[Edaily Reporter JAEMIN SONG ] As investment in artificial intelligence (AI) infrastructure by global Big Tech companies surges, rising memory semiconductor prices are emerging as a major factor driving up the cost of building AI servers. While increased sales of high-value-added memory and rising prices are leading to improved earnings for SamsungElectronics and SK hynix, analysts suggest that the growing investment burden on major clients could affect future memory demand.

According to industry sources on the 16th, market research firm TrendForce projected on the 14th that capital expenditures by the world’s nine largest cloud service providers (CSPs) next year will exceed $1.5 trillion (approximately 2,100 trillion won), an increase of about 60% from this year. This year’s capital expenditure forecast also stands at $957 billion, a 106% increase from the previous year.

In contrast, the growth rate for AI server shipments next year is expected to be approximately 36%. This means the growth rate in capital expenditures significantly outpaces that of server shipments. TrendForce analyzed that factors driving the increase in investment costs include not only expanding demand for AI servers but also rising prices for key components such as finished servers and memory, as well as increased advance payments to secure supply.

Similar trends are confirmed by other surveys. According to the U.S. market research firm Dell’Oro Group, global data center capital expenditures in the second quarter of this year increased by 92% compared to the same period last year. The analysis suggests that rising memory and storage prices drove up the average selling price of servers, contributing to the overall increase in investment.

AI servers utilize not only High-Bandwidth Memory (HBM) but also high-capacity server DRAM and enterprise-grade solid-state drives (SSDs) to handle high-performance computing. As AI models grow in scale and data processing volumes increase, demands for higher memory capacity and performance are also rising. Coupled with the upward trend in memory prices, this creates a structure where the cost of building a single server is also increasing.

This trend is having a two-sided impact on domestic memory companies. SamsungElectronics and SK hynix are boosting their profitability thanks to rising memory prices and expanded sales of high-value-added products such as HBM, server DRAM, and enterprise SSDs. On the other hand, server manufacturers and cloud companies are facing rising costs for procuring core components, forcing them to invest more capital to build server infrastructure of the same scale.

The issue is that these rising costs could, in the long term, affect customers’ investment efficiency and purchasing strategies. As the cost per server increases, the number of servers that can be procured with the same budget decreases. Consequently, there is a possibility that big tech companies will take steps to manage costs by adjusting the timing of server deployments and order volumes, or by optimizing component configurations such as memory.

This means that while rising memory prices may boost the earnings of SamsungElectronics and SK hynix in the short term, in the long run, they could increase the purchasing burden on customers and dampen the growth in demand. Hork Tan, CEO of Broadcom, said at an event held in Singapore on the 12th, “Although we are overcoming the shortages of semiconductor wafers and memory, securing funding will be the biggest constraint on the growth of the AI industry going forward.”