返回投资研究台 2026-07-18

research thread Brief

  • 研究线程标识: [source-id-redacted]
  • Shared Agent Route session: [source-id-redacted]
  • Root topic: 全球 AI 链条的同步脱钩风险
  • Current 研究记录序号: 8/8
  • Current topic: 政策对冲能否使A股两融去化的第二/第三档触发永不发生
  • Current question: 在两融连续11个交易日、累计1723.18亿元去化并叠加人民币中间价主动管理的情况下,降准、平准资金买入与券商风控宽限的响应速度与规模,是否足以让本报告第二/第三档触发(单日融资余额降超300亿元、隔夜CNH HIBOR升破4%、维保低于130%账户占比超5%)永不出现?若是,6.0单位CNH/HKD应提前降回4.0单位还是维持至2026-09-30?
  • Current stance: stress-test

Completed Cards

既有研究记录

  • Analyst: 全球宏观分析师
  • Stance: initial
  • Topic: 同步下修风险为6.5/10:不是全球熊市已确认,而是AI领导资产在台湾/韩国集中度、capex久期与政策不确定性下的脆弱重估。
  • Question: Investigate the topic "同步下修风险为6.5/10:不是全球熊市已确认,而是AI领导资产在台湾/韩国集中度、capex久期与政策不确定性下的脆弱重估。" and make the strongest evidence-based judgment you can.
  • Summary: 截至2026-07-18,全球AI链条同步估值下修风险偏高但有条件,核心不是当前盈利恶化,而是台湾先进制程、韩国存储/HBM、hyperscaler capex与出口管制被市场当作同一AI beta来重估。TSMC与韩国出口数据仍强,支持AI基本面未坏;但先进产能与美国逻辑芯片暴露高度集中,意味着任何供应、政策或仓位冲击都可能迅速跨区域传导。
  • Handoff: ai-infrastructure-analyst | AI基础设施capex能否支撑更高的台湾/韩国集中度风险溢价 | 如果权益市场要求更高的台湾/韩国供应链风险溢价,2026-2031年AI基础设施需求、GPU/HBM交付与数据中心利用率是否仍能支撑当前capex路径?

既有研究记录

  • Analyst: AI基础设施分析师
  • Stance: stress-test
  • Topic: 2026-27年AI基础设施capex仍有需求、GPU/HBM和利用率支撑,但无法完全抵消台湾/韩国集中度风险溢价上行对ROIC和估值倍数的压缩。
  • Question: Investigate the topic "2026-27年AI基础设施capex仍有需求、GPU/HBM和利用率支撑,但无法完全抵消台湾/韩国集中度风险溢价上行对ROIC和估值倍数的压缩。" and make the strongest evidence-based judgment you can.
  • Summary: 截至2026-07-18,本报告压力测试既有研究记录的同步估值下修逻辑:AI基础设施capex的需求与利用率基础仍强,Microsoft、Alphabet、Amazon、Meta、NVIDIA、TSMC和HBM合约信号均显示2026-27仍是供给约束而非需求崩塌。但更高台湾/韩国风险溢价会提高融资和估值折现压力,因为更多capex被组件通胀、电力接入、长期场址与地理冗余消耗。结论是capex路径短期可持续,权益倍数更脆弱。
  • Handoff: semiconductors-analyst | 2027年HBM3E/HBM4与先进封装供给曲线能否抵消台湾/韩国风险溢价 | 在更高台湾/韩国权益风险溢价下,2027年HBM3E/HBM4、CoWoS/先进封装和先进制程供给是否仍紧到足以支撑AI存储与foundry利润率,还是会出现供给空气袋触发估值下修?

既有研究记录

  • Analyst: 半导体分析师
  • Stance: stress-test
  • Topic: 2027年HBM3E/HBM4与先进封装供给曲线能否抵消台湾/韩国风险溢价
  • Question: Investigate the topic "2027年HBM3E/HBM4与先进封装供给曲线能否抵消台湾/韩国风险溢价" and make the strongest evidence-based judgment you can.
  • Summary: 截至2026-07-18,本报告压力测试既有研究记录并收窄其结论:2027年供给确实紧,但不对称。HBM合约已锁至2027(SK hynix 1Q26营业利润率72%),且到2027年底占DRAM晶圆投片约30%却只产出约13%的bit;而CoWoS供需缺口正从20%收窄至10%,2027年扩产超60%、TSMC月产能或达20万片。瓶颈由封装迁移至存储,稀缺溢价将先离开封装/foundry。对foundry而言供给紧张与台湾风险溢价是同一变量,无法互相抵消。真正的空气袋是2028年物理事件、2027年定价事件,脆弱窗口约4Q2026–2Q2027。
  • Handoff: asset-allocator | 针对可定日期的AI链条估值下修窗口(约4Q2026–2Q2027)的跨资产配置与对冲方案 | 如果AI链条的估值下修窗口可定位在约4Q2026–2Q2027,且轮动方向是撤出封装/foundry稀缺溢价、保留已合约锁定的存储,那么跨股票/信用/商品/外汇的组合应如何调整规模、对冲与节奏,才能在不放弃AI上行的前提下承受台湾/韩国集中度冲击?

既有研究记录

  • Analyst: 首席策略师
  • Stance: synthesize
  • Topic: AI上行仍应保留,但4Q2026–2Q2027应把组合从封装/foundry稀缺溢价转为“存储多头+外汇/期权/黄金保护+电力约束上行”的跨资产哑铃。
  • Question: Investigate the topic "AI上行仍应保留,但4Q2026–2Q2027应把组合从封装/foundry稀缺溢价转为“存储多头+外汇/期权/黄金保护+电力约束上行”的跨资产哑铃。" and make the strongest evidence-based judgment you can.
  • Summary: 截至2026-07-18,本报告综合前序判断:4Q2026–2Q2027的AI估值下修风险应通过跨资产“AI哑铃”处理,而不是退出AI。建议以100单位主动风险预算为基准,保留30–34单位AI上行,削减封装/foundry稀缺溢价,保留已合约锁定的存储;同时配置12–15单位外汇/期权/黄金对冲、15单位现金/利率,并在2026-10-01前完成主要减仓和第一层对冲。A股侧减拥挤硬件、转电力电网与应用软件;信用侧避免重复权益beta;商品侧买AI电力约束而非全球周期beta。
  • Handoff: fx-strategist | 台韩集中度冲击下的USD/KRW、USD/TWD代理对冲与CNH流动性安排 | 在台韩AI供应集中度风险上升、但台湾仍有大额外汇储备且韩国外汇市场成交深化的情况下,应如何设定USD/KRW、USD/TWD代理对冲、JPY/CHF冲击对冲与CNH流动性比例,才能降低组合尾部风险又不牺牲AI上行?

既有研究记录

  • Analyst: 全球宏观分析师
  • Stance: synthesize
  • Topic: 台韩AI集中度尾部应通过“KRW主对冲、TWD期权化、JPY/CHF凸性、CNH流动性”的分层货币栈管理,而不是用一笔大额亚洲货币空头替代AI仓位。
  • Question: Investigate the topic "台韩AI集中度尾部应通过“KRW主对冲、TWD期权化、JPY/CHF凸性、CNH流动性”的分层货币栈管理,而不是用一笔大额亚洲货币空头替代AI仓位。" and make the strongest evidence-based judgment you can.
  • Summary: 截至2026-07-18,本报告综合既有研究记录并给出货币执行比例:在100单位主动风险预算中,保留30-34单位AI上行,同时将13单位外汇/期权覆盖拆为USD/KRW 5.5、USD/TWD 3.0、JPY/CHF 2.0、权利金与保证金2.5;在15单位现金/利率储备中保留4.0单位CNH/HKD可调度流动性。KRW是核心代理,TWD以期权为主,JPY/CHF买全球冲击凸性,CNH用于结算和再入场,不应成为主要做空人民币押注。
  • Handoff: china-macro | CNH流动性、人民币中间价管理与A股两融压力的联动压力测试 | 如果台韩AI集中度冲击与人民币中间价管理、CNH资金波动、A股两融压力同时出现,4.0单位CNH/HKD可调度流动性是否足够,还是需要把CNH流动性比例提高到6.0单位并削减其他对冲腿?

既有研究记录

  • Analyst: 全球宏观分析师
  • Stance: stress-test
  • Topic: 4.0单位CNH/HKD流动性不足以覆盖人民币中间价管理叠加A股两融去杠杆的同日压力,应临时提高到6.0单位,但这应是流动性垫而非人民币贬值押注。
  • Question: Investigate the topic "4.0单位CNH/HKD流动性不足以覆盖人民币中间价管理叠加A股两融去杠杆的同日压力,应临时提高到6.0单位,但这应是流动性垫而非人民币贬值押注。" and make the strongest evidence-based judgment you can.
  • Summary: 截至2026-07-18,本报告压力测试既有研究记录的4.0单位CNH/HKD流动性假设并给出修正:USD/CNH和CNH HIBOR尚未显示人民币即期或资金面失控,但A股融资余额已从2026-07-13的人民币2.880236万亿元降至2026-07-16的人民币2.829025万亿元,低于研究院人民币2.86万亿元压力线。因此建议至少到2026-09-30将CNH/HKD可调度流动性提高到6.0单位,同时不削减USD/KRW核心对冲,而从USD/TWD、JPY/CHF和通用权利金/保证金储备中挪出2.0单位。
  • Handoff: ashare-strategist | A股融资盘担保品压力、AI/半导体拥挤仓位与6.0单位CNH/HKD流动性触发规则 | 在融资余额已经跌破人民币2.86万亿元压力线后,哪些A股行业和高融资余额标的最可能把AI链条外部冲击转化为被动卖出,6.0单位CNH/HKD流动性应按哪些盘中指标释放?

既有研究记录

  • Analyst: 首席策略师
  • Stance: stress-test
  • Topic: A股融资盘担保品压力、AI/半导体拥挤仓位与6.0单位CNH/HKD流动性触发规则
  • Question: Investigate the topic "A股融资盘担保品压力、AI/半导体拥挤仓位与6.0单位CNH/HKD流动性触发规则" and make the strongest evidence-based judgment you can.
  • Summary: 截至2026-07-18,本报告压力测试既有研究记录:支持6.0单位CNH/HKD额度,但否定其隐含的全市场强平连锁机制。两融11连降累计1723.18亿元至28581.56亿元,但账户级数据显示仅1/400被强平、维保低于140%者约3%。真正的转化器是电子:融资余额约6204亿元、占全市场21.5%,2026-07-13却贡献当日51.0%的去化,超配2.37倍。医药生物同日增20.98亿元,说明是轮动非离场。个股须区分反身性(胜宏科技融资/流通市值9.08%)与冲击(中际旭创444.40亿元)。6.0单位按三档盘中触发释放:电子净卖出>150亿+5连降+成交额<3万亿;余额单日降>300亿或隔夜CNH HIBOR>4%;维保<130%账户占比>5%。
  • Handoff: china-macro | 政策对冲能否使A股两融去化的第二/第三档触发永不发生 | 在两融连续11个交易日、累计1723.18亿元去化并叠加人民币中间价主动管理的情况下,降准、平准资金买入与券商风控宽限的响应速度与规模,是否足以让本报告第二/第三档触发(单日融资余额降超300亿元、隔夜CNH HIBOR升破4%、维保低于130%账户占比超5%)永不出现?若是,6.0单位CNH/HKD应提前降回4.0单位还是维持至2026-09-30?

Shared Workspace

  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/研究档案 (25896 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (2937 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (11148 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (10463 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (7832 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (14370 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (12859 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (6990 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (15381 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (14460 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (3567 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (15362 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (13865 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (3308 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (19047 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (17474 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (3543 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (13193 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (12303 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (6800 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (15851 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (15003 bytes)
  • research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/研究档案 (12222 bytes)

Archived Files

Archived files - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/研究档案 (25896 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (2937 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (11148 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (10463 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (7832 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (14370 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (12859 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (6990 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (15381 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (14460 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (3567 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (15362 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (13865 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (3308 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (19047 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (17474 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (3543 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (13193 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (12303 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/meta.json (6800 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.en.md (15851 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/research note/report.zh.md (15003 bytes) - research discussion/97f09375-f51c-48ad-a78a-a81e4aadd29b/研究档案 (12222 bytes)