasto.com/learn/prefill-decode-and-disaggregated-serving) - [Choosing between vLLM, SGLang and TensorRT-LLM](https://www.aguasto.com/learn/choosing-an-inference-serving-framework) - [From Power to Inference—Building a Distributed AI Compute Network](https://www.aguasto.com/learn/from-power-to-inference) - [Dedicated compute and shared inference: compare the whole workload](https://www.aguasto.com/insights/dedicated-and-shared-inference-economics) - [Distributed inference reliability begins with the failure boundary](https://www.aguasto.com/learn/distributed-inference-reliability) - [Core Scientific: reading contracted, energized and billed capacity](https://www.aguasto.com/research/cases/core-scientific-capacity-disclosures) - [Why Modular AI Infrastructure Matters Now](https://www.aguasto.com/research/why-modular-ai-infrastructure-matters-now) - [What Is a Modular AI Data Center?](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center) - [How Modular AI Data Centers Are Built and Commissioned](https://www.aguasto.com/learn/modular-data-center-factory-to-commissioning) - [Liquid Cooling for High-Density GPU Systems: Follow the Interfaces](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces) - [Why Electricity Availability Shapes AI Infrastructure Deployment](https://www.aguasto.com/research/electricity-availability-and-time-to-power) - [What Is KV Cache? Capacity, Reuse and Isolation](https://www.aguasto.com/learn/kv-cache-capacity-and-policy) - [Continuous Batching: Throughput Without Hiding Latency](https://www.aguasto.com/learn/continuous-batching-and-latency) - [Applied Digital: Reading the Polaris Forge 1 Delivery Milestones](https://www.aguasto.com/research/cases/applied-digital-polaris-forge-capacity-milestones) - [Crusoe Abilene: Separate Delivered Infrastructure from Expansion Plans](https://www.aguasto.com/research/cases/crusoe-abilene-delivery-and-expansion) - [Official LinkedIn](https://www.linkedin.com/company/aguasto) This is a generated representation of the public HTML page. Diagrams are conceptual. GPU and inference services are planned and are not currently available to book or call. The contact page describes the available inquiry flow.