The North American artificial intelligence (AI) market is expanding its demand for enterprise solid-state drives (eSSD) as applications evolve from AI model training and inference to a broader range of uses. The semiconductor supercycle is expected to significantly increase the share of storage devices.
According to TrendForce, a Taiwanese market research firm, major U.S. cloud service providers (CSPs) have recently raised their eSSD demand forecasts over the past two weeks.
A key feature of this demand increase is the emergence of QLC (Quad-Level Cell) products, which offer cost efficiency and high capacity, alongside the high-performance TLC products that previously drove market growth.
This shift indicates that cloud companies are not only using SSDs for high-performance tasks like AI training and inference but are also expanding their applications to a wider array of AI apps.
As a result, total eSSD orders in the fourth quarter of this year are expected to surpass the record levels set in the third quarter. TrendForce noted that the increase in demand is also impacting supply negotiations between NAND manufacturers and cloud companies. Major NAND players, including Samsung Electronics, SK Hynix, Micron, Kioxia, and YMTC, are strengthening their control over capacity allocation and pricing in the enterprise market, which is likely to lead to further price increases for eSSDs.
In the second quarter of this year, the eSSD market was led by Samsung Electronics, which generated $14.354 billion in revenue (35.1% market share), followed by SK Hynix and Solidigm with $8.631 billion (21.1%), and Micron with $6.983 billion (17.1%).
Experts suggest that while the AI market initially saw bulk purchases of eSSDs for large language model (LLM) training, demand in the second half of the year will be driven by agent AI in North America and KV cache offloading in China.
KV cache offloading refers to a technique that moves relatively lower-priority data to system memory or storage to alleviate the excessive use of high-performance memory (HBM, GDDR) during LLM inference. While U.S. AI models focus on maximizing performance through high-performance memory, refined AI models in China are shifting their research and development to achieve optimal results with reduced high-performance memory usage.
TrendForce analyzed that as agent AI becomes more prevalent, the demand for large-scale real-time data retrieval and caching is rapidly increasing. For agent AI to make autonomous decisions and actions, it must quickly access and store vast amounts of context and data, which will enhance the role of high-performance eSSDs.
In this context, cloud companies are expected to actively utilize QLC to achieve both high capacity and cost efficiency. QLC is anticipated to be widely used in building vector databases, which are core components of AI search and recommendation systems. This shift may lead to a reconfiguration of AI infrastructure design centered around QLC, moving beyond changes in purchasing patterns of a few hyperscalers.
Another significant factor driving eSSD demand is China. Chinese companies, particularly DeepSeek, are actively promoting the use of high-capacity QLC enterprise SSDs for KV cache offloading. This approach aims to enhance model inference efficiency and reduce costs by selectively moving cached data from system memory to high-speed, high-capacity QLC.
Given the challenges faced by Chinese AI companies in securing high-performance memory due to U.S. semiconductor export regulations, such system optimization is expected to become essential, further accelerating the adoption of QLC in China. In this regard, YMTC, which was previously considered less competitive against companies like Samsung and SK Hynix, is expected to benefit from an increase in market share within China.
A semiconductor industry insider stated, "The eSSD market is expected to see independent demand growth from U.S. and Chinese companies, while price increases driven by NAND manufacturers' price control are likely to continue into the fourth quarter."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.