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: AI硬件估值框架
  • Current 研究记录序号: 3/8
  • Current topic: 可交付MW溢价与迁移租金的股票定价与分化
  • Current question: 给定这张确定性vs纸面MW地图,权益市场是否已把可交付MW溢价与迁移到电力设备/电网/冷却的租金充分计入,还是在6月底拥挤回吐(GEV −5.82%、ETN −4.53%)之后仍存在可交易的估值分化?
  • Current stance: support

Completed Cards

既有研究记录

  • Analyst: TMT行业分析师
  • Stance: initial
  • Topic: 截至2026-07-18,算力估值应从GPU单因子切换为电力优先、HBM/存储次之、GPU平台兜底的三层瓶颈框架。
  • Question: Investigate the topic "截至2026-07-18,算力估值应从GPU单因子切换为电力优先、HBM/存储次之、GPU平台兜底的三层瓶颈框架。" and make the strongest evidence-based judgment you can.
  • Summary: 截至2026-07-18,本报告支持AI硬件估值从单纯GPU beta转向瓶颈分层:Time-to-Power为第一变量,HBM与存储/内存层级为第二变量,GPU仍是吞吐核心但不再单独决定估值。公开收盘数据未验证1.70为最新收盘比率,因此将其视为院内压力阈值;已验证的2026-07-16 VXN/VIX为1.63,2026-07-17为1.50。投资上应上调有电力保障、HBM/封装能力和内存采购可见性的公司,折价无电力背书或内存薄弱的算力承诺。
  • Handoff: ai-infrastructure-analyst | AI数据中心Time-to-Power折价曲线 | 哪些AI数据中心开发商已经锁定确定性MW,哪些仍依赖可能被并网、变压器、燃机或冷却瓶颈推迟的纸面GPU订单?

既有研究记录

  • Analyst: AI基础设施分析师
  • Stance: support
  • Topic: AI数据中心Time-to-Power折价曲线
  • Question: 哪些AI数据中心开发商已经锁定确定性MW,哪些仍依赖可能被并网、变压器、燃机或冷却瓶颈推迟的纸面GPU订单?
  • Summary: 支持并延伸既有研究记录的瓶颈分层:AI数据中心估值应从'谁锁定GPU'转向'谁的MW可物理交付、何时交付'。确定性MW开发商(Chevron–Microsoft Kilby 2.67GW、xAI/SpaceX Colossus表后燃机、已运营的Stargate Abilene 1.2GW、微软三里岛835MW核电PPA)绕开并网队列,应按接近铭牌估值。纸面GPU阵营受制于并网排队4–7年、变压器交期3–5年、燃机档期已排至2029–2030:16GW已宣布仅5GW在建,7GW缺口、近半滑期。注意'已签约≠已并网'(CoreWeave签约3.5GW但2026活跃仅约1.7GW)。据此给出自测算折价曲线:A档0–10%、B档15–30%、C档40–70%。印证th_T01。
  • Handoff: chief-strategist | 可交付MW溢价与迁移租金的股票定价与分化 | 给定这张确定性vs纸面MW地图,权益市场是否已把可交付MW溢价与迁移到电力设备/电网/冷却的租金充分计入,还是在6月底拥挤回吐(GEV −5.82%、ETN −4.53%)之后仍存在可交易的估值分化?

Shared Workspace

  • research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/研究档案 (25896 bytes)
  • research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/meta.json (2796 bytes)
  • research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/report.en.md (11352 bytes)
  • research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/report.zh.md (9954 bytes)
  • research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/meta.json (5635 bytes)
  • research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/report.en.md (10092 bytes)
  • research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/report.zh.md (9384 bytes)
  • research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/研究档案 (2552 bytes)

Archived Files

Archived files - research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/研究档案 (25896 bytes) - research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/meta.json (2796 bytes) - research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/report.en.md (11352 bytes) - research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/report.zh.md (9954 bytes) - research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/meta.json (5635 bytes) - research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/report.en.md (10092 bytes) - research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/research note/report.zh.md (9384 bytes) - research discussion/3f88efad-85c6-47b7-8847-d0c934e2a2b2/研究档案 (2552 bytes)