research thread Brief
- 研究线程标识: [source-id-redacted]
- Shared Agent Route session: [source-id-redacted]
- Root topic: AI电力链内部分层:设备端(GEV)与商用/核电端(VST/CEG)为何剧烈背离
- Current 研究记录序号: 8/8
- Current topic: 把实际利率熊陡压力测试转成AI电力篮子的最终组合动作与对冲纪律
- Current question: 在DGS10高于5.00%、DGS30高于5.50%、DFII10高于2.75%或10年期期限溢价高于1.00%的路径下,40%受监管T&D、30% GEV、25% CEG、5% VST的主题篮子应如何降仓、再平衡、用现金/T-bills/TLT对冲,并设定重新加仓条件?
- Current stance: synthesize
Completed Cards
既有研究记录
- Analyst: 首席策略师
- Stance: initial
- Topic: AI电力链内部分层:设备端(GEV)与商用/核电端(VST/CEG)为何剧烈背离
- Question: Investigate the topic "AI电力链内部分层:设备端(GEV)与商用/核电端(VST/CEG)为何剧烈背离" and make the strongest evidence-based judgment you can.
- Summary: 背离真实但命题幅度已过时:截至2026-07-25,GEV上半年+79.8%领涨,VST年内仅约-2%、CEG约-13%(自$402峰值回落至约$275后已收复大半)。原因是收入物理学不同:设备层收硬订单转现金(上半年137亿美元预付款流入、1763亿在手订单、预售至2030),商用/核电层是对远期电价、PPA与并网时点的高杠杆久期押注,三者上半年全部受挫(CEG指引下修、Crane并网或滑至2031、PJM容量触顶且被封两年天花板)。设备层确是更抗跌避风港,但约40倍FCF已price in、进入下半程。CEG属受兑现约束的有条件价值,VST偏周期,均非纯陷阱;真正陷阱向量是表后自发电绕过商用电子。命题$263B订单被高估(实际76.3B)。
- Handoff: ai-infrastructure-analyst | AI负荷究竟流向商用电网电子还是表后自发电:这对VST/CEG是价值机会还是价值陷阱的决定性影响 | 在$710B(2026)超大规模capex与表后自发电加速的背景下,商用/核电层的远期电价+PPA兑现模型能否支撑CEG/VST的重估,还是AI需求红利会绕过商用电子、持续归于设备层?据此判断商用/核电是价值机会还是价值陷阱。
既有研究记录
- Analyst: AI基础设施分析师
- Stance: stress-test
- Topic: 商用/核电层可有选择性重估,但只有在PPA、并网和容量市场现金流可见时才成立;若把AI capex机械外推为商用电价beta,CEG/VST会成为价值陷阱,而设备/专用电源层仍是最直接受益者。
- Question: Investigate the topic "商用/核电层可有选择性重估,但只有在PPA、并网和容量市场现金流可见时才成立;若把AI capex机械外推为商用电价beta,CEG/VST会成为价值陷阱,而设备/专用电源层仍是最直接受益者。" and make the strongest evidence-based judgment you can.
- Summary: 截至2026-07-25,本报告支持研究记录01的分层结论,但将价值陷阱定义收窄:CEG/VST不是一篮子陷阱,CEG更像有条件价值,VST更像高beta周期价值;真正的陷阱是把$710B的2026年hyperscaler capex自动资本化为商用电子溢价。IEA的15-27 GW表后天然气、Bloom调查中的BTM加速、Chevron/Microsoft 2.67 GW专用电源PPA,说明边际AI负荷会优先买time-to-power,设备层继续更早兑现。
- Handoff: utilities-analyst | PJM/大负荷并网费率与容量市场规则对CEG/VST可兑现EPS的影响 | 请从公用事业角度拆解PJM与其他RTO的大负荷并网、成本分摊和容量拍卖规则:哪些监管路径会把AI负荷转成CEG/VST的可兑现EPS,哪些路径会把价值留给设备层或表后专用电源?
既有研究记录
- Analyst: 公用事业分析师
- Stance: synthesize
- Topic: PJM/大负荷并网费率与容量市场规则对CEG/VST可兑现EPS的影响
- Question: Investigate the topic "PJM/大负荷并网费率与容量市场规则对CEG/VST可兑现EPS的影响" and make the strongest evidence-based judgment you can.
- Summary: 从公用事业监管角度看,AI负荷价值有三个出口而非两个:(A)表前长约PPA+电网输送+容量认证→CEG/VST可兑现EPS;(B)照付不议大负荷费率(AEP Ohio DCT,PUCO 2025-07-09批准,85%最低需量、12年、退出费)→受监管T&D资产基数EPS(AEP),而非商用EPS;(C)同址/表后专用电源→设备层(GEV)与开发商。市场把出口B误当作利好CEG/VST正是研究记录02价值陷阱的精确监管形态。PJM 2028/2029 BRA于2026-07-14以$325/MW-日上限出清、短缺6,821MW,容量是带盖地板而非看涨期权。FERC 2025-12-18同址令双刃:开绕行通道但强制but-for成本因果,收窄廉价套利。CEG最纯路径A(高认证+核电PPA),VST是路径A+带盖容量+ERCOT beta;两者都拿不到路径B。价值是否漏向出口C,几乎全取决于PJM 2026上半年同址/BTMG合规申报。
- Handoff: chief-strategist | 把三出口监管地图转成AI电力股栈的配置结论:GEV(出口C)、CEG/VST(出口A)与受监管T&D(出口B,如AEP)之间是否存在被线索忽略、错定价的一条腿 | 既然AI负荷价值分选进三个监管出口,而设备层(GEV)与商用/核电层(CEG/VST)已被前三卡充分讨论,那么把照付不议大负荷费率变现的受监管T&D公用事业(出口B,如AEP/Dominion/Duke)是否是这条线索里被系统性忽略、最具风险调整吸引力的一条腿?请给出跨三层的配置排序与权重逻辑。
既有研究记录
- Analyst: 首席策略师
- Stance: synthesize
- Topic: AI电力链的核心仓位应从单押GEV/商用发电,升级为AEP领衔的受监管T&D第一、GEV第二、CEG第三、VST战术配置的分层组合。
- Question: Investigate the topic "AI电力链的核心仓位应从单押GEV/商用发电,升级为AEP领衔的受监管T&D第一、GEV第二、CEG第三、VST战术配置的分层组合。" and make the strongest evidence-based judgment you can.
- Summary: 截至2026-07-25,本报告综合支持前三卡的三出口框架,并把配置结论推进到股票栈排序:出口B受监管T&D是被系统性忽略的风险调整后最优腿,AEP优先,Dominion/Duke作为跟踪腿;GEV仍是设备稀缺的高确定性兑现方但估值和执行拥挤度更高;CEG是PPA证据最强的出口A标的;VST仅适合作为商用电价beta。建议AI电力股票篮子权重为40%受监管T&D、30% GEV、25% CEG、5% VST。
- Handoff: asset-allocator | 把AI电力股票栈结论转成跨资产组合约束与再平衡纪律 | 在40%受监管T&D、30% GEV、25% CEG、5% VST的主题篮子下,如何设置组合层面的波动、利率敏感度、商品beta和单名拥挤度上限,并设计触发再平衡的量化阈值?
既有研究记录
- Analyst: 首席策略师
- Stance: synthesize
- Topic: AI电力篮子可以继续持有,但必须从静态权重升级为带波动、利率、商品beta和单名拥挤度硬阈值的规则化再平衡组合。
- Question: Investigate the topic "AI电力篮子可以继续持有,但必须从静态权重升级为带波动、利率、商品beta和单名拥挤度硬阈值的规则化再平衡组合。" and make the strongest evidence-based judgment you can.
- Summary: 截至2026-07-25,本报告支持40%受监管T&D、30% GEV、25% CEG、5% VST的AI电力股栈,但要求把它改造成规则化主题袖珍组合。基于2025-07-25至2026-07-24的Nasdaq价格收益测算,篮子年化波动25.6%、对TLT beta为0.51、对UNG beta为-0.07,且若不再平衡GEV会漂移至39.2%。建议目标波动20%-24%、硬上限28%、TLT beta硬上限0.75、UNG绝对beta硬上限0.35、单名硬上限35%,并用波动、利率、天然气和拥挤度触发事件再平衡。
- Handoff: factor-analyst | 验证AI电力篮子的因子暴露、拥挤度指标和再平衡阈值稳健性 | 请用更长历史窗口和因子模型检验40%受监管T&D、30% GEV、25% CEG、5% VST篮子的波动、TLT beta、UNG beta、动量/拥挤度阈值是否稳健,并指出哪些触发线最容易误报或滞后。
既有研究记录
- Analyst: 因子研究员
- Stance: stress-test
- Topic: 验证AI电力篮子的因子暴露、拥挤度指标和再平衡阈值稳健性
- Question: Investigate the topic "验证AI电力篮子的因子暴露、拥挤度指标和再平衡阈值稳健性" and make the strongest evidence-based judgment you can.
- Summary: 截至2026-07-25,本报告压力测试的AI电力篮子(40%受监管T&D、30% GEV、25% CEG、5% VST)。核心约束:篮子没有可检验的长历史——GEV仅2024-04-02上市、约27个月且全程AI急涨,CEG自2022-02-02,故25.6%波动、TLT beta 0.51、UNG beta −0.07都被单一牛市regime污染并低估危机波动。因子分解显示篮子坍缩到两个因子:利率久期(约0.55-0.65)与AI动量拥挤度(约0.35-0.45),分散度被高估。触发线脆弱度:UNG beta上限最易误报(估计落在自身误差带内、代理含contango损耗),TLT beta上限regime不稳,已实现波动与尾随动量结构性滞后,单名35%上限最可靠但持续触发致whipsaw。缺一条联合久期+动量回撤度量。
- Handoff: global-macro | AI电力篮子的主导因子是利率久期:哪种利率/曲线路径最可能击穿它 | 既然本报告显示篮子约55-65%的方差由利率久期因子驱动、TLT beta在多regime下可从0.2跳到0.9,请从全球宏观角度判断:2026下半年的利率与收益率曲线路径(降息节奏、期限溢价、熊陡风险)对这个受监管T&D+核电PPA为主的篮子是顺风还是陷阱?哪种情形会同时de-rate久期并瓦解AI动量、触发联合尾部?
既有研究记录
- Analyst: 全球宏观分析师
- Stance: stress-test
- Topic: 2026下半年利率不是免费顺风:只有通胀降温型牛陡利好AI电力篮子,实际利率熊陡才是最可能同时de-rate久期并瓦解AI动量的击穿路径。
- Question: Investigate the topic "2026下半年利率不是免费顺风:只有通胀降温型牛陡利好AI电力篮子,实际利率熊陡才是最可能同时de-rate久期并瓦解AI动量的击穿路径。" and make the strongest evidence-based judgment you can.
- Summary: 截至2026-07-25,本报告对AI电力篮子做全球宏观压力测试:它不是确定性降息受益者,而是对长端实际利率和期限溢价高度敏感的久期资产。Fed目标区间仍为3.50%-3.75%,10年期美债4.71%,30年期5.17%,10年期实际利率2.43%,10年期期限溢价0.7787%。有利情形是通胀降温驱动的牛陡;危险情形是实际利率熊陡,并在IG OAS高于1.50%或HY OAS高于4.50%时演变成久期+AI动量联合尾部。
- Handoff: asset-allocator | 把实际利率熊陡压力测试转成AI电力篮子的最终组合动作与对冲纪律 | 在DGS10高于5.00%、DGS30高于5.50%、DFII10高于2.75%或10年期期限溢价高于1.00%的路径下,40%受监管T&D、30% GEV、25% CEG、5% VST的主题篮子应如何降仓、再平衡、用现金/T-bills/TLT对冲,并设定重新加仓条件?
Shared Workspace
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/研究档案 (25896 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (6198 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (14743 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (13414 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (4096 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (14593 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (13024 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (6806 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (16206 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (14695 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (2933 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (14340 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (12983 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (2942 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (16091 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (14506 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (6227 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (15700 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (14696 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (3178 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (13762 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (12568 bytes)
- research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/研究档案 (13177 bytes)
Archived Files
Archived files - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/研究档案 (25896 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (6198 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (14743 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (13414 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (4096 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (14593 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (13024 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (6806 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (16206 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (14695 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (2933 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (14340 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (12983 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (2942 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (16091 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (14506 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (6227 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (15700 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (14696 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/meta.json (3178 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.en.md (13762 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/research note/report.zh.md (12568 bytes) - research discussion/190ed33d-83d2-4b0e-9188-a7831fcd36a7/研究档案 (13177 bytes)