SI Firms Linking AIDC Design, Build, and Operations Draw Attention for Integrated Capabilities
IT DAILY ·
✦ AI Summary
Domestic SI companies are strengthening infrastructure build and operations competitiveness, along with customer-tailored integrated design and operational support, in line with the expansion of the AIDC market.
Gartner said many expensive GPU clusters operate at less than 15% to 20% efficiency, and analyzed that GPU utilization is affected by network, storage, and resource management methods.
Snetsystem and GS ITM emphasized AIDC operation optimization beyond deployment through integrated GPU, network, and storage operations, power and cooling design, and control systems.
As the AIDC market expands, domestic SI companies are pushing to strengthen their competitiveness in infrastructure build and operations. Based on the technology and experience they have accumulated in their existing IT infrastructure businesses, they are showing a trend of reinforcing capabilities that cover design, build, and operations.
Domestic SI companies are expanding integrated design tailored to customers' AI workloads, while also expanding operational support tailored to customers' AI workloads. This is leading to efforts to strengthen both infrastructure design and overall operations in line with customer demand in the AIDC environment.
To improve the returns on AIDC investment, waste must be reduced in the actual usage process as well as through GPU performance. Simply securing expensive GPUs does not guarantee sufficient results, and if actual computing utilization is insufficient, the return on investment may fall short.
Gartner released the "2026 GPU Optimization Playbook" in May. Gartner noted that many expensive GPU clusters operate at less than 15% to 20% efficiency, and analyzed that low GPU utilization is a major issue causing financial waste in enterprise AI.
GPU utilization is affected not only by the computing device itself but also by surrounding infrastructure and resource management methods. If data transfer over the network and storage is delayed, GPU idle time increases. Accordingly, it is necessary to understand how GPUs assigned to each task are being used, and a GPU reallocation system that responds to demand is also needed.
In the same context, AIDC needs to reflect the actual operating scale of GPUs and workload characteristics in its power and cooling design. The Korea Energy Economics Institute published "Implications of Data Center Growth in the AI Era for Domestic Energy Consumption" late last year. The institute said the fluctuation range of AIDC power consumption is greater than that of conventional data centers, and suggested improving equipment utilization, designing power management, and optimizing workload and resource management as ways to enhance data center energy efficiency. Accordingly, AIDC needs to check IT infrastructure resource utilization and facility conditions at the same time.
As these conditions become important together, the role of SI companies with strengths in integrating a wide range of equipment and solutions into a single system is expanding.
Snetsystem supports the build and operation of integrated AI computing environments spanning GPU, network, and storage, and also handles operational support after deployment. The company has accumulated experience operating GPUaaS infrastructure for large service providers, and based on that experience it supports the stable and efficient use of customer-deployed equipment in actual services.
To do this, Snetsystem uses GPU resource management platforms such as NVIDIA Run:ai. This supports GPU sharing among multiple users and projects, allocates resources based on task priority and demand, and helps secure the resources needed for each project while reducing unused GPUs.
In surrounding infrastructure design, the focus is on timely supply of data for GPU computing. In large-scale training environments, the emphasis is on enabling smooth data exchange between GPUs, while in inference service environments, servers, networks, and storage are configured with response speed and concurrent request handling capability in mind.
When a failure occurs during operations, individual equipment is inspected. At the same time, the linkage status of GPU, network, and storage is also checked.
Snetsystem said that to raise the effectiveness of AI infrastructure investment, optimizing utilization and operational support are more important than simply introducing resources. A Snetsystem official said that investment in IT resources such as expensive GPUs has limits in improving returns on investment if those resources are not used efficiently.
The official added that the company aims to develop into a partner that supports stable operations by making infrastructure and services more specific according to each customer's AI use purpose.
GS ITM is focusing on improving infrastructure utilization efficiency through linkage between AIDC design, build, and operations. A representative example is the Doksan Data Center, an edge-type AIDC of about 4 MW, which is currently under development.
In that project, GS ITM is responsible for the overall IT infrastructure from the design stage. The starting point of AIDC design is determining the AI workloads to be performed and the GPU configuration, after which the design adjusts power supply and cooling based on the power consumption and heat generation of each rack. In this process, design considerations include responding to fluctuations in high-density GPU loads, accommodating future GPU replacements, and allowing for expansion that can support a transition to liquid cooling.
In the operations stage, GS ITM builds a control system that integrates IT infrastructure and facility data to understand GPU utilization rates and power efficiency, while also handling failure response. As the foundation for this, it cited 20 years of experience in IT outsourcing and integrated operations of group company infrastructure, along with a generative AI-based ITSM platform.
GS ITM said the role of SI companies will change from build contractors to AI infrastructure partners responsible for planning through operations. Accordingly, for supporting customer AX, it said it will center on a city-type AIDC integrated SI model that standardizes the know-how of the Doksan Data Center.
GS ITM presented AI-based integrated monitoring, operational automation, energy diagnosis, and high-efficiency initiatives as means of supporting customer AX. It said it will strengthen a support system that covers the operations stage after deployment.
Meanwhile, there were also suggestions that long-term approaches that consider the period after infrastructure securing are needed for the AIDC industry to grow sustainably. Snetsystem emphasized that time is needed from infrastructure build to actual service use and results creation, and explained that rather than judging investment value based on short-term performance, it is necessary to accumulate technical and operational experience and create new demand.
Snetsystem said it sees the importance of connecting the field experience and expertise of domestic manufacturing with AI services. Examples of AI services that solve field problems included quality inspection and process optimization. Snetsystem said the spread of such services is necessary to expand the value of AIDC utilization.
For this, Snetsystem said support for cooperation between manufacturing companies and AI and infrastructure companies is needed. The scope of cooperation covers the entire process from identifying tasks to service development, validation, and commercialization. A Snetsystem official said sustainable growth in the AIDC industry is possible if infrastructure investment and demand from industrial sites create a virtuous cycle.
GS ITM suggested that companies participating in AIDC need a culture of collaboration from the early design stage. GS ITM said there is a need for a cooperative foundation that goes beyond securing individual technologies and turns them into a single system. A GS ITM official said AIDC is a single system combining IT, power, cooling, and operations.
GS ITM pointed out that conditions supporting business are needed for collaboration to continue. A GS ITM official said transparent communication with the local community is necessary, and added that expansion of power infrastructure and consistent permitting standards are also needed. The official also said that industry and government attention is needed to foster converged talent.
Source: IT DAILY · Yang Seung-gap
Original: https://www.itdaily.kr/news/articleView.html?idxno=242693
References
This article was produced with the help of an automated content generation algorithm.
Source: IT DAILY
View originalThis article was summarized and organized by BizCrush based on the original article from IT DAILY. For exact quotations and full details, please refer to the original article.