📊 TrendForceTrendForceCodex · 实盘不做把握 7/10
北京时间·🕒 发布 07:30:47时区推断·⟳ 抓取 07:30:52·⚡ 决策 07-23 07:32:53 +2m01s·96.4 天前
南亚科技高管称AI推理将推动2026年全球存储市场创新高
这是一条存储行业景气预测与公司技术路线介绍，未披露新增订单、客户定点、量产收入或超预期财务指引。
📄 原文TrendForce↗ 原文
[News] Nanya Tech Sees Memory Market to Reach Record US$800B in 2026, Nearly Half of Semiconductor Revenue As the memor…
不做 南亚科技（2408.TW）
不做：本条是存储行业规模预测和公司技术路线宣传，缺少新增订单、客户定点、量产收入或超预期财务指引，无法形成短周期可量化交易edge。
展开详情(成交明细 / 研究档案 / 原文全文 / 决策记录链)
研究档案
🔎 新鲜度:最早出现 2026-07-22 · 扩散 medium · price-in 风险 medium ↗
关键理由
- 这是一条存储行业景气预测与公司技术路线介绍，未披露新增订单、客户定点、量产收入或超预期财务指引。
- 南亚科技被点名为潜在受益方，但事件来自行业媒体转述高管观点，短周期内难以形成可量化的新增收入或利润传导。
- AI推理带动DRAM需求属于已交易的行业主线，单条市场规模预测不足以支撑可回测的同向交易。
主要风险
- 若后续公司正式披露UWIO DRAM客户、量产时间或资本开支超预期，需作为新催化重新评估。
- 存储上行周期可能继续利好相关个股，但本条本身更偏行业背景而非离散交易事件。
原文全文
展开原文 · 3,117 字
[News] Nanya Tech Sees Memory Market to Reach Record US$800B in 2026, Nearly Half of Semiconductor Revenue
As the memory upcycle broadens from HBM and leading-edge DRAM to legacy products, Taiwanese DRAM maker Nanya Technology is emerging as a key beneficiary as AI-driven demand expands across the market. Citing Nanya Tech Senior Vice President Joseph Wu, Liberty Times reports that AI inference is expected to propel the global memory market value, including DRAM and NAND, to a record US$800 billion in 2026.
According to Wu, memory’s share of total semiconductor industry revenue could reach nearly 50% this year, significantly higher than the roughly 30% level seen historically.
According to Liberty Times, Wu said that as AI shifts from training to inference, demand for 3D-stacked memory — such as HBM — is expected to grow further over the next one to two years, with HBM projected to account for nearly 10% of total DRAM bit demand this year, while conventional server DRAM is expected to represent more than 40%.
Together, HBM and server DRAM already account for more than 50% of DRAM demand, and their combined share is expected to rise above 60% in the future, Wu noted.
Wu, per Next Apple , further highlighted the growing memory intensity of AI systems: For AI servers, a single GPU may require 96 high-density DRAM chips, while a single CPU platform could require more than 200 high-density DRAM chips as AI inference gains traction.
Nanya Tech Bets on Customized UWIO DRAM
However, Nanya Tech is not targeting the traditional HBM market. While HBM remains essential for AI training, Wu noted that the rise of AI inference is shifting the bottleneck toward memory bandwidth and power efficiency, as edge devices handle massive user interactions and KV cache workloads, Mirror Daily reports.
Due to its limited I/O capacity and higher power consumption, HBM is not an ideal fit for power- and latency-sensitive devices such as AI PCs, smartphones, and portable devices, Wu added.
To address this, Nanya is developing its customized UWIO DRAM, which leverages Wafer-on-Wafer (WoW) 3D IC stacking technology to directly integrate DRAM with CPUs or ASICs, Mirror Daily adds. The architecture reportedly delivers 5–10x higher bandwidth while reducing energy consumption per bit transferred to one-third to one-tenth that of HBM.
On the other hand, the company is accelerating its capacity expansion with a new wafer fab in Taishan, northern Taiwan, where total investment is expected to reach NT$500 billion (US$17 billion), according to Mirror Daily. The facility will reportedly introduce EUV lithography, with next year’s capital expenditure expected to surpass NT$200 billion. Set to start production ramp-up early next year, the new fab is expected to help Nanya narrow the technology gap with leading global memory makers, the report adds.
Read more
[News] DRAM Rally Lifts Nanya Tech’s 2Q Gross Margin to Nearly 80%, Closing in on Global Memory Leaders
(Photo credit: Nanya Technology)
Please note that this article cites information from Liberty Times , Next Apple and Mirror Daily .
As the memory upcycle broadens from HBM and leading-edge DRAM to legacy products, Taiwanese DRAM maker Nanya Technology is emerging as a key beneficiary as AI-driven demand expands across the market. Citing Nanya Tech Senior Vice President Joseph Wu, Liberty Times reports that AI inference is expected to propel the global memory market value, including DRAM and NAND, to a record US$800 billion in 2026.
According to Wu, memory’s share of total semiconductor industry revenue could reach nearly 50% this year, significantly higher than the roughly 30% level seen historically.
According to Liberty Times, Wu said that as AI shifts from training to inference, demand for 3D-stacked memory — such as HBM — is expected to grow further over the next one to two years, with HBM projected to account for nearly 10% of total DRAM bit demand this year, while conventional server DRAM is expected to represent more than 40%.
Together, HBM and server DRAM already account for more than 50% of DRAM demand, and their combined share is expected to rise above 60% in the future, Wu noted.
Wu, per Next Apple , further highlighted the growing memory intensity of AI systems: For AI servers, a single GPU may require 96 high-density DRAM chips, while a single CPU platform could require more than 200 high-density DRAM chips as AI inference gains traction.
Nanya Tech Bets on Customized UWIO DRAM
However, Nanya Tech is not targeting the traditional HBM market. While HBM remains essential for AI training, Wu noted that the rise of AI inference is shifting the bottleneck toward memory bandwidth and power efficiency, as edge devices handle massive user interactions and KV cache workloads, Mirror Daily reports.
Due to its limited I/O capacity and higher power consumption, HBM is not an ideal fit for power- and latency-sensitive devices such as AI PCs, smartphones, and portable devices, Wu added.
To address this, Nanya is developing its customized UWIO DRAM, which leverages Wafer-on-Wafer (WoW) 3D IC stacking technology to directly integrate DRAM with CPUs or ASICs, Mirror Daily adds. The architecture reportedly delivers 5–10x higher bandwidth while reducing energy consumption per bit transferred to one-third to one-tenth that of HBM.
On the other hand, the company is accelerating its capacity expansion with a new wafer fab in Taishan, northern Taiwan, where total investment is expected to reach NT$500 billion (US$17 billion), according to Mirror Daily. The facility will reportedly introduce EUV lithography, with next year’s capital expenditure expected to surpass NT$200 billion. Set to start production ramp-up early next year, the new fab is expected to help Nanya narrow the technology gap with leading global memory makers, the report adds.
Read more
[News] DRAM Rally Lifts Nanya Tech’s 2Q Gross Margin to Nearly 80%, Closing in on Global Memory Leaders
(Photo credit: Nanya Technology)
Please note that this article cites information from Liberty Times , Next Apple and Mirror Daily .
决策记录链 · 只追加
每次领取产生一条记录,重新决策追加一条并以 supersedes 指向被替代的记录(ADR 0014)。
这条决策早于决策账本(2026-10-08 起逐条追加),没有追加记录。