← INTC 研究档案

INTC · Valuation DR

深潜研究原文 · 永久存档,不回改

INTC — Phase 1.5 Pre-cryst Valuation DR

Generated: 2026-07-26T14:31:16Z
Round: 4 | research_version: 1
Source: Claude Code CLI (dr_dispatcher, phase=phase1_5_valuation)
Independence: xbrl_context=None (ADR-0034 §4.2)


INTC 独立估值分析(Phase 1.5 Pre-crystallization)

一、Segment 识别逻辑

Phase 1 三轮研究材料清晰呈现 Intel 的四个报表可辨识板块,加上一个非合并权益资产。基于经济性质差异(收入模式、周期性、竞争位置、盈利质量),我把 Intel 拆为四个 segment 独立估值:

  1. CCPG(Client Computing & Physical AI Group) — 消费级/商用 PC CPU 为主业,Physical AI 是 optionality 但当前贡献 negligible
  2. DCAI(Data Center & AI) — Xeon 服务器 CPU + 新兴 ASIC 代工,Gaudi 已被承认为 dead-end
  3. Intel Foundry — 制造板块,会计上分离,经济上是内部服务 + 极小外部
  4. Altera 49% Stake + All Other — Altera 已剥离控股权,本质是 non-consolidated 权益资产 + 尾部业务

内部转让定价的处理:三轮材料反复强调 CCPG/DCAI 的分部利润是"按近似市价从 Foundry 采购晶圆"后的口径,本质是 Foundry 在承担全部制造固定成本。若直接把三个板块的分部利润用 fabless multiples 加总,会系统性高估(因为 Foundry 独立看是负值,且带巨额折旧)。我的处理方式:对 CCPG/DCAI 采用保守的 multiple 区间(承认其账面利润率含隐性 Foundry 补贴),Foundry 独立作为 real option + 部分清算价值估算。


二、Per-segment 估值方法与关键选择

Segment 1 — CCPG

方法:EV/Sales × P/E 混合三角化。

关键推算
- 营收年化基础:Q2 2026 $8.9B,Q1 假设 $7.5B(Phase 1 未直接给出),年化约 $33–35B
- Q2 分部运营利润率 ~26%($2.3B)— 但含转让定价补贴,"真实"margin 估 15–20%
- PC 市场结构性收缩(IDC 预测 -11%)+ AMD 持续侵蚀笔记本至 25%、桌面至 35%+ + ARM PC 逐步渗透 = 长期营收温和衰减
- Physical AI/Robotics 是 optionality(130+ 商业设计伙伴),但贡献 negligible

Multiple 选择
- 参考同类"结构性衰减但有 optionality"的消费半导体标的,我给 P/E 范围 8–15x(不用 25x growth multiple,因为不是 growth story),EV/Sales 0.9–2.5x(对应 mature/late-cycle 消费半导体)
- Bear:份额继续流失、AMD 进一步吃份,$32B × 1.0x = $32B
- Base:份额缓慢流失但 ASP 稳定、Panther Lake 顺利,$35B × 1.7x = $60B
- Bull:Panther Lake 打开 AI PC 换机潮 + Physical AI 兑现 optionality,$38B × 2.4x = $90B

Segment 2 — DCAI

方法:EV/Sales × normalized P/E 混合。

关键推算
- Q1 2026 $5.1B + Q2 2026 $6.3B = H1 $11.4B,全年年化 $23–26B
- Q2 39.5% margin 存在明显非可持续成分:Gaudi 减值消失($512M base effect)+ Xeon 6 集中出货 ASP +48%(大客户 concentration)+ 供给紧张下高价出货
- 校准 Q4 2025 只有 5.1% margin — 说明 margin 波动区间极大
- 长期竞争:AMD EPYC 营收份额 46.2%(历史峰值),Venice 已发布;ARM server CPU 结构性侵蚀(Graviton/Axion/Cobalt);ASIC 内部化压缩 attach rate
- 积极信号:Agentic AI 可能拉高 CPU:GPU ratio;ASIC 代工业务年化 $1B+ 增长快;Diamond Rapids 2027 有机会

Multiple 选择
- Normalized DCAI margin 假设 22–28%(远低于 Q2 峰值),normalized operating income $5–7B
- 净利润 ~$4–5.5B(税率 20% + 分摊利息),apply 12–20x P/E
- EV/Sales 交叉验证:$24B × 1.5–5.0x = $36–120B
- Bear:AMD 继续吃份 + Q2 是 anomaly,$22B × 1.6x = $35B
- Base:Xeon 6 结构性受益 AI attach 但 margin 回落至 25%,$26B × 3.7x = $95B
- Bull:Agentic AI CPU:GPU ratio 兑现 + ASIC 代工业务规模化,$30B × 5.2x = $155B

Segment 3 — Intel Foundry

方法:real option + 部分资产回收价值(liquidation floor)+ 消耗性 DCF。

关键推算
- Q2 2026 营收 $5.8B(约 95% 内部),年化 $22B
- 运营亏损 -$2.1B/季度 = 年化 -$8B;主因固定成本摊薄不足 + 18A 良率爬坡
- 存量净 PP&E 估计 $50–60B(累计资本开支 $95B 减折旧)— 提供 partial liquidation floor
- CHIPS Act 已到账 $2.2B + 政府股权转换 $8.9B(本质是 sunk equity injection)
- 外部客户 pipeline:Fortinet(Intel 4 小规模)+ 14A 两个未名 prospects(承诺窗口 H2 2026–H1 2027)
- 折旧压力:约 $14–15B/年(占公司 D&A 70–75%)
- 管理层承诺 2027 年底 break-even(依赖内部产品,外部只需 low-to-mid single billions)

估值框架
- Bear:18A 外部客户全部延后 + 14A 承诺不兑现 + 持续 3–4 年亏损累计 $20–25B — 但资产可部分回收(战略资产),floor 值 = $15B
- Base:管理层 2027 break-even 大致兑现 + 少量外部订单 + 2030 年运营利润 $2–3B,DCF back = $50B
- Bull:14A 签下 1–2 家超大规模客户 + 政府/国防订单放量 + 2030 年外部收入 $8–10B、blended margin 20% → 估值向 TSMC 折价的 3–4x sales 靠拢 = $110B

Segment 4 — Altera 49% Stake + All Other

方法:sum of parts(Altera 用近期交易 EV 反推 + IPO uplift 概率加权)。

关键推算
- Altera 51% 股权交易 EV $8.75B(Silver Lake,2025-09) → 49% 隐含 ~$4.3B
- 2026 年底 IPO 预期若 tape 顺利,估值可能上修至 $6–10B EV → 49% = $3–5B
- All Other Q2 2026 营收 $0.7B YoY -33%(收缩中),价值极小 $1–3B

三档
- Bear:$5B(Altera IPO 推迟/低估值 + All Other 零价值)
- Base:$9B(Altera IPO 如期,估值持平剥离 EV + All Other 尾部价值 $3B)
- Bull:$13B(Altera IPO 溢价上市 AI/国防题材 + All Other 价值实现)


三、企业级汇总

Segment EV 加总

Segment Bear Base Bull
CCPG 32 60 90
DCAI 35 95 155
Intel Foundry 15 50 110
Altera + All Other 5 9 13
Sum EV 87 214 368

Net Debt 调整
- Long-term debt $46.6B(Phase 1 R2)
- Cash 估计 $25–30B(含近期 Altera $4.3B + CHIPS 拨付 $2.2B + 现有流动性)
- Net Debt ≈ $20B

Equity Value(对全体股东):Bear $67B / Base $194B / Bull $348B

政府 ~10% 股权稀释:CHIPS Act 转换的 $8.9B 股权 + 5 年期 warrant($20 行权价可再取 5%)。若视作现有普通股股东被稀释 10%:
- Bear ≈ $60B
- Base ≈ $175B
- Bull ≈ $315B

我在 JSON 中报告的 total_*_b 采用稀释后 common equity 口径(即最终归属于普通股股东的价值)。


四、Bull/Base/Bear 差异化的核心 levers

三档 spread 主要由以下 lever 驱动:
- DCAI 长期 margin(22% vs 28% vs 35%)— Q2 39.5% 显然非可持续
- Intel Foundry 外部客户转化(0 vs $2–3B vs $8–10B 年化)
- CCPG 份额趋势(继续流失 vs 稳定 vs Physical AI 打开 optionality)
- 中国政策(11-10 关税暂停到期后是延续/升级/加剧管制)— 未单独 quantify 但反映在 CCPG/DCAI 各档

Bull vs Bear 约 5.3x spread 反映结构转型的极高不确定性:Foundry 从 -$8B 变成 +$3B 就是 $30B+ 估值差;DCAI margin 波动 10 pts 就是 $30B 估值差。这不是普通稳定业务,估值区间必须宽。


五、Deliberately 跳过 / 未做的事

  1. 未 anchor 市值:Phase 1 材料未提及 INTC 当前市值,我也未去猜测。我的 $175B base 是完全从 fundamentals 推算的,可能与市场价存在显著 disconnect — 这是设计特征。
  2. 未使用 Phase 1 出现的隐含 anchor:如 R2 提到"股东权益 $87.5B"这是账面数不是市值,但我仅将其作为杠杆背景,未用它 anchor 估值。R3 提到 "长期债务 $46.6B" 和 CHIPS 股权 $8.9B 我用于 net debt 调整和稀释处理。
  3. 未做单元经济学详细拆解:如 Foundry 每片晶圆 marginal contribution、DCAI 每 SKU BOM 结构 — Phase 1 材料不足以支撑,我用 aggregate margin 判断代替。
  4. 未 apply peer multiple table:TSMC/AMD/Nvidia 等对标估值我 deliberately 未套用。我的 multiples 是基于"这个业务应该值多少"的 first-principles 判断,不是"peer 是这个数所以我也是"。
  5. 未 quantify Escrowed Shares $12.5B non-cash mark-to-market:因 Phase 1 未清晰说明这些股份最终归属谁,我保守假设它不是普通股股东的 real value 减损(会计事项)。

六、关键假设(Phase 2 falsifiability pointers)

见下方 JSON key_assumptions

{
  "schema_version": 1,
  "ticker": "INTC",
  "produced_at": "2026-07-26T14:45:00Z",
  "total_bull_b": 315,
  "total_base_b": 175,
  "total_bear_b": 60,
  "epistemic_tag": "EST",
  "segments": [
    {
      "segment_name": "CCPG (Client Computing & Physical AI Group)",
      "methodology": "EV/Sales 1.0-2.4x blended with implied P/E 8-15x on subsidy-adjusted earnings",
      "bull_b": 90,
      "base_b": 60,
      "bear_b": 32,
      "narrative": "Annualized revenue ~$34B. Applied EV/Sales 1.0x (bear) to 2.4x (bull) reflecting structural PC decline, AMD share erosion (25% laptop / 35%+ desktop), and Physical AI optionality (Intel Robotics, 130+ design partners) as upside lever. Bull case assumes Panther Lake unlocks AI PC refresh cycle."
    },
    {
      "segment_name": "DCAI (Data Center & AI)",
      "methodology": "EV/Sales 1.6-5.2x blended with normalized P/E 12-20x on cyclically-adjusted earnings",
      "bull_b": 155,
      "base_b": 95,
      "bear_b": 35,
      "narrative": "Annualized revenue $22-26B. Q2 2026 39.5% margin heavily reflects Gaudi impairment base-effect + Xeon 6 ASP surge; normalized margin 22-28%. Bull case assumes Agentic AI CPU:GPU ratio thesis materializes + ASIC foundry scale-up; bear reflects AMD Venice share pressure + margin regression to Q4 2025 anomaly-adjacent levels."
    },
    {
      "segment_name": "Intel Foundry",
      "methodology": "Real option overlay + partial liquidation floor + DCF back on 2028+ break-even scenario",
      "bull_b": 110,
      "base_b": 50,
      "bear_b": 15,
      "narrative": "Currently loss-making at -$8B/yr annualized. Bear reflects PP&E floor (~$50-60B gross) discounted for continued cash burn and government equity subordination. Base assumes 2027 break-even per mgmt guidance driven by internal Panther Lake / Diamond Rapids volume. Bull requires 1-2 hyperscaler wins on 14A (mgmt cites 2 prospects) and external revenue ramping to $8-10B by 2030."
    },
    {
      "segment_name": "Altera 49% Stake + All Other",
      "methodology": "Precedent transaction (Silver Lake 51% at $8.75B EV) + probability-weighted IPO uplift + residual All Other value",
      "bull_b": 13,
      "base_b": 9,
      "bear_b": 5,
      "narrative": "Altera 49% stake baseline value ~$4.3B implied by 2025 Silver Lake transaction; IPO uplift adds $0-3B in bull. All Other (residual NEX/edge products) valued at $1-3B on ~$3B annualized declining revenue."
    }
  ],
  "key_assumptions": [
    "DCAI Q2 2026 39.5% operating margin is not sustainable; normalized long-term margin is 22-28% due to Gaudi base-effect absence in FY27+ and AMD Venice pressure. Falsified if Q3-Q4 2026 DCAI margin stays >35%.",
    "Intel Foundry achieves operating break-even by end-2027 primarily via internal product volume (Panther Lake, Diamond Rapids) without material external customer contribution. Falsified if Q3-Q4 2026 Foundry op loss fails to narrow below -$1.5B/quarter.",
    "At least 1 of 2 mgmt-cited external 14A customer prospects converts to volume production commitment by H1 2027. Falsified if no external 14A commitment publicly announced by Q2 2027 earnings.",
    "Panther Lake (18A) achieves stable production yield without quality issues that would impair external foundry credibility. Falsified by any material Panther Lake recall or shipment miss in H2 2026.",
    "China revenue continues secular decline from $12.7B (FY25) toward $8-10B annualized floor without triggering acceleration below $7B. Falsified if Nov-10 tariff pause expiry triggers cliff-drop of >$3B annualized in single quarter.",
    "Net debt ~$20B (long-term debt $46.6B minus estimated cash $25-30B including recent Altera proceeds and CHIPS injection). Falsified by material additional financing round or cash burn exceeding CapEx guidance.",
    "US Government ~10% equity stake (CHIPS conversion) remains passive without operational interference on foundry customer selection or foreign order acceptance. Falsified by any government-directed strategic constraint on commercial decisions."
  ]
}