(Yicai) Aug. 28 -- Huawei Technologies has brought out a new lineup of high-performance, large-capacity solid-state drives designed to optimize artificial intelligence workloads and bridge the gap between snowballing dataset volumes and available storage capacity.

The new SSD series, made up of the Huawei OceanDisk EX, SP and LC, offers massive storage capacity of up to 245 terabytes per drive, the largest single-disk capacity available in the AI industry, Shenzhen-based Huawei said at a launch event in Shanghai yesterday.

As AI use expands, data libraries are shifting from simple text to multimodal formats like images, audio, and video, meaning that the size of datasets is exploding. AI inference tasks now involve long, multimodal sequences that worsen performance bottlenecks.

Massive data storage shortfalls are becoming acute, Zhou Yuefeng, Huawei vice president and head of its data storage product line, said at the event.

Training a 671-billion-parameter large language model, for instance, requires 3.5 petabytes of data, while the total amount of multimodal data on the global internet has already ballooned to 154 zettabytes, far exceeding what traditional storage media can handle, he said.

The largest version of the popular DeepSeek R1 AI model, for instance, has 671 billion parameters. Training or fine-tuning such a model requires about 13.4 terabytes of memory, which cannot be done on a single machine and in turn greatly limits model training efficiency and flexibility.

South Korea’s Samsung and SK Hynix, the US’ Micron Technology and SanDisk as well as Japan’s Kioxia were the top five makers of enterprise-grade solid-state drives in the first quarter, according to market research firm TrendForce. China has mainly relied on hard disk drives and lags behind on advanced storage technologies despite the rapid growth in demand.  

Although hard-disk drives are still the dominant form of server storage, solid-state drives are gaining ground fast thanks to their energy savings, high efficiency, and low operating costs, TrendForce analyst Ao Guofeng said, adding that by 2028, they should account for 20 percent of all AI server storage.

Editor: Kim Taylor