Projections indicate that South Korea will maintain its status as a memory powerhouse through 2031.

Josephine Liu, an analyst at French market research firm Yole Group, shared this outlook on the 4th (local time) at the memory event "FMS 2026" held at the Santa Clara Convention Center in California, USA.

She cited the South Korean government's plan to double wafer production capacity within five years, alongside aggressive investments by Samsung Electronics and SK Hynix, as the basis for this projection.

Meanwhile, the landscape of memory consumption is undergoing significant shifts driven by U.S. semiconductor regulations against China.

The U.S. share of memory consumption surged by 15 percentage points from 37% in 2021 to 52% last year, while China's share, hit directly by the regulations, dropped by 6 percentage points from 35% to 29% over the same period.

The combined memory consumption share of the United States and China increased by 9 percentage points from 72% to 81%.

Analyst Liu diagnosed that the memory semiconductor market has moved past the training-focused "AI 1.0" stage and entered the inference-focused "AI 2.0" stage.

She explained that in the AI 2.0 stage, product development cycles have shortened dramatically, placing pressure on memory manufacturers to release new products in 12 months instead of the traditional 18-month development period.

As a result, corporate agility—the ability to respond swiftly to market environments—has emerged as a core competitiveness, according to the analysis.

Projections were also raised that the influence of Nvidia, which has virtually monopolized the high-bandwidth memory (HBM) market used in AI chips, will relatively decline.

Avril Wu, Senior Vice President at Taiwanese market research firm TrendForce, forecasted that "Nvidia's HBM market share will drop from 66% last year to around 58% this year."

This is because major cloud service providers, including Google with its proprietary Tensor Processing Unit (TPU), are aggressively increasing the adoption of application-specific integrated circuits (ASICs) instead of Nvidia chips.

The technological catch-up by Chinese memory manufacturers was also mentioned.

Yole Group explained that an analysis of DRAM products from ChangXin Memory Technologies (CXMT) showed they have reached the 1z process (third-generation 10-nanometer class) that industry leaders completed back in 2019.

(Photo: Yonhap News)