📰 The InformationThe InformationCodex · 实盘不做把握 5/10不做原因:证据不足
北京时间·🕒 发布 18:56:57时区推断·⟳ 抓取 18:57:04·⚡ 决策 09-04 20:06:59 +1h09m·52.9 天前
DeepSeek据报计划在内蒙古数据中心部署至少16万颗华为昇腾AI芯片
事件方向利好国产AI算力和华为昇腾生态，但直接受益方华为不可交易；可交易A股映射标的缺少本项目供货、金额和份额的一手证据，且消息称交付可能超过一年，短线正期望不足。
📄 原文The Information↗ 原文
DeepSeek Plans Major Huawei Chip Order in New AI Data Center
DeepSeek plans to install at least 160,000 Huawei AI chips at a data center in Inner Mongolia, Northern China, Bloomber…
不做 华为昇腾产业链映射篮子（N/A）
方向偏正面但直接受益方华为未上市；A股映射标的缺少本项目一手供货和金额证据，且交付周期可能超过一年，短线无足够edge。
不做 寒武纪（688256.SH）
DeepSeek采用华为芯片不是寒武纪订单，主题外溢不足以抵消华为竞争强化带来的不确定性。
不做 高新发展（000628.SZ）
虽有昇腾服务器生态联想，但未核实参与DeepSeek内蒙古集群供货，盘后明牌题材不适合无证据追高。
不做 拓维信息（002261.SZ）
昇腾生态伙伴逻辑存在，但缺少项目订单、份额和收入弹性证据，不能把主题名单当作交易催化。
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
这是一条新鲜且有影响力的Bloomberg/The Information转述报道，显示DeepSeek大规模转向华为昇腾推理集群，对国产AI芯片链方向偏正面。但订单未获公司官方确认，金额和交付节奏不明，关键瓶颈还在先进存储和华为产能；华为未上市，而寒武纪、高新发展、拓维信息、神州数码等可交易标的都缺少与该项目直接绑定的供货证据，因此不下可执行新仓。
消息传播链
- TNW称该消息来自Bloomberg，并指出DeepSeek未确认订单规模、芯片型号和交付计划。
- The Edge转发Bloomberg称DeepSeek拟部署至少160000颗华为顶级加速器，项目位于内蒙古数据中心。
- 本轮信号来源，明确新增至少160000颗华为芯片、主要用于推理、交付受华为产能限制。
- 7月已有DeepSeek内蒙古大型AI数据中心报道，说明数据中心本身不是今天的新信息。
关键理由
- 16万颗华为AI芯片若落地，将验证DeepSeek在推理端采用国产昇腾生态，强化国产算力需求预期。
- 华为官方此前披露Ascend 950DT面向推理解码和训练，且采用自有高带宽存储方案，说明技术路线与本事件方向一致。
- 现有可量化基数不足：订单金额、上市供应商份额和交付年度未披露，无法估算对候选公司收入/利润/市值的短期增量。
主要风险
- 报道基于知情人士，DeepSeek和华为未正式确认合同，若后续否认或规模低于报道，主题溢价会快速回吐。
- 先进存储和关键零部件短缺可能使完整交付超过一年，削弱短线订单兑现和A股供应链弹性。
反向证据
- 做多前查反向 · 来源 web · 日期 2026-09-04 · 影响 weakens
TNW称DeepSeek未确认订单规模、芯片型号、交付计划或合同签署，削弱订单交易强度。
检索:DeepSeek Huawei 160,000 chips unconfirmed shortage delivery more than a year 原文 ↗ - 做多前查反向 · 来源 web · 影响 weakens
可见报道指向HBM/先进存储短缺限制华为AI芯片产量，订单完整交付可能拉长。
检索:Huawei Ascend HBM shortage production low hundreds thousands 2026 原文 ↗ - 做多前查反向 · 来源 web · 日期 2026-07-07 · 影响 weakens
中文资料能找到昇腾产业链概念名单，但未坐实高新发展、拓维信息、神州数码等参与DeepSeek内蒙古项目。
检索:华为昇腾 产业链 A股 供应商 DeepSeek 项目 订单 原文 ↗
先例如何影响了这次决策
对本次决策的影响:先例不足使本次不把主题外溢升级为ENTER，候选仅保留为后续若出现一手供货证据再评估。
方向:没有找到足够高相似、可锚定到A股具体交易标的的同类先例；方向只能作为国产算力主题正面外溢参考。
涨幅量级:缺少同市场、同机制、同标的供应链确认的高相似先例，不能据此设定可交易tp/sl幅度。
形态:该类未确认媒体报道加产业链映射容易出现高开低走，尤其当直接受益方不可交易、二阶标的缺少订单坐实时。
工具执行结果 · 成功 2 · 降级 0 · 失败 0 · 取消 0
- web_search 成功 ? · 0 ms
- web_search 成功 ? · 0 ms
原文全文
展开原文 · 1,633 字
DeepSeek plans to install at least 160,000 Huawei AI chips at a data center in Inner Mongolia, Northern China, Bloomberg reported, citing people familiar with the situation.
The project would support China’s push to replace Nvidia silicon in light of U.S. chip restrictions. DeepSeek would mainly use the Huawei chips to run AI models after training, a process known as inference, according to Bloomberg. DeepSeek’s V4 model has already been optimised to run on Huawei hardware. The planned cluster would be comparable with large U.S. AI data centers by chip count. However, Huawei’s chips are less powerful than the Nvidia ones, which DeepSeek still uses for model training.
The chip purchase plan comes as DeepSeek seeks $7.4 billion in a second funding round, largely to finance research and computing infrastructure. The company spent 11 billion yuan on AI infrastructure in the first seven months of 2026, up from 1.2 billion yuan in all of 2025, [The Information reported](https://www.theinformation.com/articles/deepseeks-revenue-reaches-70-million-july-tenfold-jump-2025). Most of that spending went toward renting servers and buying chips and other equipment.
Securing enough Huawei chips, however, remains a challenge. Shortages of advanced memory and other key parts could limit Huawei’s production to the low hundreds of thousands of chips this year. DeepSeek may have to wait more than a year for its full order as Huawei also needs to supply other customers, according to Bloomberg.
However, component shortages could limit Huawei’s production output and extend the timeline for when DeepSeek can receive the chips.
The project would support China’s push to replace Nvidia silicon in light of U.S. chip restrictions. DeepSeek would mainly use the Huawei chips to run AI models after training, a process known as inference, according to Bloomberg. DeepSeek’s V4 model has already been optimised to run on Huawei hardware. The planned cluster would be comparable with large U.S. AI data centers by chip count. However, Huawei’s chips are less powerful than the Nvidia ones, which DeepSeek still uses for model training.
The chip purchase plan comes as DeepSeek seeks $7.4 billion in a second funding round, largely to finance research and computing infrastructure. The company spent 11 billion yuan on AI infrastructure in the first seven months of 2026, up from 1.2 billion yuan in all of 2025, [The Information reported](https://www.theinformation.com/articles/deepseeks-revenue-reaches-70-million-july-tenfold-jump-2025). Most of that spending went toward renting servers and buying chips and other equipment.
Securing enough Huawei chips, however, remains a challenge. Shortages of advanced memory and other key parts could limit Huawei’s production to the low hundreds of thousands of chips this year. DeepSeek may have to wait more than a year for its full order as Huawei also needs to supply other customers, according to Bloomberg.
However, component shortages could limit Huawei’s production output and extend the timeline for when DeepSeek can receive the chips.
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。