Artificial intelligence (AI) data centers are expanding their scope from gathering servers in one hu..

Artificial intelligence (AI) data centers are expanding their scope from gathering servers in one huge facility to a decentralized structure that places computing resources where necessary. In line with AI reasoning and the spread of Sovereign AI, IA plans to grow a modular data center (MDC) that can quickly build power and cooling facilities into a new business.

IA announced on the 24th that it proposed that the Open Computing Project (OCP) needs to revise the "expanded 300kW MDC" standard established in 2019 to suit the AI computing environment at the "2026 OCP Korea Tech Day" held at COEX in Gangnam, Seoul on the 21st.

Unlike existing data centers, AI servers have recently been equipped with a large number of high-performance accelerators such as graphics processing units (GPUs) and neural network processing units (NPUs). It is explained that the power consumption and heat generation of existing MDCs should be redesigned for AI servers as the server's degree of integration increases.

On the same day, IA also unveiled its own modular AI data center reference design 'NeoCube'. It is a full stack structure that combines power and cooling infrastructure, AI accelerators such as GPUs and NPUs, as well as private clouds and AI platforms and services. It will link DCE Solution's thermal management and cooling technology with Trino Technology's power semiconductor and IA Cloud's cloud infrastructure (IaaS) technology.

It also brought up new indicators to evaluate the power efficiency of data centers along with AI computational performance. IA presented a draft "PF-eq" that represents the actual amount of AI computation compared to facility power. It is not the official standard adopted by the OCP as an initial concept currently presented by the IA.

The key is whether to demonstrate performance in an actual data center environment. IA is currently producing a 25kW Neocube concept verification (POC) facility. It plans to verify AI computational performance, power and cooling efficiency, and operational stability after being built early next month. After that, it will increase its scale to 100kW to promote market supply. The period from design to construction was set within three months.

The data obtained during the demonstration process will also be used to discuss the standardization of MDC by OCP. Han Sung-yong, CEO of AI, said, "AI data center is moving to a stage where not only scale but also power, cooling efficiency, construction speed, and actual computational efficiency must be evaluated together. Based on the empirical data of Neocube, we plan to participate in preparing MDC standards and evaluation standards suitable for AI environment."