📰 BloombergBloombergQwen · 实盘不做把握 8/10
北京时间·🕒 发布 11:32:20时区推断·⟳ 抓取 11:42:53·⚡ 决策 08-07 11:53:13 +10m20s·81.3 天前
彭博发表综述性报道:北京推动国产替代下中国AI芯片设计商有望迎来销售丰收,但全文均为既有旧闻、无新增催化
这篇彭博报道是行业回顾综述,无新增离散事实:BI调查为6月、摩根士丹利70%自给率测算为3月旧报告、2万亿元数据中心计划为6月旧报道、华为910C年产60万颗为去年报道、摩尔线程H1营收+149%指引为上月已披露
📄 原文Bloomberg↗ 原文
China’s AI chip designers are expecting bumper sales this earnings season off the back of Beijing’s push to get local f…
不做 寒武纪(688256.SH)
彭博稿为零新增信息的行业综述,盘面亦未增量定价;真正催化是明日(8/8)半年报,但A股业绩追多硬闸要求归母净利同比>10倍且未被price-in——2025H1基数10.38亿元下H1净利10倍以上几乎不可能,且本篇报道本身就是财报预期充分化的证据,财报前基于综述买入不予允许;待财报落地后由新信号另行研判
不做 海光信息(688041.SH)
仅在6月BI调查中被提及'被大量受访企业评估',无任何新增订单/指引/政策;国产算力为最拥挤共识、市值约6765亿元,软叙事无市值级增量,今晨盘面+0.7%亦证伪增量催化
不做 摩尔线程(688795.SH)
报道引用的'H1营收最高+149%'为公司上月已披露指引,属旧闻复述、已price-in;本稿无新增事实,不构成做多催化
不做 天数智芯(09903.HK)
'H1销售增至三倍'是外部预期而非公司公告,中报尚未披露、财报前猜测不予允许;该股盘中已+6.8%、20日日波动率约9.9%、近月单日±7~14%为常态,基于综述稿追高波动次新股风险收益比极差
不做 沐曦股份(688802.SH)
文章仅提及其上月WAIC展示万卡机架(旧事),无任何新增离散催化;2025年12月上市的次新高波动标的,纯主题联动且无增量信息,不做
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
🔎 新鲜度:最早出现 2026-08-07 · 扩散 high · price-in 风险 high ↗
消息传播链
- 彭博综述稿于北京时间11:32(A股午休前、港股上午盘中)发布,通篇引用既有调查与旧报道,未披露任何新订单、新政策或新指引;发布后A股国产算力龙头上午收盘微跌/微涨、无增量反应市场反应:发布后寒武纪上午收-0.9%、海光+0.7%、摩尔线程+1.3%;港股天数智芯+6.8%但该股近20日单日波动常态±7~14%,无法确证由本稿驱动
关键理由
- 这篇彭博报道是行业回顾综述,无新增离散事实:BI调查为6月、摩根士丹利70%自给率测算为3月旧报告、2万亿元数据中心计划为6月旧报道、华为910C年产60万颗为去年报道、摩尔线程H1营收+149%指引为上月已披露
- 唯一在前的可交易事件是寒武纪(688256.SH)明日(8/8)半年报,但A股业绩追多需满足净利同比>10倍+未price-in双硬闸:2025H1归母10.38亿元为基数、2026H1净利10倍以上基本不可能,且本篇报道本身就是财报预期充分化的体现
- 国产算力替代主题已被市场交易多轮、极度拥挤:寒武纪(688256.SH)市值约7339亿元、海光信息(688041.SH)约6765亿元,今晨A股盘面寒武纪-0.9%、海光+0.7%,未对该报道做任何增量定价
主要风险
- 若误将综述当新催化追入,买在主题炒作的高波动区(寒武纪/天数智芯20日日波动率5.8%/9.9%,动辄单日±10%以上),财报兑现或传闻扰动下回撤风险极大
- 寒武纪2026年2月曾因营收指引'小作文'单日跌超9%,高位拥挤持仓对任何负面扰动高度敏感
反向证据
- 做多前查反向 · 来源 web · 日期 2026-02-03 · 影响 weakens
2026年2月寒武纪曾因'2026营收200亿小作文'盘中跌超14%、公司紧急澄清不实——该股对传闻与预期差极度敏感,综述稿追多的风险收益比差
检索:寒武纪 半年报 业绩预告 8月8日 2026 营收 预期 原文 ↗ - 做多前查反向 · 来源 kb_search · 日期 2026-07-22 · 影响 weakens
内部库近两周充满国产算力超级周期路演/纪要(国联民生'三支箭'、'算力即国力'等),该主题为卖方与资金最拥挤共识,本报道无任何超出已有库内容的新增信息
检索:国产AI芯片 自主可控 寒武纪 海光 订单 北京 要求 国产替代(kb_search 7月20日以来) - 做多前查反向 · 来源 stock_quote · 日期 2026-08-07 · 影响 invalidates
报道发布后首个可交易时段(A股上午盘)寒武纪-0.88%、海光+0.67%、摩尔线程+1.34%,均未出现增量定价;市场未把本稿当新催化
检索:A股国产算力龙头今日盘面对本报道的增量反应
工具调用 · 历史记录,结果未知(5 种)
- search_processed_events · 结果未知
- web_search · 结果未知
- normalize_ticker · 结果未知
- kb_search · 结果未知
- stock_quote · 结果未知
原文全文
展开原文 · 4,497 字
China’s AI chip designers are expecting bumper sales this earnings season off the back of Beijing’s push to get local firms to use homegrown components and reduce their reliance on American tech.
Cambricon Technologies Corp. is projected to post surging revenue for the first half of the year with up-and-coming Shanghai Iluvatar CoreX Semiconductor Co. expected to see sales triple over the same period. Beijing-based Moore Threads Technology Co. said last month it expected revenue for the first six months of 2026 to grow as much as 149%.
Chinese firms are ramping up orders after Beijing ordered local tech firms to find alternatives to Nvidia Corp.’s cutting-edge products. Washington has for years restricted China’s access to advanced chips, but it recently cleared sales of Nvidia’s H200, the most powerful artificial intelligence processors ever allowed by the US to be sold in the country. However, US officials said last month that the actual number of Nvidia processors shipped to China since then has been “trivial.”
Privately-held Huawei Technologies Co. and Alibaba Group Holding Ltd.’s chip unit T-Head are also expected to boost their domestic market shares at a time when Nvidia and US chipmaker Advanced Micro Devices Inc. are virtually locked out of the market. Huawei is sharply ramping up production of its most advanced artificial intelligence chips in 2026, aiming to produce about 600,000 of its marquee 910C Ascend chips, Bloomberg News reported last year.
As a result, Chinese companies expect to spend 46% of their AI accelerator budget in the next 12 months to acquire locally-made chips, compared to 30% now, according to a survey of Chinese technology executives by Bloomberg Intelligence in June. AI accelerators produced by Cambricon and Hygon Information Technology Co. were being evaluated by a large pool of respondents to the survey.
In the longer term, Chinese chipmakers are set to benefit from Beijing’s ambitious data center buildout that aims to spend 2 trillion yuan ($295 billion) over the next five years. The massive project will rely on local suppliers including Huawei for at least 80% of technology such as AI chips. That would push China’s self-sufficiency level in AI chips to 70% at the end of this decade, up from 42% in 2025, according to estimates from Morgan Stanley.
Read More: China Preps $295 Billion Plan to Fund Nationwide AI Buildout
Nvidia boss Jensen Huang has consistently advocated for the use of his firm’s chips in China to ensure America’s lead in cutting-edge technology. His arguments alerted Beijing to the prospect of over-relying on foreign technology and creating vulnerabilities in the nation’s quest for global AI leadership, according to Morningstar analyst Phelix Lee.
“Most people would privately agree that having access to Nvidia would always be ideal, but in the absence of that, the trend I’m seeing is more Chinese chips being used,” Lee said.
China-made chips, when it comes to single-unit performance, are still years behind those built by Nvidia. It’s partly the result of a yearslong US-led campaign that restricts the most advanced chipmaking machines including the state-of-the-art extreme ultraviolet lithography systems made by Dutch company ASML Holding NV from flowing to China, the kind of gear that industry leader Taiwan Semiconductor Manufacturing Co. uses to make the world’s most powerful AI accelerators.
As customers start to shift from using chips for training AI models to integrating them into their businesses, Chinese chipmakers are winning orders from their ability to generate tokens at competitive pricing, Morgan Stanley analysts including Charlie Chan said in a research note at the end of July.
“Purchasing decisions are increasingly driven by deployment economics rather than absolute peak silicon performance,” Chan wrote.
Chinese chipmakers also have concluded that they can compensate for a lack of computing power in a single chip with scale. At China’s premier AI summit held in Shanghai last month, Chinese chipmakers from Huawei to Moore Threads and MetaX Integrated Circuits Shanghai Co. showcased their server racks capable of integrating tens of thousands chips.
“What they are trying to reassure is that we have enough chips and we have the infrastructure to link all these chips together so we can outnumber you,” Morningstar’s Lee said. “It also demonstrates how China, given enough time, will eventually narrow the gap to much narrower levels than today.”
Cambricon Technologies Corp. is projected to post surging revenue for the first half of the year with up-and-coming Shanghai Iluvatar CoreX Semiconductor Co. expected to see sales triple over the same period. Beijing-based Moore Threads Technology Co. said last month it expected revenue for the first six months of 2026 to grow as much as 149%.
Chinese firms are ramping up orders after Beijing ordered local tech firms to find alternatives to Nvidia Corp.’s cutting-edge products. Washington has for years restricted China’s access to advanced chips, but it recently cleared sales of Nvidia’s H200, the most powerful artificial intelligence processors ever allowed by the US to be sold in the country. However, US officials said last month that the actual number of Nvidia processors shipped to China since then has been “trivial.”
Privately-held Huawei Technologies Co. and Alibaba Group Holding Ltd.’s chip unit T-Head are also expected to boost their domestic market shares at a time when Nvidia and US chipmaker Advanced Micro Devices Inc. are virtually locked out of the market. Huawei is sharply ramping up production of its most advanced artificial intelligence chips in 2026, aiming to produce about 600,000 of its marquee 910C Ascend chips, Bloomberg News reported last year.
As a result, Chinese companies expect to spend 46% of their AI accelerator budget in the next 12 months to acquire locally-made chips, compared to 30% now, according to a survey of Chinese technology executives by Bloomberg Intelligence in June. AI accelerators produced by Cambricon and Hygon Information Technology Co. were being evaluated by a large pool of respondents to the survey.
In the longer term, Chinese chipmakers are set to benefit from Beijing’s ambitious data center buildout that aims to spend 2 trillion yuan ($295 billion) over the next five years. The massive project will rely on local suppliers including Huawei for at least 80% of technology such as AI chips. That would push China’s self-sufficiency level in AI chips to 70% at the end of this decade, up from 42% in 2025, according to estimates from Morgan Stanley.
Read More: China Preps $295 Billion Plan to Fund Nationwide AI Buildout
Nvidia boss Jensen Huang has consistently advocated for the use of his firm’s chips in China to ensure America’s lead in cutting-edge technology. His arguments alerted Beijing to the prospect of over-relying on foreign technology and creating vulnerabilities in the nation’s quest for global AI leadership, according to Morningstar analyst Phelix Lee.
“Most people would privately agree that having access to Nvidia would always be ideal, but in the absence of that, the trend I’m seeing is more Chinese chips being used,” Lee said.
China-made chips, when it comes to single-unit performance, are still years behind those built by Nvidia. It’s partly the result of a yearslong US-led campaign that restricts the most advanced chipmaking machines including the state-of-the-art extreme ultraviolet lithography systems made by Dutch company ASML Holding NV from flowing to China, the kind of gear that industry leader Taiwan Semiconductor Manufacturing Co. uses to make the world’s most powerful AI accelerators.
As customers start to shift from using chips for training AI models to integrating them into their businesses, Chinese chipmakers are winning orders from their ability to generate tokens at competitive pricing, Morgan Stanley analysts including Charlie Chan said in a research note at the end of July.
“Purchasing decisions are increasingly driven by deployment economics rather than absolute peak silicon performance,” Chan wrote.
Chinese chipmakers also have concluded that they can compensate for a lack of computing power in a single chip with scale. At China’s premier AI summit held in Shanghai last month, Chinese chipmakers from Huawei to Moore Threads and MetaX Integrated Circuits Shanghai Co. showcased their server racks capable of integrating tens of thousands chips.
“What they are trying to reassure is that we have enough chips and we have the infrastructure to link all these chips together so we can outnumber you,” Morningstar’s Lee said. “It also demonstrates how China, given enough time, will eventually narrow the gap to much narrower levels than today.”
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。