📰 DigiTimesDigiTimesDoubao · 实盘不做把握 7/10不做原因:证据不足
北京时间·🕒 发布 09-10 12:15:36·⟳ 抓取 09-10 12:46:24 +30m48s·⚡ 决策 09-15 13:01:11 +5d00h·42.2 天前
京东云拟以摩尔线程GPU建设十万卡国产智算集群，但消息已扩散数日且缺少合同金额与交付节奏
事件对摩尔线程具方向性利好，但当前公开口径是京东云拟建设/规划十万卡集群，并非披露金额、付款和交付节奏的正式重大合同；消息已传播约5天，摩尔线程在消息当日及随后两个交易日继续下跌，说明市场并未给予正向重估，当前没有可执行的短线正edge。
📄 原文DigiTimes↗ 原文
JD Cloud plans 100,000-GPU cluster built on Moore Threads chips
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 摩尔线程（688795.SH）
十万卡项目目前是规划性合作而非金额、付款和交付节奏明确的正式合同；消息9月9日已明牌传播，随后股价连续下跌且公司现金流和存货压力较大，当前没有未被定价的短线正edge。
不做 京东集团（09618.HK）
项目对京东云战略有正面意义，但对京东集团整体市值和短期业绩的增量不清晰，且资本开支回报存在不确定性。
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
京东云选用摩尔线程GPU规划十万卡集群，验证国产训练GPU进入头部云厂商大规模集群；但无合同金额、无交付约束且消息已被广泛报道，直接标的公告后持续走弱，因此不追多。
消息传播链
- 较早报道京东云总裁曹鹏在JDD大会公布与摩尔线程等伙伴规划十万卡国产算力集群。
- 报道京东云拟建设摩尔线程十万卡GPU智算集群，确认会议信息已在9月9日下午广泛传播。
- DigiTimes补充摩尔线程财务、库存、现金流及国产GPU竞争分层信息，属于后续英文产业媒体整理而非首发。
- 2026-09-10 15:00 CST市场反应行情数据非一手摩尔线程9月10日收跌5.97%，9月11日再跌1.70%，9月14日再跌3.31%，利好没有带来短线正收益。市场反应:消息后首个完整交易日至9月14日，688795.SH连续走弱并创阶段新低。
关键理由
- 十万卡级集群若全部转化为摩尔线程订单，相对其2026年上半年17.36亿元收入具备显著潜在弹性。
- 头部云厂商采用国产训练GPU具备标杆意义，但现阶段仍停留在规划和合作披露，尚不能按确定订单建模。
主要风险
- 项目可能分期建设或调整规模，且未披露合同金额、验收和付款条件，实际收入确认可能远低于十万卡标题量级。
- 摩尔线程经营现金流为负、存货高企且股价处于明确下行趋势，利好披露后未能反弹显示市场更关注兑现与估值风险。
反向证据
- 做多前查反向 · 来源 web · 日期 2026-09-11 · 影响 weakens
消息广泛传播后的9月10日，摩尔线程仍下跌5.97%，主力资金净流出约3亿元，未出现利好驱动的正反馈。
检索:688795 摩尔线程 9月10日 十万卡 股价 主力资金 净流出 原文 ↗ - 做多前查反向 · 来源 web · 日期 2026-09-10 · 影响 weakens
DigiTimes转述分析称，公司剔除政府补助和投资收益后仍亏损1.51亿元，经营现金流-21.69亿元，存货从2025年底的13.32亿元增至35.50亿元。
检索:摩尔线程 2026 上半年 经营现金流 存货 政府补助 亏损 原文 ↗ - 做多前查反向 · 来源 web · 日期 2026-09-15 · 影响 weakens
公开报道主要表述为拟建设、规划建设或订单口径，但未检索到披露合同金额和分批交付义务的上市公司正式公告。
检索:摩尔线程 十万卡 正式合同 金额 交付 公告 原文 ↗
先例如何影响了这次决策
对本次决策的影响:因缺少干净对齐先例且直接价格反应为负，结论从“主题利好”下调为不交易，而不是设定追涨仓位。
方向:未获得足够同市场、同相位且能锚定官方首发的国产GPU十万卡集群订单先例，不能据此外推涨幅。
涨幅量级:检索到摩尔线程3月签署6.6亿元夸娥智算集群合同等较低相似案例，但公开搜索结果多为门户转载且金额、客户和事件相位与本次十万卡规划差异较大，未用于设定止盈止损。
形态:本次消息后直接标的连续下跌，比泛化先例更能说明当前市场对该规划口径的承接较弱。
工具执行结果 · 成功 19 · 降级 0 · 失败 0 · 取消 0
- search_processed_events 成功 stock_tools HTTP · 16 ms
{"result": "search_processed_events:agent query=`jd cloud moore threads 100000 gpu cluster joyai | 京东云 摩尔线程 10万张 国产gpu 集群 | 摩尔线程 京东云 十万卡 全功能gpu 订单`；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。"} - direct_instrument_candidates 成功 stock_tools HTTP · 20 ms
{ - web_search 成功 doubao_search · 1,464 ms
豆包搜索结果（原始查询:摩尔线程 股票代码 上市 京东云 10万卡 集群 2026 9月10日；API SearchType:web） - kb_search 成功 stock_tools HTTP · 2,229 ms
{ - query_modeling_kb 成功 stock_tools HTTP · 46 ms
{"result": "modeling_kb 命中 10 条 (库快照 2026-07-27 · AI 算力链 19 页 · 本次命中证据最新 2026-05-21 · 其中 ⟨预测⟩ 7 条) —— 这是**定期重生成的快照**,不是活数据;比快照更晚的事实以你手上的信号/kb_search 为准。\n\n• [A] — Blackwell 家族小计（GPU 颗） · 表1：分型号 GPU 出货/产出量（百万颗 GPU，日历年） · CY2025 = ~6.0 百万颗 ⚠️源间数字有冲突[1][2][3][4][5][6][7][8][9][10][11]\n 依据: 直接取黄仁勋2 - stock_quote 成功 stock_tools HTTP · 76 ms
{ - stock_quote 成功 stock_tools HTTP · 40 ms
{ - web_search 成功 doubao_search · 1,525 ms
豆包搜索结果（原始查询:京东云 摩尔线程 十万卡 国产智算集群 官方 9月9日 曹鹏；API SearchType:web） - web_search 成功 doubao_search · 1,404 ms
豆包搜索结果（原始查询:摩尔线程 产业链 供应商 服务器 PCB 封装 HBM 概念股 688795；API SearchType:web） - web_search 成功 doubao_search · 3,484 ms
豆包搜索结果（原始查询:688795 摩尔线程 9月10日 9月11日 十万卡 股价 反应；API SearchType:web） - web_search 成功 doubao_search · 1,505 ms
豆包搜索结果（原始查询:摩尔线程 招股书 前五大供应商 晶圆代工 封装测试 PCB 服务器 ODM；API SearchType:web） - web_search 成功 doubao_search · 1,432 ms
豆包搜索结果（原始查询:摩尔线程 十万卡 集群 服务器 产业链 受益股 工业富联 浪潮信息 中科曙光；API SearchType:web） - get_stock_profile 成功 stock_tools HTTP · 41 ms
{"result": "🏷️ **证券档案**\n\n### 688795\n- ⚠️ 无法解析为股票代码。请换公司全名或直接给代码（如 300750.SZ / NVDA）。\n\n### 688256\n- ⚠️ 无法解析为股票代码。请换公司全名或直接给代码（如 300750.SZ / NVDA）。\n\n### 9618.HK\n- ⚠️ 无法解析为股票代码。请换公司全名或直接给代码（如 300750.SZ / NVDA）。\n\n> 数据来源: 代码/名称/上市状态=内部股票池 stock_universe.csv · A股行业=朝阳永续(申万/中信/中金/GICS) · 港股行业=通联 - search_case_library 成功 stock_tools HTTP · 31 ms
{"result": "search_case_library「688795.SH,9618.HK」—— 与这些标的相关的自家决策案例(本页 10 条,该标的全历史相关记录共 78 条；ENTER 只按该标的自己的腿统计):\n◆ 688795.SH：相关历史决策 45 条 · 其中该标的 ENTER 2 条(4.4%) · 已结算 2 笔 · 胜率 0% · 中位 -3.99% · 均值 -3.99% ⚠️只有 2 笔,样本太薄,只能当线索不能当规律\n◆ 9618.HK：相关历史决策 33 条 · 其中该标的 ENTER 0 条(0.0%) · 我们从没在这只票上 ENTER 过\n↑ 胜 - web_search 成功 doubao_search · 1,448 ms
豆包搜索结果（原始查询:国产GPU 十万卡 集群 订单 寒武纪 摩尔线程 股价 公告 2026；API SearchType:web） - web_search 成功 doubao_search · 1,321 ms
豆包搜索结果（原始查询:摩尔线程 66亿 合同 订单 2026 3月 公告 股价 次日；API SearchType:web） - web_search 成功 doubao_search · 1,594 ms
豆包搜索结果（原始查询:寒武纪 大规模 订单 智算集群 合同 公告 2026 股价；API SearchType:web） - kb_search 成功 stock_tools HTTP · 2,278 ms
{ - web_search 成功 doubao_search · 1,181 ms
豆包搜索结果（原始查询:site:sse.com.cn 688795 关于签订日常经营重大合同的公告 2026-009 6.6亿元；API SearchType:web）
原文全文
展开原文 · 5,921 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
ICT
JD Cloud plans 100,000-GPU cluster built on Moore Threads chips
Jingyue Hsiao
, DIGITIMES, Taipei
Sep 10, 2026, 12:15
0
Credit: AFP
JD Cloud said on September 9 that it plans to build a 100,000-card cluster of full-function GPUs supplied by Moore Threads, a deployment the company describes as the first time domestic GPUs have gone into a 100,000-card core computing cluster at a leading Chinese AI cloud provider. The plan was announced at JD's 2026 Global Technology Explorers Conference in Beijing, according to Yicai .
A domestic stack for physical AI
Cao Peng, chairman of JD Group's technology committee and president of JD Cloud, said the company is pushing a "physical AI" strategy in which high-performance, reliable domestic computing is the core support, and that JD chose Moore Threads for "its scarce capability in domestic training-grade GPUs and its mature delivery experience with large-scale clusters." The two sides will work across chips, cloud platform, and model training to support JD's JoyAI model family, closing a loop from data through training and simulation to deployment.
The cluster is a step up from what is already running: JD Cloud has built a 10,000-card domestic cluster with Moore Threads and other partners, and the new capacity will be opened to outside users for large-model training, inference, and embodied AI, TechNode reported. Cao also said JD Cloud's campaign to collect 10 million hours of human activity data is proceeding, feeding the embodied models it is training.
Policy is pulling in the same direction. China's Ministry of Industry and Information Technology recently issued its telecommunications industry plan for the 15th Five-Year Plan period, calling for orderly deployment of 10,000-card and 100,000-card-plus intelligent computing clusters and stronger adaptation to domestic chips.
Validation for a supplier still short of profit
Zhang Jianzhong, Moore Threads' founder, chairman, and chief executive, called the order "the strongest validation of our technical strength and engineering delivery capability, and a key milestone for domestic intelligent computing moving to commercial scale."
It also lands on a company whose finances remain early-stage. Moore Threads posted first-half 2026 revenue of CNY1.736 billion (US$258.01 million), up 147.42% year on year, with net loss narrowing 95.73% to CNY11.56 million. Stripping out government grants and investment gains, it still lost CNY151 million, and operating cash flow was negative CNY2.169 billion, according to an analysis published by Huxiu . Cloud computing products accounted for 97.5% of revenue, and inventory rose to CNY3.550 billion from CNY1.332 billion at the end of 2025.
That profile contrasts with Cambricon, which reported first-half revenue of CNY5.996 billion and net profit of CNY2.311 billion. A Huatai Securities note cited in the same analysis argued the divergence marks the end of industry-wide losses in domestic GPUs and the start of a tiering phase in which leading firms convert orders into results. Brokerages cited by Huxiu also warn the window created by US export controls on Nvidia's high-end training chips may run only another 12 to 18 months.
Article
edited by Jerry Chen
AI Search
Related stories
Moore Threads builds China-made embodied AI stack with Lightwheel.ai
Moore Threads hits first profit, AI compute orders drive growth
China's GPU Four race to power large models, eye inference surge
Moore Threads takes on CUDA with MUSA as China's GPU race shifts to developer ecosystems
Commentary: Moore Threads' record IPO echoes Cambricon's trajectory as investors bet on unproven profitability
Show more
Categories
AI Chips + components IC design, distribution IT + CE Server, IPC, cloud computing, IoT
Tags
2026 chairman China chips Cloud data Moore Threads revenue training
Share this article
Other links
Print
Related topics
Global supply chain: Key components
Global supply chain: Mobile, telecom, computing
Global supply chain: Emerging tech, businesses
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
Column: When digital trust breaks — quantum risk rewrites supply chain resilience
Wuliangye skips carmaking, turns liquor waste into silicon-carbon anodes
The 1,000-watt problem: a TSMC veteran on the brute force forging Nvidia's AI chips
eCloudvalley expands enterprise AI Agent portfolio
CXMT rides AI boom to 80% operating margin
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
ICT
JD Cloud plans 100,000-GPU cluster built on Moore Threads chips
Jingyue Hsiao
, DIGITIMES, Taipei
Sep 10, 2026, 12:15
0
Credit: AFP
JD Cloud said on September 9 that it plans to build a 100,000-card cluster of full-function GPUs supplied by Moore Threads, a deployment the company describes as the first time domestic GPUs have gone into a 100,000-card core computing cluster at a leading Chinese AI cloud provider. The plan was announced at JD's 2026 Global Technology Explorers Conference in Beijing, according to Yicai .
A domestic stack for physical AI
Cao Peng, chairman of JD Group's technology committee and president of JD Cloud, said the company is pushing a "physical AI" strategy in which high-performance, reliable domestic computing is the core support, and that JD chose Moore Threads for "its scarce capability in domestic training-grade GPUs and its mature delivery experience with large-scale clusters." The two sides will work across chips, cloud platform, and model training to support JD's JoyAI model family, closing a loop from data through training and simulation to deployment.
The cluster is a step up from what is already running: JD Cloud has built a 10,000-card domestic cluster with Moore Threads and other partners, and the new capacity will be opened to outside users for large-model training, inference, and embodied AI, TechNode reported. Cao also said JD Cloud's campaign to collect 10 million hours of human activity data is proceeding, feeding the embodied models it is training.
Policy is pulling in the same direction. China's Ministry of Industry and Information Technology recently issued its telecommunications industry plan for the 15th Five-Year Plan period, calling for orderly deployment of 10,000-card and 100,000-card-plus intelligent computing clusters and stronger adaptation to domestic chips.
Validation for a supplier still short of profit
Zhang Jianzhong, Moore Threads' founder, chairman, and chief executive, called the order "the strongest validation of our technical strength and engineering delivery capability, and a key milestone for domestic intelligent computing moving to commercial scale."
It also lands on a company whose finances remain early-stage. Moore Threads posted first-half 2026 revenue of CNY1.736 billion (US$258.01 million), up 147.42% year on year, with net loss narrowing 95.73% to CNY11.56 million. Stripping out government grants and investment gains, it still lost CNY151 million, and operating cash flow was negative CNY2.169 billion, according to an analysis published by Huxiu . Cloud computing products accounted for 97.5% of revenue, and inventory rose to CNY3.550 billion from CNY1.332 billion at the end of 2025.
That profile contrasts with Cambricon, which reported first-half revenue of CNY5.996 billion and net profit of CNY2.311 billion. A Huatai Securities note cited in the same analysis argued the divergence marks the end of industry-wide losses in domestic GPUs and the start of a tiering phase in which leading firms convert orders into results. Brokerages cited by Huxiu also warn the window created by US export controls on Nvidia's high-end training chips may run only another 12 to 18 months.
Article
edited by Jerry Chen
AI Search
Related stories
Moore Threads builds China-made embodied AI stack with Lightwheel.ai
Moore Threads hits first profit, AI compute orders drive growth
China's GPU Four race to power large models, eye inference surge
Moore Threads takes on CUDA with MUSA as China's GPU race shifts to developer ecosystems
Commentary: Moore Threads' record IPO echoes Cambricon's trajectory as investors bet on unproven profitability
Show more
Categories
AI Chips + components IC design, distribution IT + CE Server, IPC, cloud computing, IoT
Tags
2026 chairman China chips Cloud data Moore Threads revenue training
Share this article
Other links
Related topics
Global supply chain: Key components
Global supply chain: Mobile, telecom, computing
Global supply chain: Emerging tech, businesses
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
Column: When digital trust breaks — quantum risk rewrites supply chain resilience
Wuliangye skips carmaking, turns liquor waste into silicon-carbon anodes
The 1,000-watt problem: a TSMC veteran on the brute force forging Nvidia's AI chips
eCloudvalley expands enterprise AI Agent portfolio
CXMT rides AI boom to 80% operating margin
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。