AI Server SSD Market Rising Demand, Supply Updates, and Future Outlook
TapTechNews August 12th news, TrendForce Consulting released a report today. Due to the significant increase in AI demand, in the recent two quarters, AI server-related customers have further requested additional orders of Enterprise SSD (enterprise solid-state drive) from suppliers. In order to meet the supply of SSD in AI applications, upstream suppliers have accelerated the process upgrade and started to plan to launch 2YY products, which are expected to enter mass production in 2025.
The report shows that the situation of AI server customers adding orders to suppliers has led to a cumulative increase of more than 80% in the contract price of Enterprise SSD from the fourth quarter of 2023 to the third quarter of 2024. SSD plays an important role in the development of AI. In addition to the process of AI model training, SSD is not only responsible for storing model parameters, including constantly updated weights and biases, another important application is to create checkpoints to regularly save the progress of AI model training, and it can be recovered from a specific point even if the training is interrupted. The above functions rely quite a lot on high-speed transmission and write durability, so customers mainly choose to use 4TB/8TB TLC (triple-level cell flash memory) SSD products to meet the demanding training process requirements.
The report points out that as more generated information is displayed in the form of videos or photos, the corresponding data storage volume also increases accordingly. Therefore, TLC/QLC (quad-level cell flash memory) 16TB and above and other large-capacity SSDs have become the main products adopted for AI inference.
In 2024, in addition to the significant increase in the demand for products above 16TB in the AI server SSD market, as the arrival of products such as NVIDIA H100, H20 and H200, relevant customers have further increased the order demand for TLC 4 and 8TB Enterprise SSD. According to the estimation of TrendForce Consulting, the purchasing capacity of AI-related SSD this year will exceed 45 EB (TapTechNews note: 1EB = 1024PB ≈ 1.048 million TB), and in the next few years, AI servers are expected to drive the annual growth rate of SSD demand to exceed 60% on average, and the proportion of AI SSD demand in the entire NAND Flash (flash memory) has the opportunity to increase from 5% in 2024 to 9% in 2025.
In the supply market, due to the unchanged trend of using large-capacity SSD products in inference servers, suppliers have begun to accelerate the process upgrade and plan to mass produce products with 2YY/3XX-layer processes starting from the first quarter of 2025, and launch 120TB Enterprise SSD products.