2026年9月22日 星期二
INTC - Case Study - Bernstein illustrative manufacturing burden Improvement
可以。今次我當佢係一個 **one-time shadow forecast**,完全唔改我哋正式 forecast;之後只有 Intel 真係喺 results 入面交到持續 margin / cost improvement,我先會考慮將呢個 mechanism 正式搬入 model。
先定義清楚:你講嘅 **50% / 45%**,我理解係 Bernstein 嗰個 illustrative manufacturing burden 由目前約 **$60 per $100 Products revenue**,改善到:
> **Base-bull:$50**
> **Bull:$45**
唔係話 Intel corporate GM 變50%/45%。
我繼續用上次個中性 capture assumption:cost gap改善唔會100%變成profit,假設 **70%真正落到 operating profit**;15% tax;約5.4B diluted shares。
## 先睇純 EPS 槓桿
如果 burden由 **$60 → $50**:
> 10pts cost improvement × 70% capture
> = **+7pts operating profit**
如果由 **$60 → $45**:
> 15pts × 70%
> = **+10.5pts operating profit**
所以唔同 Products revenue level,大概係:
| Products revenue | $60→$50 EPS uplift | $60→$45 EPS uplift |
| ---------------: | -----------------: | -----------------: |
| **$80B** | **+$0.88** | **+$1.32** |
| **$100B** | **+$1.10** | **+$1.65** |
| **$120B** | **+$1.32** | **+$1.98** |
| **$150B** | **+$1.65** | **+$2.48** |
而家 Intel Products annualised revenue大概係 ~$60B級,所以去到2028–30,如果CPU cycle令Products revenue升到$80–120B以上,呢個 cost recovery就會變得非常material。Q2 2026 Products已經係約$15.1B quarterly revenue、31.8% segment OM,所以個starting base唔細。
---
# 對我哋原本 forecast 有幾大影響?
我哋最新正式 forecast係:
| Horizon | Base | Bull |
| ------- | -------: | -------: |
| **12M** | $140–170 | $190–235 |
| **18M** | $170–210 | $245–305 |
| **24M** | $195–240 | $295–370 |
| **36M** | $250–310 | $390–485 |
今次 shadow case我會假設:
> **Base:逐步走向 $50 burden**
> **Bull:逐步走向 $45 burden**
而且唔會假設立即完成;2027只係market開始price,2028–30先逐步反映落earnings。
### One-time adjusted shadow forecast
| Horizon | 正式 Base | **Shadow Base — burden→$50** | 正式 Bull | **Shadow Bull — burden→$45** |
| ------- | -------: | ---------------------------: | -------: | ---------------------------: |
| **12M** | $140–170 | **$155–190** | $190–235 | **$215–270** |
| **18M** | $170–210 | **$190–240** | $245–305 | **$280–355** |
| **24M** | $195–240 | **$220–275** | $295–370 | **$335–425** |
| **36M** | $250–310 | **$280–350** | $390–485 | **$435–550** |
呢張表唔係新official forecast,只係問:
> **如果Bernstein所講嘅 synthetic-IDM gap 真係開始被Intel收窄,原本估值大概可以多幾多?**
答案係:
> **Base大概多 $15–40/share**
> **Bull大概多 $25–65/share**
越長期越大,因為 revenue base越大,而cost improvement係按整個Products revenue base乘落去。
---
## 點解股價impact可以咁大?
假設去到2028左右:
> Products revenue ~$90–100B
如果burden去到$50:
> EPS大概多 **$1.0–1.1**
假設市場畀 sustainable earnings 20–22x:
> 額外值大約 **$20–24/share**
如果Bull去到$45:
> EPS大概多 **$1.5–1.65**
用22–25x:
> 額外大概 **$33–41/share**
去到2029–30,如果Products revenue已經$120B甚至更高:
> $50 case:+$1.3 EPS左右
> $45 case:接近+$2 EPS甚至更多
咁估值槓桿自然再放大。
---
# 最有意思其實係 EPS path
我哋之前一直討論:
> CPU Products有冇機會去到 **$6 EPS**
> Bull甚至$7–8。
如果只靠:
> volume + ASP + mix
可能要一個非常強CPU cycle。
但如果再加:
> manufacturing burden下降
件事會容易好多。
例如原本某個Bull year:
> Products EPS = **$6.0**
再加 $60→$50:
> 可能變 **$7.0–7.3**
如果去到$45:
> 甚至可能 **$7.5–8.0+**
所以呢個sensitivity真正有趣之處係:
> **$8 EPS唔一定需要Intel多賣一倍CPU。**
>
> 有一部分可以來自同樣收入,但製造成本正常化。
---
## 而且呢個同CPU boom係乘法,不係加法
最bullish組合係:
> CPU TAM ↑
>
> * ASP ↑
> * Intel supply ↑
> * Products revenue ↑
> * 18A / Intel 3 yield ↑
> * die size / packaging cost ↓
如果只有cost improvement但revenue唔增,value有限。
但如果Intel由 ~$60B Products revenue:
> → $100B
> → $120B
同時每$100 revenue製造負擔:
> $60 → $50 → $45
咁 profit expansion會非常非線性。
呢個先係點解Bernstein嗰個 synthetic IDM comparison值得研究。
---
# 但有一個我一定要壓住嘅地方
我唔會假設:
> $60 → $45一定發生。
因為嗰個 gap入面有好多嘢:
* Foundry utilization
* yield
* die size
* packaging
* test
* node economics
* product architecture
* depreciation
* internal transfer pricing
任何一樣做唔到,都可能令gap只收窄少少。
所以我會將今次最合理嘅 one-time conclusion寫成:
> **$50 burden係一個可信嘅 Base-Bull sensitivity。**
>
> **$45係一個較強Bull,但未到Extreme。**
>
> **$30以下先係接近synthetic IDM world-class execution嘅Extreme scenario。**
---
### 對我哋原本forecast最簡單嘅結論
如果 Intel真係逐步由 ~$60 burden改善至:
> **$50**:原本 Base valuation大概有 **10–15%額外上行空間**
>
> **$45**:原本 Bull valuation大概有 **10–20%額外上行空間**
而最明顯嘅impact係 **18–36M**,唔係12M。
所以你之前話:
> 「暫時只當bullish thinking,等results真係見margin改善先入正式forecast」
我完全同意。
但做完今次shadow run之後,我會話:**如果未來Intel連續幾季證明呢條cost curve真係開始落,呢個唔係小修正,係可以令2028–30 EPS結構性高一截嘅第二條大earnings lever。**
INTC - case study - CPU product matching AMD valuation
係,呢個 case study 好值得做。先講我睇完 AMD 最新 valuation 同幾份 analyst SOTP 之後最重要嘅結論:
> **我哋長期用 20–22x 去估 Intel Products,的確比市場目前給 AMD CPU franchise 嘅 implied multiple 保守一截。**
>
> 但 AMD $1T market cap 絕對唔可以理解成「市場俾 CPU 50x」。AMD GPU 已經唔再係純 optionality,市場今日係同時 price 緊 **EPYC + Instinct + future rack-scale AI systems**。
今次只做 **one-time case study**,完全唔改我哋正式 forecast。
---
## 先睇最反直覺嘅比較
AMD 9月21日第一次突破 **$1T market cap**,股價 $615.52。Intel 同日約 **$640B market cap**。([Reuters][1])
但睇最新季度:
| Q2'26 | Intel Products | AMD whole company |
| ----------------------------------- | -------------: | -----------------: |
| Revenue | **$15.14B** | **$11.54B** |
| Segment / non-GAAP operating profit | **$4.82B** | **$3.09B** |
| Operating margin | **31.8%** | **26.8% non-GAAP** |
Intel Products數據係Intel自己10-Q:CCPG $8.88B + DCAI $6.26B,Products OI $4.82B。 AMD Q2 revenue $11.54B、non-GAAP operating income $3.09B。([AMD Newsroom][2])
所以你個直覺完全合理:
> **點解一間目前Products earnings仲大過AMD全公司嘅Intel,market cap反而得AMD大約64%?**
答案主要唔係「今日賺幾多」。
係市場認為:
> **AMD未來earnings增長更快、更乾淨、更容易convert成FCF,而且execution風險低好多。**
---
# 但首先要修正一點:AMD GPU已經唔係「未materialize」
AMD Q2 Data Center revenue已經 $6.7B,EPYC + Instinct共同推動;AMD更明確講 **Instinct sales >2x YoY**。([StockAnalysis.com][3])
而Street而家估得非常進取:
Citi估 AMD AI GPU revenue:
> **2027 $33B**
> **2028 $50.8B**
Wells Fargo更加估:
> **2027 $40.6B**
> **2028 $63B**。([Investing.com][4])
所以AMD $1T入面,好大部分已經係:
> **GPU future earnings capitalization**
而唔係CPU。
呢點同Intel嘅:
14A / EMIB-T / HBM base die
唔完全同級,因為AMD Instinct已經ship緊、revenue已經material,而且有大型announced deployments。
---
# 咁市場究竟俾AMD CPU幾多?
呢度有一份非常有用嘅資料。
Citi今年6月做過 AMD SOTP:
> **Data Center GPU:$281/share**
> **Data Center CPU:$204/share**
> client/gaming/embedded另外計
> net cash約 $35/share。([Investing.com][4])
AMD約1.63B shares,所以單係Citi個:
> **Data Center CPU business = 約 $333B valuation**
而GPU:
> 約 **$458B**。
呢個已經直接證明:
> 市場/analysts並唔係只畀AMD CPU 20x左右。
---
## 可以反推AMD server CPU大概係幾多 PE
Wells Fargo估 AMD server CPU revenue:
> 2026 $16.0B
> 2027 $20.5B
> **2028 $25.0B**。([Investing.com UK][5])
AMD冇披露純EPYC operating margin,所以我哋只能做 sensitivity。
如果2028:
> Revenue = $25B
> operating margin = **35–40%**
> tax ≈15%
咁純server CPU after-tax earnings大概:
> **$7.4–8.5B**
Citi CPU valuation:
> ~$333B
代表大概:
> **39–45x 2028 CPU earnings**
注意:呢個唔係Citi公開話「我用42x PE」,而係我用佢SOTP valuation反推嘅 **implied multiple**。
而且Citi可能本身係用更遠期2030 earnings / TAM做DCF或SOTP,所以唔應將42x視為精確target multiple。
但方向非常清楚:
> **AMD server CPU franchise係用遠高過20–22x嘅valuation framework去估。**
---
# 如果計埋Ryzen,我會用幾多做AMD CPU proxy?
Data Center CPU ~40–45x implied可能偏高,因為:
AMD client Ryzen應該唔值40幾倍;
所以將:
> EPYC + Ryzen
合埋做一個整體 CPU Products franchise,
我會比較舒服用:
> **約30–40x**
>
> 中間值大約 **35x**
做今日 AMD-equivalent CPU multiple。
呢個係我今次 case study 用嘅數。
唔係話AMD市場「官方PE=35x」,而係:
> **用AMD SOTP + current CPU earnings/revenue economics推返出嚟嘅合理proxy。**
---
# 同一個2028 horizon其實仲有個幾有趣結果
AMD而家約$1T。
MarketScreener consensus估:
> 2027 net income ~$21.9B
> 2028 ~$33.5B。([MarketScreener UK][6])
即:
> **2027 PE ≈46x**
> **2028 PE ≈30x**
所以市場其實係look-through 2026,price 2027/28。
Intel consensus 2028 net income反而只係約 **$13.2B**。([MarketScreener][7])
用Intel $640B current market cap:
> consolidated 2028 PE ≈ **48x**
即係有件好有趣嘅事:
> **Intel whole company今日其實唔比AMD便宜。**
真正平嘅係:
> **Intel Products SOTP multiple**
而唔係Intel consolidated PE。
原因係Foundry loss將consolidated earnings食咗好多。
Intel Q2 Products OI:
> **+$4.82B**
但Intel Foundry:
> **-$2.09B** operating loss。
所以市場唔肯直接將Products當AMD咁估。
---
# 點解AMD CPU值得30–40x,而我哋一直只畀Intel 20–22x?
我覺得主要係以下幾件事,而唔係AMD CPU本身神奇好多。
| | AMD | Intel |
| --------------------------- | ------------------ | ---------------------- |
| Manufacturing risk | **TSMC承擔** | Intel自己承擔 |
| Capital intensity | 低 | **非常高** |
| Server CPU share trajectory | 持續gain | 過去持續loss,最近先可能轉折 |
| Product execution | Zen cadence非常穩 | 過去多年delay留下discount |
| Near-term supply | TSM allocation限制 | 自己fab/substrate限制 |
| FCF conversion | 高 | Foundry / CapEx大量吸cash |
| GPU optionality | **已開始materialize** | AI GPU仍較early |
| Foundry optionality | 無 | **Intel獨有巨大upside** |
| Packaging optionality | 較少 | **EMIB-T有額外價值** |
而AMD自己已經公開目標:
> > 50% server CPU revenue share
> > Data Center >60% CAGR
> > AI business >80% CAGR。([Advanced Micro Devices, Inc.][8])
加上Q2 EPYC cloud同enterprise sales都 >70% YoY,市場自然願意提前畀較高multiple。([The Motley Fool][9])
Intel今日雖然CPU demand亦爆緊,但市場仍然問:
> Clearwater/Diamond execution得唔得?
> 18A成本幾時落?
> Foundry幾時breakeven?
> CapEx仲要幾多?
> dilution仲會唔會有?
所以係 **confidence discount**。
---
# 但有一點我覺得我哋20–22x可能真係太保守
Intel Products今日已經係一個:
> ~$60B annualized revenue business
> ~32% operating margin
而且Intel自己internal foundry transfer pricing本身就係想接近market economics,所以Products某程度已經係用「fabless-like」方式去衡量。
如果去到2028:
CPU demand仍然強
* server ASP保持
* Clearwater / Diamond正常
* 18A cost下降
* Products margin改善
* Foundry唔再拖累investor confidence
咁我唔覺得市場永遠只會畀:
> **20–22x**
25–30x其實好合理。
而如果 Intel Products真係做到同AMD CPU franchise類似嘅:
> growth visibility + execution confidence
**30–35x**係完全值得test嘅。
---
# 而家做你要嘅 one-time forecast
### ⚠️ 唔係正式forecast
我用以下假設:
> Intel Products原本 valuation multiple ≈ **22x**
> Partial AMD parity = **30x**
> Full AMD-like CPU parity proxy = **35x**
我哋之前2030 Products model roughly有:
> Base net income ≈ **$28B**
> Bull ≈ **$40–42B**。
即 roughly:
> Base Products EPS ≈ **$5.2**
> Bull Products EPS ≈ **$7.5**
而 Foundry / EMIB / other SOTP **完全唔改**。
因為市場唔會一夜之間由22x升35x,我再假設multiple rerating逐步發生:
> 12M:25%
> 18M:50%
> 24M:75%
> 36M:100%
### One-time AMD multiple parity case
| Horizon | **正式 Base** | **30x Products** | **35x AMD-like** | **正式 Bull** | **30x Products** | **35x AMD-like** |
| ------- | ----------: | ---------------: | ---------------: | ----------: | ---------------: | ---------------: |
| **12M** | $140–170 | **$150–180** | **$157–187** | $190–235 | **$205–250** | **$214–259** |
| **18M** | $170–210 | **$191–231** | **$204–244** | $245–305 | **$275–335** | **$294–354** |
| **24M** | $195–240 | **$226–271** | **$246–291** | $295–370 | **$340–415** | **$368–443** |
| **36M** | $250–310 | **$292–352** | **$318–378** | $390–485 | **$450–545** | **$488–582** |
再強調:
> **右邊兩欄係今日一次性 valuation experiment,唔係我哋新forecast。**
---
# 我覺得真正interesting係24M
假設去到2028年尾:
Intel已經證明:
> CPU shortage唔係一年cycle
> DCAI earnings持續升
> 18A product economics改善
> 14A正常
> Foundry breakeven接近
> EMIB-T開始有真正external revenue
咁問題就會變成:
> **點解Intel Products仍然只值20–22x,而AMD CPU值30–40x?**
如果到時答案只剩:
> 「因為Intel以前做得差」
呢個discount係好容易收窄。
所以我認為最realistic嘅rerating step唔係:
> 22x → 40x
而係:
> **22x → 25x → 28x → 30x**
其實淨係去到 **30x**,24M Base由:
> $195–240
已經可以變成:
> **~$226–271**
而Bull:
> $295–370
變:
> **~$340–415**
完全唔需要Extreme Foundry成功。
---
# 如果真係 full AMD CPU parity 呢?
到36M:
Base:
> **~$318–378**
Bull:
> **~$488–582**
呢個數字睇落非常大,但要記住,我哋唔係話 Intel突然多賺咗好多錢。
只係:
> **同一份 earnings,market由22x改用35x估。**
例如Base:
> $5.2 × 22 = ~$114 Products value/share
> $5.2 × 35 = **~$182**
淨multiple expansion已經:
> **+$68/share**
Bull:
> $7.5 × 22 = $165
> $7.5 × 35 = **$263**
即:
> **接近+$100/share**
呢個就係AMD同Intel今日valuation gap真正有意思嘅地方。
---
# RED TEAM:點解我唔會今日正式改成30x?
因為AMD個premium唔係免費。
AMD證明咗:
> Zen execution
> EPYC share gain
> fabless capital efficiency
> TSMC manufacturing predictability
> FCF
> GPU growth
好多年。
Intel今日只係開始重新證明:
> Products execution正常返。
而且Intel Foundry仲要食大量cash。
所以我會認為:
> **20–22x = 今日合理 execution-discount multiple**
>
> **25–30x = turnaround被證明後合理 range**
>
> **30–35x = Intel Products真正被市場視為AMD級 growth franchise**
>
> **40x+ = 要非常強 CPU supercycle + sustained share/margin execution**
所以我唔會將35x放入Base。
---
## 但今次case study令我有一個幾清楚嘅新觀察
我以前比較集中研究:
> Intel earnings可以升幾多?
但AMD $1T提醒咗一件事:
> **Intel有兩個 rerating lever,而唔係一個。**
第一個:
> **E:earnings上升**
CPU volume / ASP / margins。
第二個:
> **P/E:multiple上升**
由 turnaround CPU company:
> 20–22x
變成 AI infrastructure growth CPU franchise:
> 25–30x甚至更高。
所以:
> **Price = Earnings × Multiple**
兩樣如果同時向上,股價反應會係乘法。
呢個其實可能比我哋之前只focus EPS更加重要。
而我覺得今日 Intel $640B vs AMD $1T 呢個比較真正值得研究嘅唔係:
> 「Intel應唔應該都立即值$1T?」
而係:
> **如果2027–28 Intel Products earnings同execution開始越來越似AMD,市場仲有冇理由畀佢接近50% multiple discount?**
我覺得呢個問題,未來12–24個月會變得愈來愈重要。
**正式forecast維持不變;以上純屬 AMD-multiple parity one-time sensitivity。**
[1]: https://www.reuters.com/business/amd-becomes-latest-chipmaker-reach-1-trillion-valuation-ai-demand-2026-09-21/?utm_source=chatgpt.com "AMD joins $1 trillion club as chipmakers rally on AI-driven demand"
[2]: https://newsroom.amd.com/news/amd-2q-2026-earnings/?utm_source=chatgpt.com "AMD Reports Second Quarter 2026 Earnings - AMD Newsroom"
[3]: https://stockanalysis.com/stocks/amd/transcripts/660937-q2-2026/?utm_source=chatgpt.com "Advanced Micro Devices (AMD) Q2 2026 Earnings Call Transcript & Audio"
[4]: https://www.investing.com/news/stock-market-news/amd-emerging-as-a-legit-second-source-in-the-gpu-market-says-citi-4739340?utm_source=chatgpt.com "AMD “emerging as a legit second source in the GPU market” says Citi By Investing.com"
[5]: https://uk.investing.com/news/stock-market-news/wells-fargo-raises-amd-stock-price-target-on-server-cpu-strength-93CH-4750137?ampMode=1&utm_source=chatgpt.com "Wells Fargo raises AMD stock price target on server CPU strength By Investing.com"
[6]: https://uk.marketscreener.com/quote/stock/AMD-ADVANCED-MICRO-DEVICE-19475876/finances/ "AMD (Advanced Micro Devices): Financial Data Forecasts Estimates and Expectations | AMD | US0079031078 | MarketScreener UK"
[7]: https://www.marketscreener.com/quote/stock/INTEL-CORPORATION-4829/finances/?utm_source=chatgpt.com "Intel Corporation: Financial Data Forecasts Estimates and Expectations | INTC | US4581401001 | MarketScreener"
[8]: https://ir.amd.com/news-events/press-releases/detail/1266/amd-unveils-strategy-to-lead-the-1-trillion-compute-market-and-accelerate-next-phase-of-growth?utm_source=chatgpt.com "AMD Unveils Strategy to Lead the $1 Trillion Compute Market and Accelerate Next Phase of Growth :: Advanced Micro Devices, Inc. (AMD)"
[9]: https://www.fool.com/earnings/call-transcripts/2026/08/11/amd-amd-q2-2026-earnings-call-transcript/?utm_source=chatgpt.com "AMD (AMD) Q2 2026 Earnings Call Transcript | The Motley Fool"
訂閱:
文章 (Atom)