Energy savings behind cloud migration: an interview with Lim Yong-jun

AI-driven coding tools are often discussed in terms of productivity, yet there is an overlooked shift occuring in the servers, cooling systems, and idle infrastructure that cloud migration and automated scheduling are increasing efficiency. In this interview, Lim Yong-jun of SK AX discussed how moving from in-house servers to the cloud has cut energy waste from underused infrastructure and how IT infrastructure can build more sustainable AI.

First, could you tell us about the work you’re currently doing?

I’m currently with SK Corporation’s AX division. Our main responsibilities involve IT operations within SK; specifically, we focus on system integration (SI)—that is, external development—as well as the operation and improvement of existing systems. My area of responsibility used to cover a wide range of tasks, but recently I’ve been focusing primarily on the OS domain, managing the operating systems for SK Group companies, particularly those in the Innovation Division.

As AI coding services continue to advance, have there been any changes in the way you actually work? Could you tell us about those?

The company’s perspective and my personal perspective are on somewhat different levels, but from the perspective relevant to this interview topic—which is the company’s perspective—the impact is actually much stronger. While this isn’t directly within my area of responsibility, in the infrastructure sector, we’re primarily adopting cloud technologies. From an AI perspective, this means using AI to manage operational plans for systems—such as scheduling and data management in infrastructure data centers—through the adoption of the cloud. And for those of us in the OS or SI , those responsible for application operations or developers are primarily engaged in what’s called “AI coding”—using AI to automate the coding process.

Currently, our company is also focused on developing tools for AI coding. As a result, rather than viewing potential issues as problems, we’re focusing on the future benefits. In fact, IT departments are currently allocating significant resources to this specific workforce. Since operations are based on human labor, unit costs have skyrocketed, and we’ve faced situations where we had to pay more than double what we previously spent.

As a result, we’re now exploring ways to address these issues through AI coding. Since AI coding isn’t based on human standards—but rather on our accumulated know-how and specific prompts to automatically generate code—our reliance on human labor has significantly decreased, leading to benefits such as lower development costs and reduced expenses.

From the perspective of AI services, does building a more efficient system also lead to energy savings or improved efficiency?

Actually, regarding the workforce—as I mentioned earlier—there’s currently a bit of a gap between the company’s perspective and my own view on AI. As far as the work I’m currently responsible for goes, while you’re asking from an ESG perspective, it doesn’t really have a significant impact. If we were to force a connection here, since we’re doing AI-based development and coding using computers without human intervention, there are certainly costs associated with the space that people would otherwise occupy, as well as electricity and other resources—you know, the basic supplies needed when staff are present. For example, we’d need office space and PCs, and there would be commuting involved. When viewed in that broader context, those factors do exist, but strictly speaking, they aren’t major issues.

However, from the company’s perspective, they are quite significant. Basically, since most of our operations involve in-house development, when we developed systems internally, we had to purchase all the infrastructure—that is, server equipment—for each system. Now, as anyone who’s used a personal computer knows, you don’t use it to its full capacity. The same goes for systems—they aren’t used 100% of the time. We maintain this excess capacity for stability, and that excess capacity is actually quite substantial. Why is that? Because when purchasing infrastructure servers, there’s a minimum capacity requirement. As a result, from the company’s perspective, some systems end up being incredibly small. This actually results in a lot of unused, idle equipment. However, we’ve now migrated all of this to the cloud. The biggest advantage of the cloud is that the amount of infrastructure allocated varies depending on traffic levels during the day and night.

This allows us to eliminate all idle resources. For example, we manage a job application website. The job application site isn’t used at all during the rest of the year—it’s only used during the specific recruitment period. But the traffic volume during that time is enormous, with hundreds of thousands of applicants. So, we only need to scale up the infrastructure during that period, and we can take advantage of discounted rates. There’s no need to maintain it year-round. By managing things this way, we’re seeing more and more benefits emerge in the cloud environment. We also manage data centers, which require massive infrastructure. For these facilities, AI enables us to optimize cooling, predict server load, and shut down unnecessary servers—all through predictive analytics that allow for such scheduling.

Although we’re not a large company, one of the main advantages we’ve seen through our operations is definitely operational automation. This includes monitoring and proactively identifying issues like system failures, which increases operational efficiency. As a result, we’ve been able to reduce our workforce, and there’s been a significant decrease in resource waste. In fact, even before AI was introduced, one of the major changes we’d already implemented was significantly reducing our use of paper. Payment services and the like have all shifted to electronic payments, and meetings, contracts, and other documents are now all handled electronically. As for meetings—we’re in a conference room right now—we conduct many of them via video conferencing. We don’t have to move people around, and although we did have a period of remote work, that’s no longer the case now. Those are some of the advantages we’ve seen.

You’ve mentioned a wide range of benefits, but on the flip side, it seems that as cloud usage increases, power consumption at data centers is also rising. In the age of AI, what technologies do you think are important for improving energy efficiency?

My view is that, contrary to what you might think, power consumption hasn’t actually increased that much. In fact, it has decreased. The reason is, as I mentioned earlier—or rather, not at the very beginning but later on—the lifespan of these systems is long. Since there’s an abundance of idle equipment across various systems, a lot of unnecessary electricity was being consumed. By consolidating all of this into a single large unit—the “cloud”—and adopting a structure where resources are allocated incrementally from within, we’re moving in that direction.

What’s most needed here is effective scheduling. What I mean is, as I just mentioned, a job application site, for example, experiences a surge in traffic only during specific periods. Infrastructure needs to be allocated accordingly, but it’s impossible for humans to handle all this scheduling manually. I think what’s most needed is automation that schedules resources based on historical data—specifically, who accesses the system during off-peak hours, how much traffic there is, and how the resources are utilized. Based on this scheduling, we also need to manage server cooling—since servers generate a tremendous amount of heat, and the costs associated with cooling are enormous. I believe there’s a significant need for technologies that automate all of this scheduling. In my opinion, automation of this kind of scheduling is very much needed.

Finally, even if you don’t have a direct answer right now, looking ahead, where do you think AI will contribute the most from an environmental perspective in the era of carbon neutrality?

Since I work in IT, I’d say it’s the IT sector. The reason is that there’s so much potential for reduction in IT. These IT companies are currently all labor-intensive; they operate on a human-based model and rely primarily on infrastructure equipment. For AI to advance at all, IT equipment is what needs to be established as the foundation. IT technology and IT equipment are the starting point. Systems like Gemini or ChatGPT aren’t just created out of thin air because people talk about them. They’re developed by specialists at those companies, who also implement the necessary infrastructure and equipment.

The quality of this equipment, the development process, the sourcing—and the code, code, code… what do you call it? The ability to improve the quality of code—source code, that is—must start with IT. Only then can other processes be automated. That’s why I believe IT is the fundamental starting point. I think that for all technologies—whether business teams are handling these tasks or performing other work—we fundamentally need to build a robust IT foundation. That’s why I believe this is the most essential thing to establish.