存储颗粒九月涨价传到A股模组厂要多久，江波龙和佰维存储的单季毛利率在三季报里会不会见到
- TrendForce的DRAM合约价环比涨幅：2026年一季度93%到98%，二季度约80%，三季度约40%，四季度预估20%
- 国家统计局9月CPI里，数据存储设备价格的环比涨幅比8月的2.1%还小
- TrendForce预估四季度DRAM合约价还要环比涨20%，如果落地，四季度售价端重新加速，而成本池按加权平均只会慢慢跟，四季度毛利率完全可能掉头向上——九月涨价的利润不是消失了，是被推到了四季度和明年一季度
- 江波龙三季度收入139亿到159亿元，跟二季度的141.80亿元比是少2%到多12%
- 三季度扣非净利润37.53亿到47.53亿元，比二季度的61.05亿元少22%到39%
- 已经出了三季报的德明利把方向写实了：三季度单季毛利率32.89%，二季度38.84%，一个季度掉5.95个百分点
存储颗粒九月涨价传到A股模组厂要多久，江波龙和佰维存储的单季毛利率在三季报里会不会见到
The desk answers已核实Walnut, from public filings
三季报里见不到九月涨价的钱，只会见到涨价减速的代价。江波龙三季度收入139亿到159亿元，跟二季度的141.80亿元比是少2%到多12%；三季度扣非净利润37.53亿到47.53亿元，比二季度的61.05亿元少22%到39%。收入没退，利润退了两三成，这个缺口只能落在毛利率上。已经出了三季报的德明利把方向写实了：三季度单季毛利率32.89%，二季度38.84%，一个季度掉5.95个百分点。
先把问题里的一个前提掰正。九月不是新一轮涨价的起点，是涨幅一路收窄过程中的一个月。TrendForce的DRAM合约价环比涨幅：2026年一季度93%到98%，二季度约80%，三季度约40%，四季度预估20%。国家统计局9月CPI里，数据存储设备价格的环比涨幅比8月的2.1%还小。九月确实在涨，但斜率只剩一季度的五分之一，四季度还要再砍一半。拿"新一轮涨价"去推传导时点，方向会推反。
时滞本身不长。江波龙9月17日到21日的投资者关系活动记录里，公司自己把会计口径讲清楚了："公司采用月末一次加权平均法确定发出存货的实际成本，存储晶圆价格上涨的影响已动态反映在实际营业成本中。"月末一次加权平均的意思是，九月买的颗粒，九月就进了成本池，没有先进先出那种半年以上的滞后。
进了池子不等于改了毛利率，要看新采购占池子多大比重。江波龙二季度营业成本54.93亿元，存货从179.61亿元升到257.77亿元，加进78.16亿元采购，单季采购约133亿元，相当于期末257.77亿元存货的51.6%。按这个比例，一轮20%的采购涨价，一个季度把加权平均成本推高约10%，两个季度基本把全池换一遍。所以"传多久"的答案是1到2个季度——不是半年，但也不是当月。
问题在于这两个季度正好撞上售价端减速。DRAM合约价三季度涨约40%，二季度涨约80%；成本端消化的是二季度按80%涨幅买进来的晶圆，售价端只按40%提。一头涨得快、一头提得慢，毛利率只能往下走。这就是江波龙在业绩预告里的原话："受前期较高价格基数影响，DRAM和NAND Flash价格涨幅较前期明显收窄，存储价格上涨对公司盈利的边际贡献显著趋缓。"
四家模组厂放一起看，方向一致，节奏不同。德明利：一季度57.42%，二季度38.84%，三季度32.89%，一路往下；三季度末存货289.86亿元，只够3.71个季度（337天）的成本。江波龙：一季度55.53%，二季度61.26%，三季度毛利率要等10月29日董事会审议三季报，但扣非净利率已经从二季度的43.05%掉到23.60%到34.19%。朗科科技：一季度25.25%，二季度14.67%。唯一的例外是分销商香农芯创，一季度9.12%升到二季度13.23%——它不做制造，低价库存按市价卖，涨价直接变成价差，不吃成本滞后这一刀。
佰维要单独说，因为它至今没给数。10月24日那份《关于公司近期生产经营情况的公告》通篇没有一个季度数字，只有"DRAM、NAND Flash等主要存储产品价格显著上涨"这类描述，也没有业绩预告。它二季度单季毛利率55.98%，一季度53.30%，方向跟江波龙一样是二季度见高；6月末存货181.68亿元，相当于上半年营业成本的5.16个季度、467天。存货厚只决定涨价被摊薄多少，决定不了方向；它三季度毛利率，现在从公开文件推不出来。
有个反方论点必须放上来，因为它成立的概率不低。TrendForce预估四季度DRAM合约价还要环比涨20%，如果落地，四季度售价端重新加速，而成本池按加权平均只会慢慢跟，四季度毛利率完全可能掉头向上——九月涨价的利润不是消失了，是被推到了四季度和明年一季度。证据是德明利存货只够3.71个季度、三季度先掉，江波龙存货够5.21个季度、缓冲更厚、摔得更晚。
但这个反方论点有个缺口。江波龙自己说三季度经营性现金流量净额转正、期末资产负债率降到50%以下（二季度末56.13%），说明它三季度没继续加库存，反而在收。便宜库存不再补，四季度成本跟涨价的距离会拉近，重新扩张的力度要打折。这是"利润推到四季度"最脆弱的地方。
三季度毛利率能倒推到这一步。江波龙三季度每1个百分点毛利率，等于1.39亿到1.59亿元毛利；三季度扣非比二季度少的13.5亿到23.5亿元，按二季度15.4%的有效税率折算成税前，是15.98亿到27.80亿元，正好对应11到17个百分点的毛利率。前提是费用率11.0%、减值率5.9%、有效税率15.4%三项不变，那样三季度单季毛利率落在44%到50%之间，比二季度的61.26%掉11到17个点。这个数是倒推的，前提写明，10月29日就能验。
证伪条件三条。江波龙三季报单季毛利率如果到55%以上，说明三季度的利润落差主要不是毛利造成的，而是费用或减值，上面的归因就错了。TrendForce明年1月把四季度DRAM合约价涨幅修正到10%以下，"售价端重新加速、利润推到四季度"这条就断了。佰维三季报单季毛利率如果环比上升，说明模组厂之间存在产品结构和客户结构的实质差异，不能拿江波龙和德明利的方向去套。
来源：深圳市江波龙电子股份有限公司《2026年前三季度业绩预告》（2026年10月15日，公告编号2026-117）；《投资者关系活动记录表》编号2026-010（2026年9月22日披露，记录9月17日至21日调研）；《投资者关系活动记录表》编号2026-011（2026年9月29日）；H股《董事会会议召开日期》公告（2026年10月16日，股份代号09976）；《2026年半年度报告》《2026年第一季度报告》；深圳佰维存储科技股份有限公司《关于公司近期生产经营情况的公告》（2026年10月24日，公告编号2026-100）；《2026年半年度报告》《2026年第一季度报告》；深圳市德明利技术股份有限公司《2026年第三季度报告》（2026年10月26日，公告编号2026-093）；朗科科技《2026年半年度报告》；香农芯创《2026年半年度报告》；TrendForce集邦咨询DRAM合约价季度环比预估（经《21世纪经济报道》2026年10月23日《透视22份半导体成绩单》引述）；国家统计局2026年9月CPI数据（经《每日经济新闻》2026年10月14日《9月存储价格涨幅收窄》引述）。
How long does it take for the September memory chip price increase to reach A-share module makers, and will Jiangbolong and Baiwei Storage's single-quarter gross margins show it in the third-quarter report?
The third-quarter report will not show the money from the September price increase; it will only show the cost of the price increase decelerating. Jiangbolong's third-quarter revenue is 13.9 billion to 15.9 billion yuan, which compared with 14.180 billion yuan in the second quarter is down 2% to up 12%; third-quarter net profit excluding non-recurring items is 3.753 billion to 4.753 billion yuan, down 22% to 39% from 6.105 billion yuan in the second quarter. Revenue has not retreated, but profit has retreated by 20% to 30%, and this gap can only land on gross margin. Demingli, which has already released its third-quarter report, has made the direction concrete: third-quarter single-quarter gross margin was 32.89%, versus 38.84% in the second quarter, a drop of 5.95 percentage points in one quarter.
First, correct one premise in the question. September is not the starting point of a new round of price increases; it is a month in the process of the increase rate narrowing all the way. TrendForce's DRAM contract price quarter-on-quarter increase: 93% to 98% in the first quarter of 2026, about 80% in the second quarter, about 40% in the third quarter, and an estimated 20% in the fourth quarter. In the National Bureau of Statistics' September CPI, the quarter-on-quarter increase in data storage device prices was even smaller than August's 2.1%. September was indeed rising, but the slope was only one-fifth of the first quarter, and the fourth quarter will cut it by another half. Using "a new round of price increases" to infer the transmission timing will push the direction the wrong way.
The time lag itself is not long. In Jiangbolong's investor relations activity records from September 17 to 21, the company itself explained the accounting basis clearly: "The company uses the monthly weighted average method to determine the actual cost of inventory issued, and the impact of rising storage wafer prices has been dynamically reflected in actual operating costs." Monthly weighted average means that wafers bought in September enter the cost pool in September; there is no first-in-first-out lag of more than half a year.
Entering the pool does not mean the gross margin has changed; it depends on how large a share new purchases are of the pool. Jiangbolong's second-quarter operating cost was 5.493 billion yuan, inventory rose from 17.961 billion yuan to 25.777 billion yuan, adding 7.816 billion yuan of purchases, with single-quarter purchases of about 13.3 billion yuan, equivalent to 51.6% of the 25.777 billion yuan of inventory at period end. At this ratio, a round of 20% purchase price increases would push weighted average cost up by about 10% in one quarter, and two quarters would basically turn over the whole pool. So the answer to "how long does it transmit" is 1 to 2 quarters—not half a year, but not the same month either.
The problem is that these two quarters happen to coincide with deceleration on the selling price side. DRAM contract prices rose about 40% in the third quarter and about 80% in the second quarter; the cost side is digesting wafers bought in the second quarter at an 80% increase, while the selling price side is only raised by 40%. One end rises fast and the other is raised slowly, so gross margin can only move down. This is Jiangbolong's exact wording in its earnings forecast: "Affected by the earlier relatively high price base, the price increases of DRAM and NAND Flash have narrowed significantly compared with the earlier period, and the marginal contribution of storage price increases to the company's profit has slowed markedly."
Looking at the four module makers together, the direction is the same but the pace differs. Demingli: 57.42% in the first quarter, 38.84% in the second quarter, 32.89% in the third quarter, declining all the way; inventory at the end of the third quarter was 28.986 billion yuan, enough for only 3.71 quarters (337 days) of cost. Jiangbolong: 55.53% in the first quarter, 61.26% in the second quarter; the third-quarter gross margin must wait for the board meeting on October 29 to review the third-quarter report, but the net margin excluding non-recurring items has already fallen from 43.05% in the second quarter to 23.60% to 34.19%. Langke Technology: 25.25% in the first quarter, 14.67% in the second quarter. The only exception is the distributor Shannong Xinchuan, which rose from 9.12% in the first quarter to 13.23% in the second quarter—it does not do manufacturing, and sells low-priced inventory at market prices, so price increases directly become price spreads and it does not take the hit from cost lag.
Baiwei must be discussed separately because it has still not given figures. Its October 24 "Announcement on the Company's Recent Production and Operating Conditions" contains not a single quarterly number throughout, only descriptions such as "prices of major storage products such as DRAM and NAND Flash have risen significantly," and there is no earnings forecast either. Its second-quarter single-quarter gross margin was 55.98%, and first-quarter was 53.30%, with the direction the same as Jiangbolong's, peaking in the second quarter; inventory at the end of June was 18.168 billion yuan, equivalent to 5.16 quarters, or 467 days, of first-half operating cost. Thick inventory only determines how much the price increase is diluted; it cannot determine direction; its third-quarter gross margin cannot now be inferred from public documents.
One opposing argument must be put forward because its probability of holding is not low. TrendForce estimates that DRAM contract prices will rise another 20% quarter-on-quarter in the fourth quarter. If that materializes, the selling price side will accelerate again in the fourth quarter, while the cost pool will only follow slowly under weighted average, so fourth-quarter gross margin could very well turn upward—the profit from the September price increase has not disappeared, but has been pushed to the fourth quarter and the first quarter of next year. The evidence is that Demingli's inventory is enough for only 3.71 quarters and it fell first in the third quarter, while Jiangbolong's inventory is enough for 5.21 quarters, its buffer is thicker, and it falls later.
But this opposing argument has a gap. Jiangbolong itself said that third-quarter net operating cash flow turned positive and the debt-to-asset ratio at period end fell below 50% (56.13% at the end of the second quarter), indicating that it did not continue to add inventory in the third quarter but was instead pulling back. Cheap inventory is no longer being replenished, so in the fourth quarter the distance between cost and price increases will narrow, and the strength of re-expansion must be discounted. This is the most fragile point in "profit pushed to the fourth quarter."
The third-quarter gross margin can be back-calculated to this point. For Longsys in the third quarter, each 1 percentage point of gross margin equals 1.39 yi yuan to 1.59 yi yuan of gross profit; the 13.5 yi yuan to 23.5 yi yuan by which third-quarter excluding non-recurring items is less than the second quarter, converted to pre-tax at the second quarter's 15.4% effective tax rate, is 15.98 yi yuan to 27.80 yi yuan, which corresponds exactly to 11 to 17 percentage points of gross margin. The premise is that the expense ratio of 11.0%, the impairment ratio of 5.9%, and the effective tax rate of 15.4% all remain unchanged; on that basis, the third-quarter single-quarter gross margin falls between 44% and 50%, down 11 to 17 points from the second quarter's 61.26%. This figure is back-calculated, the premises are stated, and it can be verified on October 29.
There are three falsification conditions. If Longsys's third-quarter report shows a single-quarter gross margin of 55% or above, that means the third quarter's profit shortfall was mainly caused not by gross profit but by expenses or impairment, and the attribution above is wrong. If TrendForce revises its fourth-quarter DRAM contract price increase to below 10% in January next year, the line that "the selling-price side re-accelerates and profit is pushed into the fourth quarter" breaks. If Biwin's third-quarter report shows a single-quarter gross margin rising quarter-on-quarter, that means there are substantive differences in product mix and customer mix among module makers, and the direction of Longsys and Demingli cannot be applied to them.
Sources: Shenzhen Longsys Electronics Co., Ltd. "2026 First Three Quarters Earnings Forecast" (October 15, 2026, announcement no. 2026-117); "Investor Relations Activity Record" no. 2026-010 (disclosed September 22, 2026, recording the September 17 to 21 research visit); "Investor Relations Activity Record" no. 2026-011 (September 29, 2026); H-share "Date of Board Meeting" announcement (October 16, 2026, stock code 09976); "2026 Semi-Annual Report" and "2026 First Quarter Report"; Shenzhen Biwin Storage Technology Co., Ltd. "Announcement on the Company's Recent Production and Operations" (October 24, 2026, announcement no. 2026-100); "2026 Semi-Annual Report" and "2026 First Quarter Report"; Shenzhen Demingli Technology Co., Ltd. "2026 Third Quarter Report" (October 26, 2026, announcement no. 2026-093); Netac Technology "2026 Semi-Annual Report"; Shannon Chip Creation "2026 Semi-Annual Report"; TrendForce Consulting DRAM contract price quarter-on-quarter estimates (cited in "A Look Through 22 Semiconductor Report Cards," 21st Century Business Herald, October 23, 2026); National Bureau of Statistics September 2026 CPI data (cited in "September Storage Price Increases Narrow," National Business Daily, October 14, 2026).
先把九月的坐标定住。DRAM合约价环比涨幅，2026年一季度93%到98%、二季度约80%、三季度约40%、四季度预估20%，九月站在一条走完四分之三的斜坡末端。江波龙在业绩预告里的原话是"受前期较高价格基数影响，DRAM和NAND Flash价格涨幅较前期明显收窄，存储价格上涨对公司盈利的边际贡献显著趋缓"。江波龙用月末一次加权平均法结转成本，九月买的晶圆九月就进成本池；而它卖给客户的价格跟着季度合约走，三季度这一档在季度初就锁定了。两头一挤，九月对三季度报表只剩一个负方向。
德明利把三个数的关系摆得很清楚。三季度收入116.45亿元，环比增24.2%；毛利额38.30亿元，环比只增5.2%；毛利率32.89%，环比少5.95个百分点；净利27.29亿元，环比增2.2%。毛利率退、利润还在创新高，缺口被费用和产品结构抵掉了。所以"见到涨价的钱"和"见不到涨价的钱"两种说法都不对，准确的说法是：钱在，增量停了。三季度多做出来的22.73亿元收入只多带来1.90亿元毛利，每多做一块钱收入，多出八分钱毛利。
这个数比毛利率提前一个季度转向。二季度德明利收入环比多18.34亿元，毛利额反而少6.87亿元，每多一块钱收入是倒亏三毛七；同一季度江波龙每多一块钱收入多出七毛五毛利，佰维多六毛五，香农芯创多两毛一。谁先转负，谁就先见顶：德明利一季度毛利率57.42%见顶，江波龙二季度见顶，佰维到现在还没见顶。所以"传导时滞"根本不是一个统一的季度数，它是每家存货里"便宜料占多少"换算出来的区间——这个分层报表不披露，只能从每多做一块钱收入带来多少毛利里反推。拿存货天数去算拐点会算错，江波龙和德明利就是反例。
往上游看一层，就知道这条链上谁在截钱。东芯股份上半年毛利率67.66%，一季度53.17%；北京君正上半年51.75%，一季度43.49%，两家二季度都还在往上走。东芯在半年报里写"海外存储原厂加速退出利基型市场，供给端收缩态势明显……产品市场价格大幅攀升"；君正写的是另一件事——"受制于长协的约束，尽管消费市场已大幅涨价，公司的DRAM产品仍基本执行原有销售价格，2026年第一季度开始，国内外客户陆续开始进入调价周期"。原厂按季度合约定价，设计厂被长协锁住，模组厂按月加权，每一层有自己的重定价周期。涨价不是一条水流过去，是每一层各自重签一次合同。
供给这一端短期也不会松。江波龙在9月的调研记录里说，"受过往周期影响，海外原厂仍维持相对审慎的NAND资本开支计划，更多通过技术升级实现相对温和的NAND供给增长，因此存储产业供需关系仍将维持偏紧态势"。这句话对模组厂是双刃：售价有支撑，成本也降不下来。四季度不会是"涨价结束"，而是"涨价变成纯成本"——提价的空间被下游需求限住，买料的价格还在往上，中间那段加工费被两头挤薄。
这就解释了模组厂2026年一二季度那轮暴利从哪来：用的是2025年下半年买的料，卖的是2026年的价。德明利一季度、江波龙二季度的毛利率高点，都是库存账龄的产物，不是当期加工能力变强了。九月的涨价对它们意味着下一批料更贵，也就是负的。价格传到成本池是一个季度，成本传到毛利率再一个季度，但这两个季度是利润往回吐的两个季度，不是利润进来的两个季度。
现金那一栏更能说明这门生意。上半年经营现金流，江波龙-31.51亿元、佰维-62.61亿元、德明利-67.36亿元，三家合计流出161.48亿元；同一条链上收现金的是分销商，香农芯创+57.31亿元，朗科科技+0.11亿元。德明利三季度单季又流出69.55亿元，存货从年初70.58亿元加到289.86亿元，短期借款从34.72亿元加到101.50亿元，三季度单季财务费用2.92亿元、二季度1.46亿元，资产负债率70.10%。
反方最强的两条必须摆上来。第一条，德明利每多做一块钱收入只多八分钱毛利，可能是产品结构变了，不是价差没了：公司自己写"企业级存储业务收入规模及收入占比进一步提升"，企业级是低毛利、高周转的生意，占比抬升会把毛利率和增量毛利一起拉低。第二条，经营现金流为负在扩张期是正常的，289.86亿元存货对应的是"在手订单量快速增长"，是战略备货不是积压。第一条我接受一半：它能改变四季度毛利率的绝对水平，改不了方向，结构是渐进的，不会一个季度把毛利率从六十拉到四十几。第二条有硬伤——同一份三季报里，合同负债从6.24亿元降到4.05亿元，降35.06%，原因是"预收客户货款减少"。订单真在快速增长，预收款该往上走。备货的钱也是借的：短期借款增加66.78亿元，存货增加219.28亿元，利息已经在吃净利。
江波龙10月29日董事会审议后才披露三季报，方向跟德明利一致，只是起点更高、落差更大。佰维不能拿这两家的方向去套。它二季度每多做一块钱收入还多六毛五毛利，6月末存货181.68亿元，上半年经营现金流-62.61亿元；10月24日那份《关于公司近期生产经营情况的公告》通篇没有季度数字，只有行业趋势、战略和回购。它的三季度毛利率从公开文件里推不出来，谁现在给一个数都是猜。
已经定价了多少，看8月以来的走势。江波龙从8月17日419.26元跌到10月26日233.00元，回撤44.4%；佰维从256.78元到157.40元，回撤38.7%；德明利从437.33元到341.12元，回撤22.0%。德明利10月16日见289.77元后连涨四天到353.05元，涨21.8%，三季报公布当天跌3.38%。毛利率环比下滑这件事，9月到10月已经定价完了。没被定价的是现金流那一栏：-136.91亿元的经营现金流出和289.86亿元存货之间是什么关系，读毛利率的人多，读这一栏的人少。
证伪条件四条。德明利四季度每多做一块钱收入带来的毛利回到三毛以上，说明三季度的八分钱是企业级结构造成的，上面的归因要改。德明利四季度存货环比不再增加、单季经营现金流收窄到-20亿元以内，说明备货结束、利润开始变现，分叉窗口比预期短。江波龙10月29日三季报的存货余额如果低于二季度末，说明它也在收手，"四季度利润回来"那条路才成立。佰维三季报如果每多做一块钱收入还能多出五毛毛利，说明厚库存确实能把时间差拉长，那"模组厂分化是库存函数"这个判断成立，同涨同跌的框架就不适用。
来源：深圳市德明利技术股份有限公司《2026年第三季度报告》（2026年10月26日，公告编号2026-093）；深圳市江波龙电子股份有限公司《2026年前三季度业绩预告》（2026年10月15日，公告编号2026-117）；江波龙《2026年9月17日-21日投资者关系活动记录表》（2026年9月22日披露，记录编号2026-010）；江波龙H股《董事会会议召开日期》（2026年10月16日，股份代号09976）；江波龙《2026年半年度报告》《2026年第一季度报告》；深圳佰维存储科技股份有限公司《关于公司近期生产经营情况的公告》（2026年10月24日，公告编号2026-100）；佰维《2026年半年度报告》《2026年第一季度报告》；东芯半导体股份有限公司《2026年半年度报告》（2026年8月22日）《2026年第一季度报告》；北京君正集成电路股份有限公司《2026年半年度报告》（2026年8月29日，公告编号2026-067）《2026年第一季度报告》；香农芯创《2026年半年度报告》；朗科科技《2026年半年度报告》；TrendForce集邦咨询DRAM合约价季度环比预估（经《21世纪经济报道》2026年10月23日《透视22份半导体成绩单》引述）。