2024/10/17

TSMC 3Q24 earnings thought










GPU, another IOT?
TSMC claims working with all the customers so that we know the demand is true. TSMC seeing the demand is true due to a lot of customers are coming TSMC asking for capacity.


this is by far the most persuasive methodology that saying every customer are working with them, and are having the broadest and deepest view.


However, seems no one had remembered that IOT was also described as an hot topic, but no one cares anymore. now only accounts 7%.


What will happen if all the demand of GPU 1) crushed, or disappear, 2) commoditized, if we dont need anything that is as big die size or can be replaced by things that are using right now, such as current smartphone AP capacity that replaced the GPU demand, might see a lot of capacity redundant eventually.


long-term CAGR refuse to guide, I assume to be below 15-20%, as 2024 y/y are too strong already, the growth might not sustain.

AMD are guiding CAGR to be 70+% y/y in 2023-2027.



Client's pressure forming a margin cap?
Client margin, 75% from NVDA, TSMC gross margin improved from 53.2% to 57.8%.


2024/10/2

Glass Fiber Shortage











Taiwan glass shortage
LDK glass
Taiwan glass
Nittobo
Asahi fiber glass



玻纖布方面也主攻高階產品低介電（Low DK）技術，為全球少數供應商之一，下半年目標全球市占逼近30％。


台玻
- Starting from July, monthly revenue are still showing very moderate y/y growth. No significant increase.
- Low DK product already being qualified, shipping to US big customers.
- But why the revenue data did not increase accordingly?
- 大陸玻璃純鹼期貨price rally sharply.

- 林伯豐指出，目前台玻在高階玻纖布產品開發上，已研發出第一代低介電玻纖布的DK值在10GHz下有4.58，第二代DK值4.3的產品也已完成，並持續取得國際終端大廠認證採用。

他表示，目前全球能供應高階Low DK玻纖布僅三家，包括日本日東紡、美國AGY，以及台玻；目前集團市占率約兩成，下半年在產品放量出貨下，有望挑戰三成市占，集團獲利狀況將同步獲得改善，公司將持續努力、提升獲利能見度。

據傳，台玻Low DK玻纖布已獲得銅箔基板（CCL）大廠台光電認證，並應用在5G行動通訊基礎建設、高階AI伺服器、車載裝置、高階顯卡等上。
- 不過台玻強調，長期而言，認為要到2025、2026年，整體玻璃纖維布銷售狀況才有望恢復正常水準。
- 台光電在AI伺服器的玻纖布應用中，台玻供貨比重約在25~35%






Nittobo's point of view
- Strong sales of Special Glass yarn and cloth were driven by continued healthy demand for
data center applications. Sales and profit increased (Nittobo, Aug 2024, upward revision).
- we believe that supply chain inventory build-up for T-glass was concentrated in the first quarter, and that demand will be adjusted to the actual level in the second quarter (Nittobo, Aug 2024, upward revision).

- Constructed new Special Glass factory in Taiwan
- NER needs will increase as high-end equipment in data centers begin to switch from 400GbE to 800GbE.

- industrial demand has no sign of recovery, might be a potential upcoming catalyst once every industrial demand comes back, echoing MCU suppliers'(STMicro, Infineon) view.

- 2020 Jul big fire interrupting the supply. 2019 Glass fiber shortage.


建榮 5340

- revenue significantly increase starting March.
- stock price rally in  May up to $6x, but not sustaining and fall back to $45.
- Nittobo price didnt increase much.
- AI伺服器需求上來時公司產品尚未獲得取得第一波認證，未能第一時間跟上趨勢，直到去年第4季客戶端完成相關認證才開始銷售, 公司本身也希望今年下半年能搭上趨勢。
- 今日建設工程順利完成，對於日東紡集團的營運具有重大戰略意義，未來新廠生產之低介電玻纖紗(Low DK)除供應國內廠商外，並將透過另一在臺投資公司建榮工業材料進行技術轉移
- Low Dk(NE glass)系列製品開發，其主要高頻高速低損耗應用，用於AI伺服器等產品。下一代玻璃纖維布方面，建榮NE Glass已經開始生產，T Glass仍在準備階段


富喬 1815

- revenue significantly increase starting March.
- stock price sharply increase, but with low volume.


南亞 1303
- technology might falling behind.
- stock price sharply increase.


德宏 5475
-monthly revenue bad




customer in shortage
AMZN
Tesla/ Starlink
- order being cancelled might be due to the small volume.


Who made the PCB for AI servers?
日東紡的Low-Dk玻纖布在2019年就曾出現缺貨狀況，當時最大苦主便是台燿
目前台系三雄對於日東紡的仰賴程度，台燿仍然比重最高，其次是聯茂，台光電則使用其他日系業者的材料
CCL轉嫁成本的能力通常並沒有IC載板那麼高，對獲利肯定會帶來壓力
台系CCL(銅箔基板)三雄台燿、台光電、聯茂

低熱膨脹(Low-CTE)玻纖布更是大量使用在高階的ABF載板中，主要由日東紡提供玻纖布給日立化成(Hitachi Chemical)，生產成載板用CCL材料後再交給載板廠

高階供給更加吃緊，也會連帶衝擊ABF載板的供應，恐導致市場ABF載板供不應求態勢更加嚴重，加上市場傳出，ABF載板廠普遍也有持續調高報價的企圖心，受此事件影響，恐怕讓ABF載板報價有持續向上走揚的可能性。

2024/9/23

Qualcomm buying intel scenario analysis

WSJ discussing about QCOM might be interested to buy Intel, details TBD.

  • To Buy
    • Gain share in PC.
    • customer and business.
    • Own foundry.
    • QCOM aiming for the hyperscale server GPU, aiming at server market. Dominating all the server CPU (90%+) market share, which they never had chance to get in.
    • Creating new narratives/ story on AI ecosystem.

  • Not to Buy
    • No need to gain share in PC, as QCOM can get into the market from AI PC sector. Even AI PC QCOM are not very successful.
    • Even Intel itself are struggling on the foundry technology development.
    • Also Qualcomm are using both TSM and Samsung as foundry. Pretty stabilize on TSM's outsource. However, QCOM had faced certain issues 2 years ago when using Samsung foundry, that had given MTK's dimensity some opportunities back then. Also QUALCOMM are suing Samsung's business in order to get the socket in Samsung Galaxy flagship, which has no meaning to buy an new foundry.
    • Intel modem business in iPhone failed. QCOM are not aiming at the communication business.
    • AI business trend might flow to the "low power" in the near-term, which makes buying intel less urgent.

  • Status
    • Anti-trust hardly pass in China according to privous experience.
    • Intel wants to work on GPU. while are not successful neither.
    • Timing is strange that the stock price might get lower going forward.
    • Intel book value ~$27.
    • QCOM GM 55%.
    • Intel GM 3x-4x%

  • Synergies
    • Cost saving, similar business as IC design sector, can cut many redundant position, and outsource chip making to TSM.
    • broaden customer base, revenue per customer increase.
    • Opportunity to break into AI business.

  • Financial modeling
    • TBD
    • adding chart to the title and financial chart.
    • Cathay line: mobile eye and Automotive, wait Intel foundry to spin-off.






source:
https://www.wsj.com/business/deals/qualcomm-approached-intel-about-a-takeover-in-recent-days-fa114f9d





訂閱： 意見 (Atom)