[Edaily Reporter Choi Oh-hyun ] Forecasts indicate that memory semiconductor sales will surpass non-memory sales for the first time in history this year. While the competition in artificial intelligence (AI) has traditionally focused on enhancing computational performance through components such as graphics processing units (GPUs), memory—which transmits vast amounts of data quickly and seamlessly—has now emerged as the key factor determining the overall performance of AI systems. The AI boom is shaking up the “power dynamics” of the semiconductor market.

According to market research firm Gartner on the 25th, global memory sales this year are estimated to reach $837.3 billion (approximately 1,160 trillion won), a 280.4% increase compared to last year ($220.1 billion). During the same period, sales of non-memory products, such as central processing units (CPUs), are projected to rise by 21.9%, from $589 billion to $717.9 billion. This marks the first time in the history of the semiconductor industry that memory revenue has surpassed non-memory revenue. While memory revenue accounted for only 27.1% of the total semiconductor market last year, it is projected to exceed half—reaching 53.8%—this year.

Behind this reversal lies the rapid expansion of AI infrastructure. Because AI requires repeatedly retrieving and processing vast amounts of data, simply increasing the computational power of GPUs has its limits in boosting the overall system’s performance. This is because no matter how fast the processing unit is, a bottleneck occurs if the necessary data is not supplied in a timely manner. Consequently, HBM—which rapidly transfers large volumes of data between GPUs and memory—has established itself as a core component of AI accelerators, and the importance of high-capacity DRAM required for AI servers and enterprise SSDs responsible for data storage is also increasing.

Ben Lee, a director and analyst at Gartner, explained, “As AI infrastructure expands, the demand landscape within the semiconductor industry is also being reshaped,” adding, “This demonstrates that structural changes are occurring in the areas where semiconductor value is created and in the patterns of demand across the industry as a whole.” Gartner forecasts that the AI data center ecosystem’s share of total semiconductor revenue will expand from 36.5% this year to over 53% by 2030.

This is why the current memory boom differs from those of the past. In the past, memory demand was driven primarily by cyclical economic factors, such as increases in PC and smartphone sales. In contrast, as investment in AI infrastructure expands, not only is the amount of memory per server increasing, but the proportion of high-value-added products, such as HBM, is also rising simultaneously. A new industry structure has emerged in which faster and greater amounts of memory are required as AI computing performance increases.

The memory market has experienced explosive growth, driven by steep price increases resulting from supply constraints. As demand for HBM, high-capacity server DRAM, and enterprise SSDs has risen simultaneously due to the proliferation of generative AI and Agent AI, memory manufacturers are prioritizing their limited production capacity for high-value-added AI products. In the process, supplies of general-purpose DRAM and NAND have also become tight, causing the upward price trend to spread across the entire memory market.

Lee Kyu-bok, a special professor at Inha University and former president of the Korean Institute of Semiconductor Engineers, commented on the reversal between memory and non-memory segments, stating, “Typically, the semiconductor industry structure consisted of 70% non-memory and 30% memory, but this shift is due to the recent memory shortage, which has caused unit prices to rise significantly.” He added, “Both rising prices and the structural changes driven by AI have had a simultaneous impact.”

According to a report released the previous day by market research firm TrendForce, contract prices for server DRAM are projected to rise by an additional 270% this year, following a cumulative 64% increase in the second half of last year. Prices for enterprise SSDs are also expected to surge by 235% cumulatively this year, following a rise of about 35% in the second half of last year. The report also projected that HBM contract prices could rise by an additional 70–140% by 2027.

As memory emerges as a key resource in the AI era, the standing of South Korean memory manufacturers is also expected to rise. According to Counterpoint Research, SamsungElectronics and SK hynix held a combined market share of 65% in global DRAM revenue in the second quarter of this year. Their dominance is even stronger in the HBM market. Based on first-quarter revenue this year, SK hynix accounted for 58% and SamsungElectronics for 21%, bringing their combined market share to 79%.

This memory-driven growth is highly likely to continue into next year. Gartner forecasts that memory revenue next year will reach $1.0755 trillion, exceeding non-memory revenue by $211.1 billion. Professor Lee stated, “Even if new fabs are built, production capacity does not increase immediately; therefore, the memory shortage and market growth are expected to continue at least through the first half of 2028.”