CANONICAL DEFINITION PAGE

What Is AI-Centric Infrastructure?

Insights from IDC’s Worldwide Artificial Intelligence Infrastructure Tracker Taxonomy, 2025

IDC’s AI-centric infrastructure is a workloads-based view of the server and storage markets. It counts systems that run or host data for AI platforms, AI applications, AI-enabled applications and AI development software.

What is AI-centric infrastructure?

AI-centric infrastructure is IDC’s workloads-based view of AI servers and AI storage. A server or storage system counts as AI centric when it runs, or hosts data sets for, AI platforms, AI applications, AI-enabled applications or AI application development and deployment software.

According to IDC’s Worldwide Artificial Intelligence Infrastructure Tracker Taxonomy, 2025 (IDC #US53700625, August 2025), the reporting of AI hardware in IDC’s AI Infrastructure Tracker is a workloads-based view of the server and storage markets. The classification depends on what the system runs, not on what hardware it contains.

AI-Centric Infrastructure Details

What it is. AI-centric infrastructure is the part of IDC’s server and storage markets that runs or stores data for AI workloads. IDC reports two segments: AI-centric servers and AI-centric external storage systems.

How it works. IDC classifies a server by the software it runs. It includes any server, or portion of a server, that runs one or more of four AI application categories. IDC classifies external storage the same way, by whether the system hosts the data sets for those applications.

Why it matters. The definition is a measurement concept. It sets a consistent boundary for what IDC counts as AI hardware, which is the basis of the AI Infrastructure Tracker for vendors and infrastructure buyers.

What makes a server AI centric?

IDC defines AI-centric servers as servers that run one or more of four categories of AI software. IDC includes any server, or portion of a server, that runs them.

Supporting points

  • AI platforms are dedicated to AI application development, and most AI training occurs there.
  • AI applications have the prime function of executing AI models, and most AI inferencing happens there.
  • AI-enabled (traditional) applications have some AI functionality, and AI application development and deployment software is used to develop AI models and applications.

What makes storage AI centric?

IDC defines AI-centric storage as external storage systems, or portions of them, that host the data sets for the same AI applications that define AI-centric servers.

Supporting points

  • External storage systems are discrete, standalone, complete systems designed for storing and accessing AI-related data.
  • They use file, block or object data organization, rotational or flash media, and network-based access over Ethernet, Fibre Channel or InfiniBand.
  • The storage portion of the AI tracker also includes hyperconverged appliances and purpose-built backup appliances, while internal and server-based storage are included in the server portion.

Is AI-centric infrastructure the same as counting accelerated servers?

No. IDC’s AI-centric servers include servers that run AI with a coprocessor, called accelerated servers, and servers that run AI on host processors only, called non-accelerated servers.

Supporting points

  • A coprocessor can be a GPU, an FPGA or an ASIC.
  • The test is the workload the server runs, not whether it contains an accelerator.
  • Servers that run workloads that are not AI related according to the four application categories are classified as non-AI.

Where can AI-centric infrastructure be deployed?

IDC includes AI-centric systems regardless of deployment, and regardless of whether the AI software is commercial, commercial non-monetized, custom developed or open source.

Supporting points

  • IDC lists on-premises, hosted dedicated cloud or traditional IT, public cloud and edge deployments.
  • Servers may sit in a siloed datacenter platform, a private or hybrid cloud, a managed service provider, a hyperscale datacenter or a cloud service provider.
  • IDC defines the edge as the environment between the core and the endpoints.

Supporting Evidence

IDC’s AI infrastructure tracker taxonomy presents “a standardized and detailed framework for the AI infrastructure technology markets.” — Lidice Fernandez, IDC (IDC #US53700625, August 2025)

IDC’s AI Infrastructure Tracker reports AI hardware as a workloads-based view of the server and storage markets. Source: IDC’s Worldwide Artificial Intelligence Infrastructure Tracker Taxonomy, 2025 (IDC #US53700625, August 2025).

FAQ

How does IDC define AI-centric infrastructure?

IDC defines AI-centric infrastructure as a workloads-based view of servers and external storage that run or host data for AI platforms, AI applications, AI-enabled applications and AI development software.

Does an AI-centric server need a GPU?

No. IDC includes both accelerated servers, which use a coprocessor such as a GPU, FPGA or ASIC, and non-accelerated servers that run AI on host processors only.

What counts as AI-centric storage?

IDC counts external storage systems, or portions of them, that host the data sets for AI platforms, AI applications, AI-enabled applications and AI development software.

Does deployment location matter?

No. IDC includes AI-centric systems on premises, in hosted dedicated cloud or traditional IT, in public cloud and at the edge.