DRAM、NAND、HBM及利基存储涨价证据强化，亚洲存储股已完成首轮反应

# 美股盘后催化 4/11: DRAM、NAND、HBM和利基存储涨价多源确认，存储链成为独立于GPU订单的价格催化

- 美股会话: 2026-07-20
- 方向: POSITIVE；类型: 供需/价格；规划重大性: 9/10
- 映射模式: sector_sympathy
- 美股驱动（规划器初判，需独立核验）: 事实：KeyBanc预计3Q DRAM价格环比涨15%-20%、NAND涨30%-40%，2027年HBM年度议价可能大幅上行；北京君正称三季度DRAM继续涨价、毛利率提升；东芯股份预告上半年扭亏并称SLC NAND等利基存储量价齐升，TrendForce口径显示NOR和SLC NAND上半年大涨。推断：这是存储价格周期和AI内存瓶颈的独立催化，利好HBM/DRAM/NAND/利基存储供应商，但可能提高下游AI硬件和工控汽车电子成本。

## 本催化相关美股真实收盘行情

- 英伟达 (NVDA.US): 203.28 +0.23%
- AMD (AMD.US): 503.57 +1.58%
- 美光 (MU.US): 865.46 +1.94%
- WDC.US (WDC.US): 487.42 +2.14%
- STX.US (STX.US): 802.45 +1.88%
- SNDK.US (SNDK.US): 1390.95 +2.67%

## 亚太映射假设（仅候选，必须查证）

- SK海力士 (000660.KS) [KR/同源本地股/positive]: HBM和DRAM涨价直接改善盈利和估值假设。；核验问题: SK海力士2027年HBM合约价、客户锁量、HBM4量产节奏和DRAM价格弹性。
- 三星电子 (005930.KS) [KR/竞争替代/positive]: DRAM/NAND全行业涨价改善存储周期，HBM认证若推进将提高利润弹性。；核验问题: 三星HBM客户认证进度、NAND价格弹性和存储利润敏感度。
- 东芯股份 (688110.SH) [CN/同源本地股/positive]: 公司直接披露SLC NAND、NOR、利基DRAM量价齐升并带动中报扭亏。；核验问题: 涨价持续性、客户价格传导、产品结构和砺算科技亏损影响。

## 必答问题

- 涨价主要是合约价还是现货价？
- HBM提价会否挤压GPU和云客户采购量？
- 利基存储涨价是短期供需错配还是多年供给收缩？

## 原始证据全文（17 条）

### [E0351] Eastmoney Finance | 2026-07-20 14:19 UTC | 盘中
北京君正：第三季度DRAM会持续涨价 预计毛利率将继续提升
URL: http://finance.eastmoney.com/a/202607203813700744.html
北京君正在互动平台表示，公司第二季度已经有海外客户开始涨价，第三季度DRAM会持续涨价，预计第三季度毛利率会继续提升。

### [E0352] 财联社电报 | 2026-07-20 14:19 UTC | 盘中
北京君正：第三季度DRAM会持续涨价 预计毛利率将继续提升
URL: https://www.cls.cn/detail/2431808
关于我们
网站声明
联系方式
用户反馈
网站地图
帮助
首页
电报
话题
盯盘
VIP
FM
投研
下载
全部
加红
公司
看盘
港美股
基金
提醒
2026年07月20日 22:19:55
北京君正：第三季度DRAM会持续涨价 预计毛利率将继续提升
财联社7月20日电，北京君正在互动平台表示，公司第二季度已经有海外客户开始涨价，第三季度DRAM会持续涨价，预计第三季度毛利率会继续提升。
关联个股
北京君正
-6.61%
免责声明：本文内容与数据仅供参考，不构成投资建议，更不代表财联社观点，使用前请核实。稿件基于人工智能辅助处理，编辑人工审核亦不能保证绝对无差错。如有投资者据此操作，风险自担。
阅6380
关联话题
互动平台精选
3.31W 人关注
关于我们
|
网站声明
|
联系方式
|
用户反馈
|
网站地图
|
友情链接
|
举报电话：021-54679377转617
举报邮箱：editor@cls.cn
财联社
©2018-2026
上海界面财联社科技股份有限公司 版权所有
沪ICP备14040942号-9
沪公网安备31010402006047号
互联网新闻信息服务许可证：31120170007
沪金信备 [2021] 2号

### [E0374] 每日经济新闻快讯 | 2026-07-20 14:21 UTC | 盘中
北京君正：第三季度DRAM会持续涨价 预计毛利率将继续提升
URL: http://www.nbd.com.cn/articles/2026-07-20/4483990.html
每经网首页 > 首发快讯 > 正文
每日经济新闻 2026-07-20 22:21:56
每经AI快讯，7月20日，北京君正在互动平台表示，公司第二季度已经有海外客户开始涨价，第三季度DRAM会持续涨价，预计第三季度毛利率会继续提升。
如需转载请与《每日经济新闻》报社联系。未经《每日经济新闻》报社授权，严禁转载或镜像，违者必究。
读者热线：4008890008
特别提醒：如果我们使用了您的图片，请作者与本站联系索取稿酬。如您不希望作品出现在本站，可联系我们要求撤下您的作品。
上一篇文章
长盈精密：与宇树科技不存在股权关系
下一篇文章
北京君正：第三季度DRAM会持续涨价 预计毛利率将继续提升
欢迎关注每日经济新闻APP
0

### [E0398] Tae Kim | 2026-07-20 14:27 UTC | 盘中
KeyBanc: “Expect 3Q26 DRAM pricing to increase +15-20% q/q, while NAND is expected +30-40% q/q. Supply chain partners continue to expect that DRAM and NAND pricing will continue to see significant increases over the next few quarters, yet a
URL: https://x.com/firstadopter/status/2079211557930586538
KeyBanc: “Expect 3Q26 DRAM pricing to increase +15-20% q/q, while NAND is expected +30-40% q/q. Supply chain partners continue to expect that DRAM and NAND pricing will continue to see significant increases over the next few quarters, yet at a lesser rate than what was seen in 1H26. For DRAM, after increasing +45-55% q/q in 2Q26, we expect 3Q26 pricing of +15-20% q/q and 4Q26 pricing of +15% q/q. For NAND, pricing is expected to increase at a greater rate, as the price increases have lagged DRAM; after increasing +75% q/q in 2Q26, NAND pricing is expected to increase +30-40% q/q in 3Q26 and +15% q/q in 4Q26.”

“HBM pricing is expected to increase +100-150% in 1Q27, as new pricing kicks in. Given pricing for HBM is negotiated on an annual basis, we hear expectations that HBM pricing has been set for 2027 and is expected to increase +100% y/y for HBM4 and +150% y/y for HBM3E, when new negotiated pricing for HBM kicks in. Currently, HBM pricing is dilutive to commodity DRAM GMs, as it was negotiated in 2025, but after the price reset, HBM GMs should be neutral or accretive to commodity DRAM GMs.”

### [E0719] Eastmoney Finance | 2026-07-20 15:41 UTC | 盘中
东芯股份营收增逾三倍 同比扭亏为盈 近期股价却已腰斩 | 财报解读
URL: http://finance.eastmoney.com/a/202607203813755280.html
财联社7月20日讯（记者沈娇娇张校毓）今日晚间，东芯股份发布业绩预增公告，主要受益于海外存储巨头收缩利基型市场，同时国内下游相关应用需求加速，供需关系呈现紧缺态势，产品价格攀升。另外，东芯股份战略投资的砺算科技（上海）有限公司（以下简称“砺算科技”）旗下消费显卡近期开启商业化。不过，公司股价在最近一个月经历了先扬后抑的剧烈波动，自最高点已经腰斩。

### [E0721] 财联社电报 | 2026-07-20 15:43 UTC | 盘中
东芯股份营收增逾三倍 同比扭亏为盈 近期股价却已腰斩|财报解读
URL: https://www.cls.cn/detail/2431849
①2026年上半年，东芯股份实现营收同比大幅增长334.41%至346.07%、同比扭亏为盈。 ②主要受益于海外存储巨头收缩利基型市场，供需关系呈现紧缺态势，产品价格攀升。 ③东芯股份公司股价在最近一个月经历了先扬后抑的剧烈波动，自最高点已经腰斩。
面对每天上千份上市公司公告该看哪些？重大事项公告动辄几十页几百页重点是啥？公告里一堆专业术语不知道算利好还是利空？请看财联社公司新闻部《速读公告》栏目，我们派驻全国的记者们将于公告当晚为您带来准确、快速、专业的解读。
财联社7月20日讯（记者 沈娇娇 张校毓）今日晚间，东芯股份发布业绩预增公告，主要受益于海外存储巨头收缩利基型市场，同时国内下游相关应用需求加速，供需关系呈现紧缺态势，产品价格攀升。
另外，东芯股份战略投资的砺算科技（上海）有限公司（以下简称“砺算科技”）旗下消费显卡近期开启商业化。
不过，公司股价在最近一个月经历了先扬后抑的剧烈波动，自最高点已经腰斩。
具体来看，2026年上半年，东芯股份实现营业收入14.9亿至15.3亿元，同比增长334.41%至346.07%；扣非归母净利润为6.3亿至6.7亿元，同比扭亏为盈，上年同期为-1.27亿元。
东芯股份指出，报告期内，受益于AI算力需求持续爆发及国内利基型存储技术竞争力的增强，海外存储原厂加速退出利基型市场，供给端收缩态势明显；与此同时，网络通信、工业控制、汽车电子等下游应用需求稳步复苏，利基型存储芯片供需缺口持续扩大。公司所面对的存储芯片市场的供需关系呈现结构性紧缺态势，产品市场价格攀升。
公司以SLC NAND Flash为代表的存储业务实现量价齐升，带动上半年营业收入实现增长，整体毛利率亦显著上涨，带动上半年利润水平的显著提高。从季度维度来看，公司业绩呈现环比持续向好趋势，第二季度营业收入较第一季度增长110.85%至119.20%，归母净利润较第一季度增长262.81%至291.74%，盈利能力持续提升。
东芯股份在7月7日投资者关系活动中指出，目前行业层面普遍预期三季度海外原厂产能调整带来的影响将持续显现，利基存储芯片价格具备上行支撑。而公司是中国大陆少数同时具备 NAND Flash、NOR Flash、DRAM 三大存储芯片自主研发设计能力的企业。
根据Trend Force最新存储器价格调查，由于高层数3D NAND等高附加价值产品排挤成熟制程产能，MLC NAND供给极度短缺，迫使部分工控、车用和网通客户计划改采用SLC NAND，将导致原已吃紧的SLC供应情况更加紧绷。同时，AI边缘运算、数据中心、汽车电子等对SLC的需求持续提升，供需缺口急剧扩大，预估将推升2026年下半年SLC合约价格较上半年调涨120-170%。
至于下游客户对于存储芯片涨价的接受度情况，东芯股份在7月14日投资者关系活动中表示，车规、工控等场景对供应链稳定性、产品可靠性要求较高，对产品价格波动的容忍度相对较高；此外，细分行业龙头的消费类客户对价格的接受程度也在逐步提升。总体来看，公司与下游客户的价格传导整体较为顺畅。
另外，备受投资者关注的砺算科技近期亦有新动态。
东芯股份于2024年通过自有资金2亿元战略投资砺算科技，开始布局高性能GPU领域，并于2025年追加投资约2.11亿元，持有约35.87%股权，成为单一最大股东。
据介绍，砺算科技专注于多层次（可扩展）图形渲染 GPU 芯片的研发设计，在GPU架构设计等关键技术环节拥有完整的自主知识产权，产品覆盖消费卡到专业卡。其中，专业卡已在2025年开始进行交付，消费级显卡2026年亦已公开发售。
公司2025年年报显示，砺算科技首款自研GPU芯片“7G100”首次流片成功，少量显卡已交付客户，产品量产及销售拓展等工作正在正常开展中。由于上述业务处于初期阶段、未开启大规模商业化，2025年度，砺算科技确认营收仅仅为2.48万元，东芯股份对于该投资以权益法核算确认的投资亏损约1.66亿元。
财联社记者查看砺算科技的京东自营旗舰店，Lisuan eXtreme系列创始版消费显卡在今年618期间首发价格为3299元，销量超过1000+，目前处于售罄无货的状态；另外一款显卡价格为2688元，自7月17日开始预约，预约人数已超过1000+，销量超100+。上述两款消费显卡产品在京东自营渠道形成营收超过300万元；而目前专业显卡暂未披露销量情况。
值得一提的是，东芯股份公司股价在最近一个月经历了先扬后抑的剧烈波动。公司6月30日发布的《关于股票交易风险提示》指出，股价已经连续3个交易日涨幅偏离值达6.66%，连续10个交易日累计涨幅偏离值达24.74%。随后，公司股价开始大幅下跌，从6月30日最高点209.55元已经腰斩至100.38元（今日收盘价）。

### [E0816] Eastmoney Finance | 2026-07-20 16:00 UTC | 盘中
东芯股份Q2净利环比预增262%-291% 以SLC NAND Flash为代表的存储业务量价齐升
URL: http://finance.eastmoney.com/a/202607213813764032.html
《科创板日报》7月20日讯（记者陈俊清）今日（7月20日）晚间，东芯股份公布2026年半年度业绩预告的自愿性披露公告。

### [E0817] 科创板日报 | 2026-07-20 16:00 UTC | 盘中
东芯股份Q2净利环比预增262%-291% 以SLC NAND Flash为代表的存储业务量价齐升
URL: https://www.chinastarmarket.cn/detail/2431857
①东芯股份预计2026年半年度实现营业收入14.9亿元至15.3亿元，同比增长334.41%至346.07%；归母净利润预计为6.4亿元至6.8亿元，较上年同期扭亏为盈； ②东芯股份以SLC NAND Flash为代表的存储业务实现量价齐升，带动上半年营业收入实现大幅增长。
《科创板日报》7月20日讯（记者 陈俊清） 今日（7月20日）晚间，东芯股份公布2026年半年度业绩预告的自愿性披露公告。
公告显示，东芯股份预计2026年半年度实现营业收入14.9亿元至15.3亿元，同比增长334.41%至346.07%；归母净利润预计为6.4亿元至6.8亿元，较上年同期扭亏为盈，增加7.51亿元至7.91亿元；扣非净利润预计为6.3亿元至6.7亿元，同比扭亏为盈，增加7.57亿元至7.97亿元。
从季度维度来看，该公司业绩呈现环比持续向好趋势，其第二季度营业收入预计10.11亿元至10.51亿元，环比增长111%至119%；第二季度净利润预计5.02亿元至5.42亿元，环比增长262%-291%。
东芯股份在公告中表示，报告期内，受益于AI算力需求持续爆发及国内利基型存储技术竞争力的增强，海外存储原厂加速退出利基型市场，供给端收缩态势明显。与此同时，网络通信、工业控制、汽车电子等下游应用需求稳步复苏，利基型存储芯片供需缺口持续扩大。
其表示，“公司所面对的存储芯片市场的供需关系呈现结构性紧缺态势，产品市场价格大幅攀升。公司以SLCNANDFlash为代表的存储业务实现量价齐升，带动上半年营业收入实现大幅增长，整体毛利率亦显著上涨，带动上半年利润水平的显著提高。”
值得注意的是，东芯股份每年度持续施行股权激励计划，以充分绑定核心骨干与公司的长期发展利益。2026年上半年，公司因实施股权激励计划确认的股份支付费用对当期税后利润的影响数约为3900万元。
2026年上半年以来，利基芯片持续涨价，据TrendForce数据，2026年上半年NOR Flash合约价涨幅达100%-120%，SLC NAND达130%-150%。根据Gartner 研判，NANDFlash涨价周期有望延续至2027年以后。
对于后续市场行情，行业及公司均看好利基存储芯片的涨价持续性。
东芯股份在近期接受机构调研时表示，受益于海外存储巨头聚焦大容量3D NAND Flash及HBM/DDR5等高附加值产品，SLC NAND Flash、利基型DRAM等细分市场的供给持续收缩，供需结构持续优化，产品价格呈现上升趋势。目前行业层面普遍预期三季度海外原厂产能调整带来的影响将持续显现，利基存储芯片价格具备上行支撑。
从产能供应方面来看，东芯股份近期披露的投资者关系活动记录表显示，2026年，公司正有序加大主要存储产品线的生产与备货力度，以匹配下游市场需求的持续复苏。
华龙证券在近期研报中表示，利基芯片高景气度核心原因在于供给收缩、产能延迟释放以及需求复苏的多重共振。供给端，自2025年上半年以来，受海外存储大厂持续退出消费类市场、关停2DNAND产线影响，供给明显收缩，叠加从全球存储代工产能投放节奏来看，新一轮扩产周期的大规模产能释放落地需要时间，短期内供给缺口难以有效弥补；需求端，工控、网通、车用电子等应用稳步复苏。
除核心存储主业持续高增外，东芯股份同步推进多元化技术战略布局，公司以存储为核心，向“存、算、联”一体化领域进行技术布局，对砺算的战略投资是上述布局的一大举措。
工商信息显示，东芯股份为砺算科技的第一大股东，持股比例达35.87%。砺算科技系公司联营企业，其今年一季度推出的砺算LX系列显卡可支持端、云、边场景下的主流图形渲染 及一定的AI加速计算，产品可应用于个人电脑、专业设计、AI PC、云游戏、云渲染、数字孪生等应用场景。
东芯股份高管于5月在业绩说明会上表示，砺算科技现有的GPU产品暂未搭载公司的存储产品，未来双方将结合各自产品和技术优势，探讨业务协同与合作机会。
截至今日（7月20日）收盘，东芯股份报收于100.38元/股，总市值约444.1亿元。

### [E0825] 财联社电报 | 2026-07-20 16:02 UTC | 盘中
东芯股份Q2净利环比预增262%-291% 以SLC NAND Flash为代表的存储业务量价齐升
URL: https://www.cls.cn/detail/2431857
①东芯股份预计2026年半年度实现营业收入14.9亿元至15.3亿元，同比增长334.41%至346.07%；归母净利润预计为6.4亿元至6.8亿元，较上年同期扭亏为盈； ②东芯股份以SLC NAND Flash为代表的存储业务实现量价齐升，带动上半年营业收入实现大幅增长。
《科创板日报》7月20日讯（记者 陈俊清） 今日（7月20日）晚间，东芯股份公布2026年半年度业绩预告的自愿性披露公告。
公告显示，东芯股份预计2026年半年度实现营业收入14.9亿元至15.3亿元，同比增长334.41%至346.07%；归母净利润预计为6.4亿元至6.8亿元，较上年同期扭亏为盈，增加7.51亿元至7.91亿元；扣非净利润预计为6.3亿元至6.7亿元，同比扭亏为盈，增加7.57亿元至7.97亿元。
从季度维度来看，该公司业绩呈现环比持续向好趋势，其第二季度营业收入预计10.11亿元至10.51亿元，环比增长111%至119%；第二季度净利润预计5.02亿元至5.42亿元，环比增长262%-291%。
东芯股份在公告中表示，报告期内，受益于AI算力需求持续爆发及国内利基型存储技术竞争力的增强，海外存储原厂加速退出利基型市场，供给端收缩态势明显。与此同时，网络通信、工业控制、汽车电子等下游应用需求稳步复苏，利基型存储芯片供需缺口持续扩大。
其表示，“公司所面对的存储芯片市场的供需关系呈现结构性紧缺态势，产品市场价格大幅攀升。公司以SLCNANDFlash为代表的存储业务实现量价齐升，带动上半年营业收入实现大幅增长，整体毛利率亦显著上涨，带动上半年利润水平的显著提高。”
值得注意的是，东芯股份每年度持续施行股权激励计划，以充分绑定核心骨干与公司的长期发展利益。2026年上半年，公司因实施股权激励计划确认的股份支付费用对当期税后利润的影响数约为3900万元。
2026年上半年以来，利基芯片持续涨价，据TrendForce数据，2026年上半年NOR Flash合约价涨幅达100%-120%，SLC NAND达130%-150%。根据Gartner 研判，NANDFlash涨价周期有望延续至2027年以后。
对于后续市场行情，行业及公司均看好利基存储芯片的涨价持续性。
东芯股份在近期接受机构调研时表示，受益于海外存储巨头聚焦大容量3D NAND Flash及HBM/DDR5等高附加值产品，SLC NAND Flash、利基型DRAM等细分市场的供给持续收缩，供需结构持续优化，产品价格呈现上升趋势。目前行业层面普遍预期三季度海外原厂产能调整带来的影响将持续显现，利基存储芯片价格具备上行支撑。
从产能供应方面来看，东芯股份近期披露的投资者关系活动记录表显示，2026年，公司正有序加大主要存储产品线的生产与备货力度，以匹配下游市场需求的持续复苏。
华龙证券在近期研报中表示，利基芯片高景气度核心原因在于供给收缩、产能延迟释放以及需求复苏的多重共振。供给端，自2025年上半年以来，受海外存储大厂持续退出消费类市场、关停2DNAND产线影响，供给明显收缩，叠加从全球存储代工产能投放节奏来看，新一轮扩产周期的大规模产能释放落地需要时间，短期内供给缺口难以有效弥补；需求端，工控、网通、车用电子等应用稳步复苏。
除核心存储主业持续高增外，东芯股份同步推进多元化技术战略布局，公司以存储为核心，向“存、算、联”一体化领域进行技术布局，对砺算的战略投资是上述布局的一大举措。
工商信息显示，东芯股份为砺算科技的第一大股东，持股比例达35.87%。砺算科技系公司联营企业，其今年一季度推出的砺算LX系列显卡可支持端、云、边场景下的主流图形渲染 及一定的AI加速计算，产品可应用于个人电脑、专业设计、AI PC、云游戏、云渲染、数字孪生等应用场景。
东芯股份高管于5月在业绩说明会上表示，砺算科技现有的GPU产品暂未搭载公司的存储产品，未来双方将结合各自产品和技术优势，探讨业务协同与合作机会。
截至今日（7月20日）收盘，东芯股份报收于100.38元/股，总市值约444.1亿元。

### [E1002] MarketWatch Top Stories | 2026-07-20 17:14 UTC | 盘中
As Micron’s stock rebounds, Morgan Stanley says investors shouldn’t get spooked by the past
URL: https://www.marketwatch.com/story/as-microns-stock-rebounds-morgan-stanley-says-investors-shouldnt-get-spooked-by-the-past-6996574b?mod=mw_rss_topstories
Memory chips remain a bottleneck to AI development and demand looks durable, Morgan Stanley analyst says.

### [E1161] RMBS | 2026-07-20 18:29 UTC | 盘中
As AI, high-performance computing, cloud workloads, and data-intensive enterprise applications continue to scale, server memory systems are under growing pressure to deliver more bandwidth, higher capacity, better power efficiency, and stro
URL: https://www.rambus.com/blogs/complete-guide-to-ddr5-chipsets/
As AI, high-performance computing, cloud workloads, and data-intensive enterprise applications continue to scale, server memory systems are under growing pressure to deliver more bandwidth, higher capacity, better power efficiency, and stronger reliability. DDR5 memory was designed to address these needs by building on previous generations of DRAM with higher data rates, improved channel efficiency, enhanced power management, and greater module intelligence.

However, the performance of a modern DDR5 server memory module is not determined by DRAM alone. The broader DDR5 DIMM chipset plays a critical role in enabling the speed, signal integrity, power delivery, thermal awareness, and manageability required by today’s server platforms. Components such as the Registering Clock Driver (RCD), Power Management IC (PMIC), Serial Presence Detect Hub (SPD Hub), and Temperature Sensor (TS) work together to help DDR5 DIMMs scale across current and future generations of data center infrastructure.

This blog explores how DDR5 DIMM chipsets work, why they matter, how they build upon DDR5 technology, and how emerging module architectures such as MRDIMMs are extending memory bandwidth for demanding AI and data center workloads.

Table of Contents:

What is a DDR5 DIMM Chipset?

What Is the Difference Between Server and Client DDR5 DIMM Chipsets?

How has DDR5 evolved from previous memory generations?

What are the components that make up a DDR5 server RDIMM Chipset?

What is an MRDIMM?

What are the latest trends in DDR5 DIMM Chipset Technology?

Rambus DDR5 DIMM Chipset Solutions

What is a DDR5 DIMM Chipset?

A DDR5 DIMM chipset is a collection of specialized interface, power management, and monitoring chips that enable a DDR5 memory module to operate reliably at high speeds. For a server Registered DIMM (RDIMM) the chipset includes an RCD, PMIC, SPD Hub, and two discrete TS ICs.

While the DRAM devices store data, the DIMM chipset helps manage critical functions such as signal integrity, power delivery, system configuration, thermal monitoring, and memory bandwidth optimization. As DDR5 data rates continue to increase, these supporting chips have become just as important as the memory devices themselves in achieving high performance and reliability.

In modern server platforms, DDR5 DIMM chipsets enable memory modules to support larger capacities, faster speeds, and the demanding workloads associated with artificial intelligence, cloud computing, high-performance computing (HPC), and enterprise data centers.

What Is the Difference Between Server and Client DDR5 DIMM Chipsets?

Although both server and client DDR5 modules use DDR5 DRAM, they are designed for different system requirements and therefore use different chipset architectures.

Client DDR5 DIMMs

Client DIMMs are found in desktops, laptops, workstations, and consumer devices. Their primary focus is delivering strong performance at a cost-effective price point.

Client DDR5 modules include:

PMIC

SPD Hub

Clock Driver (on newer CUDIMM, CQDIMM and CSODIMM modules) at data rates starting at 6400 MT/s

Client DIMMs typically do not require the advanced signal buffering and management technologies used in server memory because they operate with fewer DRAM devices and relatively lower capacity requirements.

Server DDR5 RDIMMs

Server RDIMMs are designed for data centers, enterprise servers, cloud infrastructure, AI systems, and HPC platforms where reliability, capacity, and scalability are critical.

DDR5 Server RDIMM chipsets include:

Registering Clock Driver (RCD)

Power Management IC (PMIC)

SPD Hub (with integrated temperature sensor)

Temperature Sensor (TS) IC x2

These additional chips help maintain signal integrity, manage power delivery, monitor thermals, and support high-capacity memory configurations that may contain dozens of DRAM devices operating at very high speeds.

How has DDR5 evolved from previous memory generations?

DDR5 was created as the successor to DDR4, with a focus on higher bandwidth, improved efficiency, and greater scalability. The JEDEC DDR5 SDRAM standard defines the specification for DDR5 devices, including functionality, electrical characteristics, packages, and signal assignments, and was created based on earlier DDR standards including DDR4.

One of the key advantages of DDR5 is its ability to support higher data rates while improving how memory resources are organized and accessed. Compared with previous generations, DDR5 introduces architectural enhancements that help increase effective bandwidth and support larger memory capacities. In server systems, these advances are especially important because processors continue to add more cores, accelerators require higher data throughput, and workloads increasingly depend on rapid access to large data sets.

DDR5 also changes the way memory modules handle power and management functions. Instead of relying solely on motherboard-based power regulation, DDR5 modules incorporate power management (the PMIC) directly on the DIMM. This shift improves local control of power delivery and helps support higher performance memory operation.

The addition of components such as the PMIC, SPD Hub, and Temperature Sensor gives the module more local intelligence than previous generations, enabling better configuration, monitoring, and system-level optimization.

Why do DDR5 DIMMs need more on-module intelligence?

As DDR5 data rates increase, the memory module becomes a more complex electrical and thermal environment. Signals must travel between the memory controller and many DRAM devices with extremely tight timing margins. Power must be delivered at low voltages with high current requirements. Thermal conditions must be monitored carefully because dense platforms can place many memory modules close together with limited airflow.

This is why DDR5 DIMMs rely so heavily on companion chips. The RCD helps buffer and distribute command, address, clock, and control signals. The PMIC supports local power regulation. The SPD Hub stores and communicates module configuration information. Temperature Sensors provide thermal feedback that can be used by the system to manage performance and reliability.

The growing importance of these chips reflects a broader shift in memory architecture. DDR5 DIMMs are not passive collections of DRAM devices. They are increasingly intelligent subsystems that combine DRAM, interface logic, power management, monitoring, and control functions on a single module.

What are the components that make up a DDR5 server RDIMM Chipset?

Server processors often communicate with hundreds of gigabytes or even terabytes of memory across multiple RDIMMs. At these scales, maintaining clean clock signals, precise timing, and stable power delivery becomes significantly more challenging than in a desktop or laptop system.

In an RDIMM:

An RCD buffers and redistributes command, address, clock, and control signals across the RDIMM.

A PMIC regulates power locally on the module.

A Temperature Sensor (TS) provides thermal telemetry to support reliability and system management. The RDIMM chipset includes two (2) discrete TS ICs.

An SPD Hub stores and communicates module identity and has an integrated temperature sensor, which along with the two discrete TS ICs above, provides three points of thermal telemetry from the RDIMM.

With these additional components, server memory can scale to the capacities and speeds demanded by modern data centers.

What is a Registering Clock Driver (RCD)?

The Registering Clock Driver, or RCD, is one of the most important chips on a DDR5 RDIMM. Its primary function is to buffer and re-drive command, address, clock, and control signals between the processor memory controller and the DRAM devices on the module.

In high-capacity server RDIMMs, the memory controller must communicate with many DRAM devices across the module. Without signal buffering, electrical loading and timing challenges would make it diff

### [E0045] The Information | 2026-07-20 13:32 UTC | 盘中
It’s a similar story if you look at the three companies on a multiple of next year’s expected earnings, with Samsung at 4.4 times, SK Hynix at 4.6 times and Micron at 6 times, according to S&P Global Market Intelligence**.** While some anal
URL: https://www.theinformation.com/articles/samsung-new-memory-chip-underdog-now
It’s a similar story if you look at the three companies on a multiple of next year’s expected earnings, with Samsung at 4.4 times, SK Hynix at 4.6 times and Micron at 6 times, according to S&P Global Market Intelligence**.** While some analysts are starting to consider the companies on a price-to-earnings basis, their history of volatility—which can plunge them swiftly into the red—makes that a difficult metric to use.

“If you ask me, from a risk-to-reward position, I do think Samsung here is the underdog, and so perhaps there’s a better chance of [its stock price] rerating up versus SK Hynix,” said Jing Jie Yu, an equity analyst at Morningstar who covers the two Korean chipmakers.

All three stocks have soared in the past year, with Micron and SK Hynix both up close to 700%, while Samsung stock is up nearly 300%. Despite not having risen as much as the other two, Samsung has fallen by more than Micron and roughly the same amount as SK Hynix in the recent selloff, which started about a month ago.

Samsung is the only one of the three not currently listed in the U.S., however, which may present a problem for some investors. Micron is a U.S. company, while Hynix—which, like Samsung, is South Korean—listed its shares on the Nasdaq last Thursday, in a $26.5 billion offering that marked the largest initial public offering ever by a foreign company in the U.S. (Samsung on Tuesday [denied a Bloomberg report](https://www.reuters.com/world/asia-pacific/samsung-explores-potential-us-listing-via-adrs-bloomberg-news-reports-2026-07-14/) that it was considering a U.S. listing.)

On the Nasdaq, Hynix shares have fallen 8% since the listing, partly a result of worries about long-term pricing and oversupply. We wrote in April that investors [were overlooking](https://www.theinformation.com/articles/sk-hynix-overlooked-memory-chip-maker) SK Hynix relative to Micron, but now that Hynix has also listed shares in the U.S., Samsung is the more obvious laggard.

**Riding the Memory Boom**

Samsung’s semiconductor unit, which primarily produces memory chips useful for AI as well as some simpler semiconductors for smartphones and cameras, made up 61% of the company’s sales in the March quarter. The rest came primarily from its sales of phones, fridges, TV sets and gaming monitors.

Investors are mainly focused on the semiconductor unit because of its extraordinary growth—226% higher revenue in the quarter and 4,781% higher operating profit relative to the year-ago period—which meant that it accounted for 94% of the company’s operating profit in the first quarter. Three years ago, Samsung’s semiconductors lost money and accounted for 22% of its revenue.

At an 11% margin, Samsung generated more free cash flow per revenue dollar last year than Micron, which had a 4% margin in its fiscal 2025 ended in August.

Micron and Hynix also experienced rocket-ship–like growth in their revenues in recent quarters. Each is benefiting from charging higher prices to customers who, at least for now, are desperate to get their hands on those chips.

Yu estimates that Samsung raised its prices for DRAM memory, a fast type of memory used to feed data to a graphics processing unit, by 42% and NAND memory, a long-term storage vault used to hold AI datasets, by 93% in the June quarter. (Samsung said earlier this month its June-quarter revenues jumped 129%, but more details will come when it reports earnings on July 30.)

In a report this month, he projected a downturn in 2029 and 2030, when the long-term agreements Samsung has signed with customers this year will likely begin to lapse.

Soaring prices are rippling through the electronics industry, forcing companies such as Apple to raise prices for hardware.

To meet surging demand, all three memory chip makers are planning to add factory capacity. SK Hynix intends to use the money it raised in its Nasdaq listing to support its goal, outlined last month, of at least [doubling](https://www.reuters.com/world/asia-pacific/samsung-explores-potential-us-listing-via-adrs-bloomberg-news-reports-2026-07-14/) its production capacity within the next five years.

Samsung is moving quickly, too; the company [told The Information](https://www.theinformation.com/briefings/samsung-races-add-chip-capacity-moves-plant-2029) recently that it aims to begin operations at its first chipmaking plant in the South Korean city of Yongin two years earlier than planned, in 2029. Micron, for its part, [upped its spending commitment](https://www.reuters.com/business/micron-invest-up-3-billion-us-chip-supply-chain-2026-07-09/) last week to $250 billion, which it plans to use for building and running new factories in the U.S. through 2035.

But it takes at least a year or two to get new manufacturing capacity up and running, which means there’s a risk the memory chip makers will end up overbuilding if demand slows in the interim. Increasing competition in the memory chip market could exacerbate the problem, as other companies besides the three largest manufacturers are also ramping up supply.

China’s CXMT, for instance, has grown its market share in DRAM to 8% in the first quarter from 3% a year earlier, [according to Counterpoint Research](https://counterpointresearch.com/en/insights/global-dram-and-hbm-market-share), and plans to [ramp up](https://www.reuters.com/world/china/chinas-cxmt-wins-3-billion-memory-supply-deal-with-tencent-sources-say-2026-06-29/) its scale to produce roughly as many DRAM chips as Micron does each month by the end of the year. CXMT’s chips are off- limits to U.S. buyers on regulatory grounds, although Apple has been [testing the chips](https://www.bloomberg.com/news/articles/2026-07-01/apple-seeks-to-buy-chinese-made-memory-chips-with-lobbying-push) in its devices and lobbying the U.S. government to change that.

Japanese firm Nanya, too, is ramping up production of its memory chips.

Still, Morningstar’s Yu thinks a scenario where supply far outstrips demand is still a few years away, largely because the earliest available source of substantial new capacity—SK Hynix’s new factory in Yongin, which should be completed by February or March 2027—probably won’t come online for another year after that, as the company needs to procure and install new equipment in the plant once it’s built.

That suggests the memory chip makers can hold onto their pricing power for a while longer. For now, though, investors might still be able to capture some remaining upside in the AI-driven memory chip boom.

“Higher prices will definitely last longer in our view through 2027 at the very least,” Yu said. “The real test really to us comes in 2028, where we don’t really have that much confidence that demand will be there.”

### [E0846] 研报平权 | 2026-07-20 16:06 UTC | 盘中
存储深度：美光40%回购潜力、现金流重估与周期风险 #公众号# #存储#
URL: https://wx.zsxq.com/dweb2/index/topic_detail/14422142554114212
存储深度：美光40%回购潜力、现金流重估与周期风险 #公众号# #存储#

### [E0884] 研报平权 | 2026-07-20 16:24 UTC | 盘中
摩根士丹利刚刚预测，美光未来两年的利润将达到 4000亿美元；瑞银则表示，凭借这些现金，该公司到 2028 年可能回购超过 40% 的股份（收藏这条）。
摩根士丹利的模型显示，美光在 CY2026 将实现 1729亿美元 的营收和 1407亿美元 的营业利润，营业利润率为 81.4%。
到 CY2027，营收将跃升至 2859亿美元，营业利润将达到 2487亿美元，营业利润率为 87%；将这两年相加，营业利润合计约为 3890亿美元
在该模型中，毛利率介于 85% 至 89.
URL: https://wx.zsxq.com/dweb2/index/topic_detail/55522452115841184
摩根士丹利刚刚预测，美光未来两年的利润将达到 4000亿美元；瑞银则表示，凭借这些现金，该公司到 2028 年可能回购超过 40% 的股份（收藏这条）。
摩根士丹利的模型显示，美光在 CY2026 将实现 1729亿美元 的营收和 1407亿美元 的营业利润，营业利润率为 81.4%。
到 CY2027，营收将跃升至 2859亿美元，营业利润将达到 2487亿美元，营业利润率为 87%；将这两年相加，营业利润合计约为 3890亿美元
在该模型中，毛利率介于 85% 至 89.5% 之间，对于一家存储芯片公司而言，这几乎是闻所未闻的数字，因为这个行业素以利润率低且具有周期性而著称。
瑞银正在同一逻辑基础上加入股东回报这一视角。
该机构表示，到 2028 年，美光可能产生超过 4000亿美元 的自由现金流。
一旦其回购限制于 2026 年 12 月到期，这些现金可能足以支持其在 2028 年底前回购超过 40% 的流通股。
因此，摩根士丹利预测的这轮由 AI 驱动的利润激增，正是瑞银认为可能为美光大幅缩减股本提供资金的来源。
美光的 HBM 在多年期合同下已全部售罄至 2026 年，CEO Sanjay Mehrotra 表示，该公司只能满足客户总需求的二分之一至三分之二。
这种供需缺口正在形成这些数字背后的定价权，而增长动力则来自 AI 从训练大规模模型转向以推理为主的智能体应用，这类应用以不同方式、在更大规模上消耗内存。
英伟达或 AMD 的每一颗 AI GPU 都需要多层堆叠的 HBM，而美光是全球仅有的三家能够规模化生产 HBM 的供应商之一。
预计 HBM 市场本身将从 2025 年的 350亿美元 增长至 2028 年的 1000亿美元，而美光的订单簿目前已经延伸至 2027 年。
即使美光只实现摩根士丹利预测数字的 70%，接近 4000亿美元 的累计营业利润，再加上 40% 以上的股份缩减，也将意味着结构性重估，而不仅仅是周期性上行。
我仍然看好存储股，Milk Road Pro 会员通过这些交易已经获得巨额收益；使用下方链接，只需 1 美元即可加入 Milk Road Pro，获取我们的完整投资逻辑！ #资讯# #存储#


### [E1079] 新浪财经7x24 | 2026-07-20 17:47 UTC | 盘中
【高盛称对冲基金以创纪录的速度抛售美国科技股】高盛集团的主经纪商业务部门指出，过去两个月对冲基金撤出美国科技股的速度达到了创纪录的水平。以Vincent Lin为首的该团队分析称，这类基金过去八周的六个星期都是该板块的净卖方。市值累计遭减持...
URL: https://wap.cj.sina.cn/pc/7x24/4998399
【高盛称对冲基金以创纪录的速度抛售美国科技股】高盛集团的主经纪商业务部门指出，过去两个月对冲基金撤出美国科技股的速度达到了创纪录的水平。以Vincent Lin为首的该团队分析称，这类基金过去八周的六个星期都是该板块的净卖方。市值累计遭减持的幅度约达10%，创出十多年前开始统计该数据以来之最。“在半导体/存储/人工智能基础设施板块持续震荡和被大幅抛售的情况下，6月初以来卖盘的持续性和规模表明科技投资者大幅减少了仓位，而且一些投降迹象也开始出现，” 该部门写道。

### [E1081] 华尔街见闻快讯 | 2026-07-20 17:48 UTC | 盘中
高盛称对冲基金以创纪录的速度抛售美国科技股，一些投降迹象也开始出现
URL: https://wallstreetcn.com/livenews/3137052
高盛集团的主经纪商业务部门指出，过去两个月对冲基金撤出美国科技股的速度达到了创纪录的水平。 以Vincent Lin为首的该团队分析称，这类基金过去八周的六个星期都是该板块的净卖方。市值累计遭减持的幅度约达10%，创出十多年前开始统计该数据以来之最。

“在半导体/存储/人工智能基础设施板块持续震荡和被大幅抛售的情况下，6月初以来卖盘的持续性和规模表明科技投资者大幅减少了仓位，而且一些投降迹象也开始出现，” 该部门写道。 高盛指出，上周科技板块表现最差，也是美国股市中净卖出最多的领域，因多头仓位减少的同时空仓攀升。大多数子板块均出现净卖出，其中科技硬件、存储及外设和IT服务板块最为严重，而半导体和软件板块的净卖出则相对较少。 高盛由Ben Snider牵头的策略师表示，“热门人工智能基础设施股令人痛苦的震荡”重新激发了人们对其它板块的兴趣。

高盛称对冲基金以创纪录的速度抛售美国科技股，一些投降迹象也开始出现

### [E1110] CNBC Top News | 2026-07-20 18:03 UTC | 盘中
Don't be fooled by an AI stock bounce. Charts suggest more downside ahead
URL: https://www.cnbc.com/2026/07/20/dont-be-fooled-by-an-ai-stock-bounce-charts-suggest-more-downside-ahead.html
The AI trade has been one of the most important sources of upside leadership for the market, but its technical profile has weakened meaningfully in recent weeks. The Global X Artificial Intelligence & Technology ETF (AIQ) has seen a significant correction in both absolute and relative terms, presenting a challenge for the group in the third quarter. AIQ has retreated nearly 19% since peaking last month. The weekly MACD has shifted to a "sell" signal, which is a bearish development for the next several weeks and suggests the correction has not yet fully run its course. The weekly stochastics are also pointing lower, reinforcing the loss of upside momentum after the ETF became extended in June. The composition of AIQ helps explain why the correction has been so important for the broader market. Its top holdings include several stocks that have been central to the AI trade, including SK Hynix , Micron , AMD , Apple , and NVIDIA . This gives AIQ meaningful exposure to semiconductors, AI infrastructure, and mega-cap technology, making it a useful proxy for the broader AI theme. The daily chart suggests AIQ is due for near-term stabilization before the next leg lower unfolds. The ETF has become short-term oversold and is testing support from the daily cloud model, which is a natural area for a bounce to unfold. This should help alleviate some of the immediate downside pressure, but it does not negate the deterioration in intermediate-term momentum on the weekly chart. We would view a near-term rebound as an interruption in the correction rather than the start of a renewed uptrend. Initial resistance is at the 50-day moving average and daily cloud, in the $64-$65 zone. A breakdown below cloud-based support near $57 would officially reverse the uptrend off the March low and increase immediate downside risk toward secondary support near $54, which is defined by former resistance. The relative chart has a similar takeaway. AIQ has broken down versus the SPX after a strong period of outperformance, confirming that leadership has shifted away from the AI trade. The ratio is oversold in the short term, so brief relative stabilization would be natural, but the loss of momentum is meaningful enough to suggest additional underperformance is likely this summer. While AIQ remains in a long-term uptrend, a significant loss of momentum suggests it is prone to an intermediate-term correction. Short-term oversold conditions should allow for stabilization in AI stocks in the very near term, but we would use strength as an opportunity to reduce exposure until momentum improves. —Katie Stockton with Will Tamplin Access research from Fairlead Strategies for free here . DISCLOSURES: None All opinions expressed by the CNBC Pro contributors are solely their opinions and do not reflect the opinions of CNBC, or its parent company or affiliates, and may have been previously disseminated by them on television, radio, internet or another medium. THIS CONTENT IS PROVIDED FOR INFORMATIONAL PURPOSES ONLY AND DOES NOT CONSTITUTE FINANCIAL, INVESTMENT, TAX OR LEGAL ADVICE OR A RECOMMENDATION TO BUY ANY SECURITY OR OTHER FINANCIAL ASSET. THE CONTENT IS GENERAL IN NATURE AND DOES NOT REFLECT ANY INDIVIDUAL'S UNIQUE PERSONAL CIRCUMSTANCES. THE ABOVE CONTENT MIGHT NOT BE SUITABLE FOR YOUR PARTICULAR CIRCUMSTANCES. BEFORE MAKING ANY FINANCIAL DECISIONS, YOU SHOULD STRONGLY CONSIDER SEEKING ADVICE FROM YOUR OWN FINANCIAL OR INVESTMENT ADVISOR. Click here for the full disclaimer. Fairlead Strategies Disclaimer: This communication has been prepared by Fairlead Strategies LLC ("Fairlead Strategies") for informational purposes only. This material is for illustration and discussion purposes and not intended to be, nor construed as, financial, legal, tax or investment advice. You should consult appropriate advisors concerning such matters. This material presents information through the date indicated, reflecting the author's current expectations, and is subject to revision by the author, though the author is under no obligation to do so. This material may contain commentary on broad-based indices, market conditions, different types of securities, and cryptocurrencies, using the discipline of technical analysis, which evaluates the demand and supply based on market pricing. The views expressed herein are solely those of the author. This material should not be construed as a recommendation, or advice or an offer or solicitation with respect to the purchase or sale of any investment. The information is not intended to provide a basis on which you could make an investment decision on any particular security or its issuer. This document is intended for CNBC Pro subscribers only and is not for distribution to the general public. Certain information has been provided by and/or is based on third party sources and, although such information is believed to be reliable, no representation is made with respect to the accuracy, completeness, or timeliness of such information. This information may be subject to change without notice. Fairlead Strategies undertakes no obligation to maintain or update this material based on subsequent information and events or to provide you with any additional or supplemental information or any update to or correction of the information contained herein. Fairlead Strategies, its officers, employees, affiliates and partners shall not be liable to any person in any way whatsoever for any losses, costs, or claims for your reliance on this material. Nothing herein is, or shall be relied on as, a promise or representation as to future performance. PAST PERFORMANCE IS NOT INDICATIVE OF FUTURE RESULTS. Opinions expressed in this material may differ or be contrary to opinions expressed, or actions taken, by Fairlead Strategies or its affiliates, or their respective officers, directors, or employees. In addition, any opinions and assumptions expressed herein are made as of the date of this communication and are subject to change and/or withdrawal without notice. Fairlead Strategies or its affiliates may have positions in financial instruments mentioned, may have acquired such positions at prices no longer available, and may have interests different from or adverse to your interests or inconsistent with the advice herein. Any investments made are made under the same terms as nonaffiliated investors and do not constitute a controlling interest. No liability is accepted by Fairlead Strategies, its officers, employees, affiliates, or partners for any losses that may arise from any use of the information contained herein. Any financial instruments mentioned herein are speculative in nature and may involve risk to principal and interest. Any prices or levels shown are either historical or purely indicative. This material does not take into account the particular investment objectives or financial circumstances, objectives or needs of any specific investor, and are not intended as recommendations of particular securities, investment products, or other financial products or strategies to particular clients. Securities, investment products, other financial products or strategies discussed herein may not be suitable for all investors. The recipient of this information must make its own independent decisions regarding any securities, investment products or other financial products mentioned herein. The material should not be provided to any person in a jurisdiction where its provision or use would be contrary to local laws, rules, or regulations. This material is not to be reproduced or redistributed absent the written consent of Fairlead Strategies.

## 研究交付要求

这是一条已经从整场材料语义拆出的独立催化，不要重新把它压回大盘总评。核实美股异动原因和传播时序；用结构化供应链、内部 KB 与公司披露坐实每个亚洲候选；逐腿查真实行情和本地市场 price-in；候选不成立也要逐条给出有证据的 AVOID。