美国美光将HBM产能翻倍，缩小与三星和SK海力士的差距
美国美光科技计划将其高带宽内存（HBM）产能较去年翻倍。该公司正进行积极的资本投资，以缩小与三星和SK海力士的差距。
原文深度解读
本文报道美国美光科技计划将高带宽内存（HBM）产能较去年翻倍，以缩小与三星电子和SK海力士的差距。文章涉及美光、三星、SK海力士、英伟达等公司，时间点为2025年9月3日及今年内。关键数字包括：美光年底月产能约10万片晶圆，去年月产能4万至5万片；12层HBM4占比从年初20%-30%升至年底50%；三星和SK海力士月产能各约15万至20万片；美光HBM4累计出货收入超10亿美元。文章与电子元器件供应链相关，聚焦HBM产能扩张、设备采购及AI内存需求。
- 美光计划到今年年底将HBM月产能增加至多6万片晶圆（按晶圆投入计算），去年月产能约4万至5万片，净增超过去年水平。
- 行业消息称，美光年底总HBM月产能约10万片晶圆，正通过增加对多家HBM制造设备供应商的采购订单来保障产能。
- 美光HBM生产中心位于台湾和新加坡的晶圆厂，设备持续交付中；12层HBM4占比预计从年初的20%-30%升至年底的50%。
- 12层HBM4将用于英伟达最新AI加速器'Vera Rubin'，美光已于今年第二季度开始量产12层HBM4。
- 美光CEO Sanjay Mehrotra在6月财报电话会上表示，12层HBM4的量产爬坡速度约为12层HBM3E的两倍，HBM4累计出货收入已超10亿美元。
- 行业估计三星和SK海力士的HBM月产能各约15万至20万片晶圆，是美光的3至4倍；若美光达成目标，差距可能缩小至原来的一半。
供应链影响
- 美光扩大HBM产能可能增加对HBM制造设备的需求，设备供应商订单量可能上升，具体取决于美光设备采购的落实速度。
- 美光在台湾、新加坡及日本广岛的产能投资可能影响当地半导体供应链布局，包括封装和测试环节，具体取决于投资进度。
- HBM产能增加可能缓解AI内存供应紧张，但整体供需平衡仍取决于三星和SK海力士的扩产计划及AI芯片需求变化。
- 12层HBM4占比提升可能影响相关材料、设备及测试服务的需求结构，具体取决于良率和量产爬坡情况。
- 美光与三星、SK海力士的产能差距缩小可能改变HBM市场竞争格局，进而影响客户采购策略，但具体影响取决于市场动态。
系统解读，仅作为判断线索，不构成备货、出货、涨价或投资建议。
美光HBM4
美国美光科技计划将其高带宽内存（HBM）产能较去年翻倍。该公司正进行积极的资本投资，直至今年年底，以增强其最新一代HBM——12层HBM4的供应能力。此举旨在通过提升目前落后于三星电子和SK海力士的HBM产能来增加市场份额。
据行业消息人士9月3日透露，美光计划到今年年底将HBM月产能（按晶圆投入计算）增加至多6万片晶圆。去年，美光的HBM产量约为每月4万至5万片晶圆，这意味着今年的净增量将超过这一水平。
一位熟悉此事的行业消息人士表示：“美光正在进行大规模设施投资，以迅速提升其HBM4产能，”并补充道，“到今年年底，它将确保每月约10万片晶圆的总HBM产能。”
为获得如此产能，美光正在增加对多家HBM制造设备供应商的采购订单（PO）。其HBM生产中心位于台湾和新加坡的晶圆厂，据称设备正在持续交付中。
12层HBM4的比例预计将显著上升。目前，12层HBM3E占美光HBM产量的大部分。尽管截至今年年初，12层HBM4仅占总产量的20%至30%，但据报道，到年底这一数字将达到50%。
美国美光将HBM产能翻倍，缩小与三星和SK的差距
12层HBM4将安装在全球最大人工智能（AI）半导体公司英伟达的最新AI加速器“Vera Rubin”中。美光已于今年第二季度开始量产12层HBM4产品。
早些时候，美光首席执行官Sanjay Mehrotra在6月份的2026财年第三季度财报电话会议上表示，“12层HBM4的量产爬坡速度约为12层HBM3E的两倍，”并透露“HBM4的累计出货收入已超过10亿美元。”
尽管美光是全球三大HBM制造商之一，但其产能仍不及三星电子和SK海力士。行业估计，三星电子和SK海力士的HBM产能是美光的3至4倍（每家公司每月约15万至20万片晶圆）。从美光的角度来看，除非它赶上这两家公司的产能，否则很难摆脱永远第三的地位。
如果美光今年达到其HBM产能目标，与三星电子和SK海力士的产能差距可能会缩小至原来的一半。
美光扩大HBM产能也证明了AI内存的强劲需求。随着以英伟达为首的AI半导体制造商积极寻求HBM，供应仍无法满足需求。在这种环境下，最大化HBM盈利能力需要供应能力，即产能。
美光目前正在全球范围内投资HBM生产设施。在美国以外，HBM封装正在台湾的铜锣晶圆厂和新加坡的伍德兰晶圆厂进行，同时也在为其日本广岛晶圆厂准备投资。
根据Counterpoint Research的数据，今年第二季度，SK海力士在全球HBM市场排名第一，市场份额为50%。三星电子占32%，而美光占18%。
Micron HBM4
Micron Technology of the U.S. will double its High Bandwidth Memory (HBM) production capacity compared to last year. The company is making aggressive capital investments through the end of the year to enhance its supply capability for 12-layer HBM4, the latest generation of HBM. This is an attempt to increase market share by boosting its HBM production capacity, which currently lags behind Samsung Electronics and SK Hynix.
According to industry sources on September 3, Micron plans to add up to 60,000 wafers per month in HBM production capacity (based on wafer input) by the end of this year. Last year, Micron's HBM output stood at around 40,000 to 50,000 wafers per month, meaning this year's net increase will exceed that level.
An industry source familiar with the matter stated, “Micron is making large-scale facility investments to rapidly boost its HBM4 production capacity,” adding, “By the end of this year, it will secure a total HBM production capacity of approximately 100,000 wafers per month.”
To secure such production capacity, Micron is increasing purchase orders (POs) to multiple HBM manufacturing equipment suppliers. Its HBM production hubs are fabs in Taiwan and Singapore, where continuous equipment delivery is reportedly taking place.
The proportion of 12-layer HBM4 is expected to rise significantly. Currently, 12-layer HBM3E accounts for most of Micron's HBM production volume. Although 12-layer HBM4 accounted for only 20% to 30% of total output as of early this year, that figure is reported to reach up to 50% by the end of the year.
US Micron to Double HBM Capacity, Closing Gap with Samsung and SK
The 12-layer HBM4 will be installed in 'Vera Rubin', the latest AI accelerator from Nvidia, the world's largest artificial intelligence (AI) semiconductor company. Micron began mass production of 12-layer HBM4 products in the second quarter of this year.
Earlier, Sanjay Mehrotra, CEO of Micron, stated during the fiscal third-quarter 2026 earnings call in June that “the mass-production ramp-up for 12-layer HBM4 is progressing about twice as fast as that of 12-layer HBM3E,” and revealed that “cumulative shipment revenue for HBM4 has surpassed $1 billion.”
Although Micron is one of the world's top three HBM manufacturers, its production capacity falls short of Samsung Electronics and SK Hynix. The industry estimates that the HBM production capacity of Samsung Electronics and SK Hynix is 3 to 4 times higher than Micron's (about 150,000 to 200,000 wafers per month for each company). From Micron's perspective, unless it catches up with the production capacity of both companies, it will be difficult to break out of its status as a perpetual third player.
If Micron meets its HBM capacity targets this year, the production gap with Samsung Electronics and SK Hynix could be narrowed to half of its previous level.
Micron's expansion of HBM production capacity is also proof of robust demand for AI memory. As AI semiconductor makers led by Nvidia actively seek HBM, supply is still unable to keep pace with demand. In this environment, maximizing HBM profitability requires the ability to supply—namely, production capacity.
Micron is currently investing in HBM production facilities across the globe. Outside the U.S., HBM packaging is taking place at the Tongluo fab in Taiwan and the Woodlands fab in Singapore, while investments are also being prepared for its Hiroshima fab in Japan.
According to Counterpoint Research, SK Hynix ranked first in the global HBM market in the second quarter of this year with a 50% market share. Samsung Electronics held 32%, while Micron accounted for 18%.