📰 DigiTimes 实时DigiTimes RealtimeDoubao · 实盘不做把握 4/10不做原因:消息已过时
北京时间·🕒 发布 09-10(台北日)仅日期·时区推断·⟳ 抓取 12:23:06·⚡ 决策 09-15 12:27:58 +4m52s·42.3 天前
DigiTimes综述：HBM紧缺推升华为昇腾、寒武纪等国产AI加速卡报价，但属5天前旧闻且寒武纪已否认涨价
信号为2026-09-10发表的行业综述，核心涨价数据引自同日Reuters，距当前(9/15)已5个自然日、3个交易日；点名的寒武纪当日即公开否认涨价，其余主角（华为/MetaX/Iluvatar）无可交易直接标的，HBM三强受益逻辑为存量共识、本文无新增订单或指引。属已扩散旧闻，无新增可交易edge。
📄 原文DigiTimes Realtime↗ 原文
中文網 繁體 简体 SUBSCRIBE My account Sign out SUBSCRIBE Trending Apple Huawei Nvidia AI Search Scaling AI Glasses Inference E…
不做 寒武纪（688256.SH）
本条为9月10日已扩散的HBM涨价综述（核心数据源自Reuters同日报道），距今约5天且公司当日否认涨价；近3个交易日股价横盘、无未定价edge，不追旧闻。
不做 HBM三强/国产算力链（SK海力士、三星、美光、华为系）（N/A）
HBM紧缺与长协锁单为多季度存量主线，本文无新增订单/指引；华为、MetaX、Iluvatar CoreX无直接可交易代码，构不成新的个股级交易。
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
结论摘要
HBM紧缺推动国产加速卡涨价的方向成立（供给受限+需求强），但这是5天前已由Reuters首发、DigiTimes整合的旧闻，唯一A股直接标的寒武纪已否认涨价且股价近几日走平，缺乏新鲜且尚未定价的个股级催化，故不做。
🔎 新鲜度:最早出现 2026-09-10 · 扩散 high · price-in 风险 high ↗
关键理由
- 涨价反映国产AI算力需求强劲且HBM成本传导顺畅，对有定价权的国产芯片商中期偏正面
- 该信息9月10日已公开并被寒武纪否认，近3个交易日寒武纪股价横盘（约1040元），未见重定价，edge已消化
主要风险
- 寒武纪否认仅针对'公告'，未排除私下报价上调，若后续订单/涨价坐实存在踏空风险
- 今日(9/15)寒武纪盘中+4.62%可能对应其他新催化，本判断仅针对本条5天前旧闻
反向证据
- 做多前查反向 · 来源 web · 日期 2026-09-10 · 影响 weakens
寒武纪证券部明确称未发布涨价公告、提示勿信传闻，直接削弱做多催化
检索:寒武纪 690 涨价 否认 公告 原文 ↗
先例如何影响了这次决策
对本次决策的影响:旧闻+当事方否认+股价已横盘，支持以STALE回避，不开仓
方向:无新鲜、对齐的个股级事件先例可测；本条为5天前行业综述而非新发订单/指引事件
涨幅量级:不足以校准幅度
形态:无
工具执行结果 · 成功 2 · 降级 0 · 失败 0 · 取消 0
- search_processed_events 成功 stock_tools HTTP · 100 ms
{"result": "search_processed_events:agent query=`china ai chip price hike hbm shortage cambricon huawei ascend | 华为昇腾 ascend 950 涨价 hbm 缺货 | 寒武纪 690 涨价 否认 hbm`；结构化索引覆盖完整 72 小时；旧记录兼容扫描近 0.0/72 小时、0 条本臂决策(完整)，返回 0 条候选。未命中只表示这些 agent query 表达式没有召回候选，不能据此证明消息首发或不是重复；按当前事件继续正常研判。"} - stock_quote 成功 stock_tools HTTP · 84 ms
{
原文全文
展开原文 · 8,184 字
中文網
繁體
简体
SUBSCRIBE
My account
Sign out
SUBSCRIBE
Trending Apple
Huawei
Nvidia
AI Search
Scaling AI Glasses
Inference Era AI
Advanced Packaging
Trending
SUBSCRIBE
CONNECT WITH US
Sign out
REALTIME NEWS
Corning sees CPO and glass core as new semiconductor battlegrounds
Tomorrow's Headlines
Sep 14, 18:47
China optical transceivers avoid US FCC curbs as NPO and CPO compete in high-speed links
Tomorrow's Headlines
Sep 14, 18:46
China's AIDC token costs hinge on access to advanced GPUs
Tomorrow's Headlines
Sep 14, 18:46
Home
Tech
Semiconductors
China's Nvidia rivals raise prices as HBM shortage bites
Sherri Wang
, DIGITIMES, Taipei
Sep 10, 2026, 14:37 , Updated: Sep 10, 2026, 17:18
0
Credit: AFP
Chinese chipmakers challenging Nvidia are sharply raising prices for current and next-generation accelerators as high-bandwidth memory (HBM) becomes scarcer and more expensive, adding another cost constraint to China's push for domestic computing hardware.
Huawei's Ascend 950DT now carries an indicated price above CNY250,000 (approx. US$37,255), up roughly 20% to 50% from quotes two months earlier, while indicated pricing for Cambricon Technologies' next-generation 690 has risen 20% to 30% over the same period. MetaX and Iluvatar CoreX have also raised prices. Reuters reported on September 10 that Huawei's Ascend 950PR and Ascend 910C have climbed to above CNY80,000 and CNY110,000, respectively, while Chinese chipmakers are increasingly relying on gray-market HBM that can cost several times more than supplies purchased outside China.
Cambricon, however, pushed back against reports of a price increase. A representative from the company's securities department told Chinese financial media on September 10 that Cambricon had not issued any price-hike announcement and cautioned investors against relying on market rumors. The response did not directly address whether pricing had changed in private customer discussions.
The increases point to a bottleneck beyond advanced logic manufacturing. SK Hynix, Samsung Electronics and Micron Technology dominate advanced HBM supply, while US export controls introduced in December 2024 restrict China's access to certain advanced HBM products.
With memory accounting for a significant share of an accelerator card's production cost, higher HBM prices are feeding through to finished hardware and raising the cost of deploying domestic alternatives to Nvidia.
Global memory supply leaves little slack
Outside China, the memory market offers limited room for additional demand.
Samsung expects demand for server DRAM, enterprise SSDs and HBM to accelerate in the second half of 2026 as infrastructure investment continues. Despite some moderation in smartphone and PC demand, the company expects the memory market to remain undersupplied, with supply constraints persisting despite efforts to increase production.
The imbalance could extend well beyond this year. Unmet demand carried over from 2026 is expected to deepen the memory shortage in 2027, with tight supply continuing into 2028. One constraint is the roughly three-and-a-half-year lead time between construction of a new fab and wafer production, limiting how quickly additional capacity can reach the market.
SK Hynix is seeing similar pressure. Additional supply requests continue to mount as major technology companies expand infrastructure investment. The memory maker has finalized long-term supply agreements with around 10 customers and is discussing additional multi-year contracts with major clients.
The company began mass shipments of HBM4 in the second quarter and plans to ramp production during the second half.
For Chinese chipmakers, that means trying to secure scarce HBM in a market where major global customers are increasingly using long-term agreements to lock in future supply. Restricted access to advanced HBM adds further pressure on both availability and price.
Huawei develops its own HBM approach
Huawei is seeking to address the memory constraint with technologies tailored to different workloads.
Its Ascend 950 series will use two proprietary HBM technologies. HiBL 1.0 will be paired with the Ascend 950PR, which targets the prefill stage of inference and recommendation systems, while HiZQ 2.0 is designed for the Ascend 950DT, targeting decode workloads and model training.
Huawei positions HiBL 1.0 as a lower-cost HBM option for workloads with relatively modest memory-bandwidth requirements. HiZQ 2.0, designed for more memory-intensive applications, provides 144GB of capacity and 4TB per second of memory bandwidth. The Ascend 950DT is scheduled for availability in the fourth quarter of 2026.
Huawei has not disclosed where or how the memory is manufactured, leaving unclear how far its proprietary approach can shield the company from broader HBM supply and pricing pressures.
The cost pressure is already visible across Huawei's product lineup. The Ascend 950PR, which sold for roughly CNY60,000 per card at the start of 2026, now fetches more than CNY80,000. The older Ascend 910C board has risen from around CNY90,000 to more than CNY110,000.
The increase is particularly notable because Huawei positioned HiBL 1.0 as a lower-cost memory solution. Higher prices for both the 950PR and the higher-performance 950DT suggest that proprietary memory technologies alone may not eliminate the broader supply and cost constraints surrounding HBM.
Higher cost for domestic compute
Demand for Chinese accelerators remains strong despite higher prices. Iluvatar CoreX has doubled planned GPU shipments to ByteDance to 100,000 units this year and diverted processors previously earmarked for internal use to help meet the internet company's demand.
US restrictions on Nvidia's most advanced products have created opportunities for domestic suppliers, but the HBM squeeze shows that designing a competitive processor is only part of the challenge.
As accelerator performance becomes increasingly dependent on high-bandwidth memory, advanced packaging and high-speed interconnects, China's ability to expand domestic computing capacity will also depend on securing enough memory at a viable cost. The latest price increases show HBM becoming another decisive test of how far China's Nvidia alternatives can scale.
Article
edited by Ysi Chen
AI Search
Related stories
Huawei's custom HBM challenges the existing memory hegemony
Huawei Ascend 950 to feature in-house HBM in 2026, advancing China's AI memory independence
Huawei charts 6-year Ascend AI chip roadmap, showcasing HBM advances and SuperNode scale
NVIDIA's Rubin CPX may reshape AI memory demand, boosting GDDR7 over HBM
HBM capacity crunch triggers price surge in DDR5 and VRAM markets
Show more
Categories
Chips + components Memory chips
Tags
Ascend China chipmakers cost HBM Nvidia production
Share this article
Other links
Print
Related topics
Global supply chain: China market
Asia
Regions
BIZ FOCUS
Sep 14, 08:00
Hauman Unveils Micro LED Testing Breakthrough for AI Optical Interconnects
Monday 14 September 2026
Scaling Photonics: Assembly, Test and Fiber Automation at SEMICON Taiwan
Friday 11 September 2026
DAS Expands in Taiwan and Unveils World First at SEMICON
Friday 11 September 2026
Toward Technologies Redefines Relay Solutions at SEMICON Taiwan 2026
MOST-READ
7 DAYS NEWS
Google's v10 TPU opens door to AMD without sidelining Broadcom, MediaTek
Update: ABF substrate maker Unimicron searched in origin-labeling probe
Qualcomm's HBC play pulls Samsung and SK hynix into a new race beyond HBM
SK keyfoundry expands 8-inch capacity as China demand strains supply
Samsung, SK Hynix step up China NAND upgrades as US tool risks grow
The 1,000-watt problem: a TSMC veteran on the brute force forging Nvidia's AI chips
Wuliangye skips carmaking, turns liquor waste into silicon-carbon anodes
eCloudvalley expands enterprise AI Agent portfolio
CXMT rides AI boom to 80% operating margin
Nvidia, MediaTek deepen 20-year chip ties
Full list
Members only
×
Sorry, the page you are trying to open is available only for our paid subscribers.
Please login to read more
New users, please register first
Notification
×
繁體
简体
SUBSCRIBE
My account
Sign out
SUBSCRIBE
Trending Apple
Huawei
Nvidia
AI Search
Scaling AI Glasses
Inference Era AI
Advanced Packaging
Trending
SUBSCRIBE
CONNECT WITH US
Sign out
REALTIME NEWS
Corning sees CPO and glass core as new semiconductor battlegrounds
Tomorrow's Headlines
Sep 14, 18:47
China optical transceivers avoid US FCC curbs as NPO and CPO compete in high-speed links
Tomorrow's Headlines
Sep 14, 18:46
China's AIDC token costs hinge on access to advanced GPUs
Tomorrow's Headlines
Sep 14, 18:46
Home
Tech
Semiconductors
China's Nvidia rivals raise prices as HBM shortage bites
Sherri Wang
, DIGITIMES, Taipei
Sep 10, 2026, 14:37 , Updated: Sep 10, 2026, 17:18
0
Credit: AFP
Chinese chipmakers challenging Nvidia are sharply raising prices for current and next-generation accelerators as high-bandwidth memory (HBM) becomes scarcer and more expensive, adding another cost constraint to China's push for domestic computing hardware.
Huawei's Ascend 950DT now carries an indicated price above CNY250,000 (approx. US$37,255), up roughly 20% to 50% from quotes two months earlier, while indicated pricing for Cambricon Technologies' next-generation 690 has risen 20% to 30% over the same period. MetaX and Iluvatar CoreX have also raised prices. Reuters reported on September 10 that Huawei's Ascend 950PR and Ascend 910C have climbed to above CNY80,000 and CNY110,000, respectively, while Chinese chipmakers are increasingly relying on gray-market HBM that can cost several times more than supplies purchased outside China.
Cambricon, however, pushed back against reports of a price increase. A representative from the company's securities department told Chinese financial media on September 10 that Cambricon had not issued any price-hike announcement and cautioned investors against relying on market rumors. The response did not directly address whether pricing had changed in private customer discussions.
The increases point to a bottleneck beyond advanced logic manufacturing. SK Hynix, Samsung Electronics and Micron Technology dominate advanced HBM supply, while US export controls introduced in December 2024 restrict China's access to certain advanced HBM products.
With memory accounting for a significant share of an accelerator card's production cost, higher HBM prices are feeding through to finished hardware and raising the cost of deploying domestic alternatives to Nvidia.
Global memory supply leaves little slack
Outside China, the memory market offers limited room for additional demand.
Samsung expects demand for server DRAM, enterprise SSDs and HBM to accelerate in the second half of 2026 as infrastructure investment continues. Despite some moderation in smartphone and PC demand, the company expects the memory market to remain undersupplied, with supply constraints persisting despite efforts to increase production.
The imbalance could extend well beyond this year. Unmet demand carried over from 2026 is expected to deepen the memory shortage in 2027, with tight supply continuing into 2028. One constraint is the roughly three-and-a-half-year lead time between construction of a new fab and wafer production, limiting how quickly additional capacity can reach the market.
SK Hynix is seeing similar pressure. Additional supply requests continue to mount as major technology companies expand infrastructure investment. The memory maker has finalized long-term supply agreements with around 10 customers and is discussing additional multi-year contracts with major clients.
The company began mass shipments of HBM4 in the second quarter and plans to ramp production during the second half.
For Chinese chipmakers, that means trying to secure scarce HBM in a market where major global customers are increasingly using long-term agreements to lock in future supply. Restricted access to advanced HBM adds further pressure on both availability and price.
Huawei develops its own HBM approach
Huawei is seeking to address the memory constraint with technologies tailored to different workloads.
Its Ascend 950 series will use two proprietary HBM technologies. HiBL 1.0 will be paired with the Ascend 950PR, which targets the prefill stage of inference and recommendation systems, while HiZQ 2.0 is designed for the Ascend 950DT, targeting decode workloads and model training.
Huawei positions HiBL 1.0 as a lower-cost HBM option for workloads with relatively modest memory-bandwidth requirements. HiZQ 2.0, designed for more memory-intensive applications, provides 144GB of capacity and 4TB per second of memory bandwidth. The Ascend 950DT is scheduled for availability in the fourth quarter of 2026.
Huawei has not disclosed where or how the memory is manufactured, leaving unclear how far its proprietary approach can shield the company from broader HBM supply and pricing pressures.
The cost pressure is already visible across Huawei's product lineup. The Ascend 950PR, which sold for roughly CNY60,000 per card at the start of 2026, now fetches more than CNY80,000. The older Ascend 910C board has risen from around CNY90,000 to more than CNY110,000.
The increase is particularly notable because Huawei positioned HiBL 1.0 as a lower-cost memory solution. Higher prices for both the 950PR and the higher-performance 950DT suggest that proprietary memory technologies alone may not eliminate the broader supply and cost constraints surrounding HBM.
Higher cost for domestic compute
Demand for Chinese accelerators remains strong despite higher prices. Iluvatar CoreX has doubled planned GPU shipments to ByteDance to 100,000 units this year and diverted processors previously earmarked for internal use to help meet the internet company's demand.
US restrictions on Nvidia's most advanced products have created opportunities for domestic suppliers, but the HBM squeeze shows that designing a competitive processor is only part of the challenge.
As accelerator performance becomes increasingly dependent on high-bandwidth memory, advanced packaging and high-speed interconnects, China's ability to expand domestic computing capacity will also depend on securing enough memory at a viable cost. The latest price increases show HBM becoming another decisive test of how far China's Nvidia alternatives can scale.
Article
edited by Ysi Chen
AI Search
Related stories
Huawei's custom HBM challenges the existing memory hegemony
Huawei Ascend 950 to feature in-house HBM in 2026, advancing China's AI memory independence
Huawei charts 6-year Ascend AI chip roadmap, showcasing HBM advances and SuperNode scale
NVIDIA's Rubin CPX may reshape AI memory demand, boosting GDDR7 over HBM
HBM capacity crunch triggers price surge in DDR5 and VRAM markets
Show more
Categories
Chips + components Memory chips
Tags
Ascend China chipmakers cost HBM Nvidia production
Share this article
Other links
Related topics
Global supply chain: China market
Asia
Regions
BIZ FOCUS
Sep 14, 08:00
Hauman Unveils Micro LED Testing Breakthrough for AI Optical Interconnects
Monday 14 September 2026
Scaling Photonics: Assembly, Test and Fiber Automation at SEMICON Taiwan
Friday 11 September 2026
DAS Expands in Taiwan and Unveils World First at SEMICON
Friday 11 September 2026
Toward Technologies Redefines Relay Solutions at SEMICON Taiwan 2026
MOST-READ
7 DAYS NEWS
Google's v10 TPU opens door to AMD without sidelining Broadcom, MediaTek
Update: ABF substrate maker Unimicron searched in origin-labeling probe
Qualcomm's HBC play pulls Samsung and SK hynix into a new race beyond HBM
SK keyfoundry expands 8-inch capacity as China demand strains supply
Samsung, SK Hynix step up China NAND upgrades as US tool risks grow
The 1,000-watt problem: a TSMC veteran on the brute force forging Nvidia's AI chips
Wuliangye skips carmaking, turns liquor waste into silicon-carbon anodes
eCloudvalley expands enterprise AI Agent portfolio
CXMT rides AI boom to 80% operating margin
Nvidia, MediaTek deepen 20-year chip ties
Full list
Members only
×
Sorry, the page you are trying to open is available only for our paid subscribers.
Please login to read more
New users, please register first
Notification
×
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。