韩国存储芯片巨头SK海力士近日对未来数年的存储市场前景发出预警。公司表示，尽管行业正在积极扩大产能，但在人工智能快速发展的推动下，高性能存储器需求增长速度仍将超过供给增速，全球存储芯片短缺现象可能持续至2030年前后。
随着生成式人工智能、大语言模型以及AI数据中心建设持续升温，HBM（高带宽内存）已成为当前半导体产业链最紧缺的核心组件之一。作为全球HBM市场的主要供应商，SK海力士认为，未来几年市场需求将继续保持强劲增长。
SK海力士指出，尽管公司正在推进多项扩产计划，包括新建生产设施和扩大先进封装能力，但存储行业扩产并非短期即可完成。先进存储产品尤其是HBM不仅需要更先进的制造工艺，还受到封装设备、产线建设周期以及人才储备等多方面限制，因此新增产能释放速度较为缓慢。
公司预计，到2030年之前，AI基础设施建设仍将保持高速发展态势。大型科技企业、云计算服务商以及主权AI项目都在持续增加对高性能计算平台的投资，而这些平台对HBM和高端DRAM的需求远高于传统服务器。
分析认为，随着AI模型规模不断扩大，每颗GPU搭载的HBM容量也在持续提升。从HBM3、HBM3E到未来的HBM4和后续产品，单个AI加速器对于存储资源的需求正呈现快速增长趋势，这进一步加剧了供需压力。
SK海力士同时表示，公司正在通过扩建先进封装产线、新建芯片工厂以及提升生产效率等方式增加供应能力。然而，即使行业主要厂商都在扩大投资，在需求持续爆发的背景下，供应追赶需求仍将面临巨大挑战。
当前，SK海力士在HBM市场保持领先地位，并已成为英伟达等AI芯片厂商的重要合作伙伴。随着各大科技公司竞相建设AI数据中心，市场对于先进存储器的争夺预计将在未来几年持续升温。
业内普遍认为，如果AI产业维持当前增长轨迹，HBM、先进DRAM以及相关封装产能将继续成为全球半导体供应链中的关键瓶颈。SK海力士此次预测也反映出产业界对于未来数年高性能存储需求持续增长的预期，即使扩产规模不断扩大，供需关系仍可能长期维持偏紧状态。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Korea ' s storage chip giant, SK Hercules, has recently given early warning of future storage market prospects for years to come. According to the company, although the industry is actively expanding its production capacity, the rapid development of artificial intelligence has driven the growth of demand for high-performance storage units to faster than supply growth, and the global stock chip shortage is likely to continue around 2030.
HBM (high bandwidth memory) has become one of the most critical core components of the current semiconductor industrial chain as the generation of artificial intelligence, large language models and the construction of the AID data centre continues to warm. As the main supplier of the global HBM market, SK Hercules believes that market demand will continue to grow strongly in the coming years.
According to SK Hercules, although the company is pursuing a number of expansion schemes, including the construction of new production facilities and the expansion of advanced containment capacity, the expansion of the storage industry is not short-term. Advanced storage products, particularly HBMs, not only require more advanced manufacturing processes, but are also constrained by a number of constraints, including sealing equipment, production-line construction cycles and talent reserves, which slows the release of new capacity.
The company expects that the development of AI infrastructure will remain high until 2030. Large technology enterprises, cloud computing service providers and the Sovereign AI Project have continued to increase investment in high performance computing platforms, which require much more HBM and high-end DRAM than traditional servers.
The analysis concluded that the HBM capacity per GPU continued to increase as the size of the AI model grew. From HBM3, HBM3E to future HBM4 and subsequent products, demand for stored resources by individual AI accelerators is increasing rapidly, further exacerbating demand and supply pressures.
At the same time, SK Hercules indicated that the company was increasing its supply capacity through the expansion of advanced containment lines, the construction of new chip plants and the improvement of production efficiency. However, even when major producers in the industry are expanding their investments, supply catch-up will continue to face enormous challenges in the context of a continuing surge in demand.
Currently, SK Hercules has taken the lead in HBM markets and has become an important partner for AI chip manufacturers such as In Weida. As major technology companies compete to build an AI data centre, market competition for advanced storage units is expected to increase in the coming years.
It is widely recognized within industry that if the AI industry maintains its current growth trajectory, HBMs, advanced DRAMs and the associated containment capacity will continue to be key bottlenecks in the global semiconductor supply chain. This projection by SK Hercules also reflects industry ' s expectation of a sustained growth in high-performance storage demand in the coming years, with the possibility of long-term supply-demand constraints even as the scale of the expansion expands.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.