"To Predict When AI Semiconductors Will Slow Down, 'Look Here': Korea Ratings Identifies New Risk Channel"

Increasing Reliance on External Funding for AI Facility Investment
'Financial Market Signals' May Precede Order Reductions
Some Impact Anticipated on the Power Equipment Industry
"Potential Limitation on Long-Term Upside"

There has been an opinion from the credit rating industry that, when measuring demand for memory semiconductors, not only industrial indicators such as exports but also financial market conditions should be taken into account. This is because as investment in artificial intelligence (AI) infrastructure becomes increasingly inseparable from financial markets, such as external funding, new risk channels are emerging in the memory semiconductor industry.

Kim Jihun, Senior Analyst at Korea Ratings, stated at the seminar titled "AI-High Interest Rates-Real Estate-Changing Market Environment and Credit Risk: Reexamining Crisis and Opportunity," held at the Korea Exchange Conference Hall on the 10th, "Signals of weakening memory demand may appear first in the financial markets rather than in the real market."

Kim Jihun, Senior Analyst at Korea Ratings, is answering questions at the seminar titled "AI-High Interest Rates-Real Estate-Changing Market Environment and Credit Risk: Reexamining Crisis and Opportunity" held on the 10th at the Korea Exchange. Photo by Kim Youngwon

Kim explained, "Recently, the dependence on external funding for AI infrastructure investment has been rising. Companies with high reliance on external funding begin construction on data centers or place orders for CPUs only after securing such funding. Thus, before actual orders begin to decline, there may be signs such as a slowdown in inter-credit transactions or structured financing issuance, or actions by financial investors such as strengthening covenant clauses."

He added, "Recently, the proportion of bond issuance by AI and big tech companies in the U.S. has risen sharply. From a portfolio perspective, as institutional investors reduce their capacity for exposure due to credit limits on certain borrowers or sectors, this could result in decreased issuance. Conversely, if issuance increases, there is also the possibility that existing required interest rates will rise."

These signals, he noted, could ultimately lead to an actual slowdown in memory demand. Kim said, "When such signals appear in the financial markets, required yields increase, and from the perspective of AI players, there is a possibility of adjusting operations, such as shortening data center contract periods or lowering rental prices. If these trends continue, we could see a slowdown in data center construction or reduced orders for memory and GPUs."

He also pointed out that simply increasing the number of long-term supply contracts among memory companies does not mean the memory cycle has been eliminated. Kim noted, "Long-term supply contracts do improve short-term visibility for volume and pricing, but they do not eliminate all risks. In an environment of high dependency on external funding, if customer credit ratings deteriorate, or if it becomes difficult to secure funding due to capital market tightening, there is a real possibility that the terms of long-term supply contracts could be renegotiated."

He concluded, "Ultimately, long-term supply contracts serve more to ease short-term volatility rather than to eliminate cycles altogether. They change the pathway and timing of risk occurrence. Since we do not believe the cycle has been eliminated, financial buffer capacity remains important when assessing the creditworthiness of memory companies."

Impact of AI Investment Structure on the Electrical Equipment Industry?

The financialization of AI investment structure also affects the electrical equipment industry, but the impact is reflected more indirectly compared to memory semiconductors, according to further analysis.

Chae Sunyoung, Senior Analyst at Korea Investors Service, said at the seminar, "As the AI investment structure has become more financialized, there are concerns that a lack of smooth funding could slow the pace of AI investment. For memory semiconductors, demand is likely to be directly reflected in products, but in the case of electrical equipment, demand is reflected indirectly through power and power infrastructure, so the speed is relatively slower."

She also emphasized that while AI data centers are a key factor in power demand, they are not the only factor. Chae said, "In advanced economies, power demand was flat for about 15 years, but even in these developed countries, demand is now expected to increase. This means that structural changes are taking place in the power industry. In the U.S., data centers account for about half of the increase in power demand, but electrification in transportation, such as electric vehicles, also accounts for a significant portion."

Chae also noted that the share of AI data centers in the order backlogs of Korean electrical equipment companies is still not significant. "Among the KRW 16.5 trillion in new orders received by the three major electrical equipment companies in 2025, only about KRW 1 trillion is related to AI data centers," she said. "Therefore, even if the pace of AI investment slows, the short-term impact on credit ratings should be limited." However, if the slowdown continues over the long term, she assessed that it "could serve as a factor limiting the upside expectation for profitability over the medium to long term."

This content was produced with the assistance of AI translation services.