AI Efficiency Gains Reshape Memory Chip Market Expectations

Shares of Samsung Electronics and SK Hynix fell by more than three percent in Seoul on Friday, marking a significant dip in the market for memory components. This drop occurred despite Micron Technology and SanDisk holding steady in positive territory during overnight U.S. trading. The sudden shift in sentiment followed an announcement from Chinese AI startup DeepSeek, which stated that its latest model requires substantially less high-bandwidth memory than previous versions.

The core of the issue lies in the efficiency of the new architecture. By reducing the need for specific types of memory chips, the development challenges the assumption that artificial intelligence growth will automatically drive a continuous surge in hardware demand. Investors are now weighing the long-term outlook for memory consumption, questioning whether the current pace of expansion can be sustained if models become more resource-efficient.

Efficiency Reduces Hardware Demand

DeepSeek unveiled its V4.1 Flash model on Thursday, highlighting a technical shift that activates only a small portion of its massive parameter set at any given time. According to reports from GN technics/ai (en-US), this approach lowers the amount of high-bandwidth memory and solid-state storage needed to run the model. The company plans to retire its more resource-intensive V4-Pro model in September, automatically routing inference tasks to the cheaper, more efficient Flash version.

This move adds a new risk to the memory industry. High-bandwidth memory is currently in high demand, with costs rising sharply due to its critical role in powering high-performance GPUs. If AI models continue to become more memory-efficient, the amount of memory required for each unit of AI computing could decline over time. This potential reduction weakens a key demand driver for the sector, even as overall AI usage expands.

Market Sentiment Remains Divided

The announcement comes at a sensitive time for South Korean memory stocks, which have struggled to recover from a steep selloff in July. Both Samsung and SK Hynix remain more than 25 percent below their record highs. Retail investors have sold over ten billion dollars worth of these two stocks this month, reflecting a cautious approach to the sector's future earnings potential.

Opinions among traders are split on the implications. Some view the recent price drops as a buying opportunity, arguing that major tech companies still need vast amounts of memory for their infrastructure. Others point to the potential for lower demand in AI flash storage as a sign of a more challenging market ahead. This divergence highlights the uncertainty surrounding how efficiency gains in software will translate into hardware sales.

Competitors Face Similar Pressure

The impact was not limited to South Korean firms. Shares of MiniMax Group and Z.AI plunged by more than eight percent in Hong Kong on Friday, indicating a broader regional reaction to the news. While U.S. peers like Micron and SanDisk held their ground initially, the overall sentiment in the sector has shifted toward caution. The market is now closely monitoring how other AI developers respond to these efficiency benchmarks.

The situation underscores a fundamental trade-off in the AI industry. While more efficient models can lower operational costs for providers, they also reduce the volume of hardware required to support those models. For memory manufacturers, this creates a complex challenge where technological progress in one area can directly impact revenue streams in another, forcing a reassessment of long-term growth strategies.