Ming-Chi Kuo

Nvidia預計在GTC 2025宣布採用CPO之Spectrum-X與Quantum-X，這已是市場共識，預期Nvidia或將提供更多細節並強化此共識。 值得關注重點： 1. 每個連接埠速度提升到1.6T。 2. 交換器採用CPO後，有利資料中心節能、維持縮放定律有效性與降低維護成本 (提升恢復與容錯能力)。 3. 參考設計供應商可望因在CPO領域取得領先優勢而得到市場更多關注，舉例如下： - 晶片：台積電 - 測試：宜特 - 光源：Lumentum - 光耦合元件：Coherent - 連接器：Senko - 組裝：鴻海/工業富聯 - 振盪器：SiTime - 光纖：Corning

@G_melo_ding @Jukanlosreve From CES up until recently, I think most people were expecting PTL to hit shelves in 2H25 (that’s the vibe Intel was giving too). https://t.co/TL7dAyoF8g

This news doesn't surprise me. The report (https://t.co/ii9nqxDTsY) aligns perfectly with my previous analysis and projections (see link below), which indicated that Intel's 18A-based PTL would begin low-volume (QS) shipments in 4Q25, initially supporting a limited number of models, and then expand to more models in 1Q26. Intel has consistently stated that PTL will launch in 2H25 (a somewhat vague timeline). Consequently, I think the market consensus will perceive PTL's full ramp-up in 1Q26 as a delay. My previous prediction: Intel's Panther Lake (PTL) Production Delayed, Unfavorable for 2H25 Sales; Patience Key to the Foundry Business https…

看到這樣的新聞我不意外，這篇報導 (https://t.co/ii9nqxDTsY) 完全符合我先前的分析跟預測 (見下方連結)，也就是採用Intel 18A的PTL預計在4Q25小量出貨 (QS) 並支援有限機型，1Q26才會看到大量機型採用。因為Intel一直對外提供的預期是PTL將在2H25出貨 （一個模糊的時程預估)，所以我認為PTL在1Q26放量這件事，對市場共識來說是遞延的。 我先前的預測： Intel Panther Lake量產時間延後，不利2H25銷售動能；晶圓代工業務推廣可能需要更多耐心等待 https://t.co/KSaraG2pCj

Developing great AI services was never going to happen overnight, so the delay of Apple Intelligence is entirely understandable. When Apple announced Apple Intelligence at last year’s WWDC, even though they knew it couldn’t be completed on schedule, this approach—which was not a good move—is still understandable given the pressure from the board and shareholders. I think the worst part is that when it came time to admit that Apple Intelligence (Siri) development wasn’t going as planned, Apple chose to break the news to the world through an unofficial channel. This is how the world’s most valuable company handles a PR crisis. What should App…

開發好的AI服務本來就非一蹴可及，Apple Intelligence延期完全可以理解。在去年WWDC宣布Apple Intelligence時，即便Apple事先已知道這無法如期完成，這種做法雖不可取，但也可以理解是因為礙於董事會跟股東的壓力。 我覺得最糟糕的是，到必須承認Apple Intelligence (Siri) 開發不如預期時，Apple竟然是透過一個非官方管道，來對全世界傳遞此訊息。這就是一個全球市值最大的公司，處理公關危機的方式。 Apple該怎麼做？當年Steve Jobs親上火線面對iPhone 4天線門的公關危機提供了很好的範例。

Since last year's WWDC, I’ve consistently held a cautious view of how much Apple Intelligence would boost iPhone shipments. Take a recent case (see link below), just two months ago, I argued that the market was too bullish on 2025 iPhone shipments and reiterated that Apple Intelligence would contribute less than expected to shipment volumes. Lately, the market consensus has shifted to a more cautious stance on iPhone shipments and Apple Intelligence (Siri), which supports my earlier predictions. Moreover, it’s clear that when Apple shared conservative shipment forecasts with its key suppliers early this year, they had already factored in App…

自去年WWDC以來，我多次重申保守看待Apple Intelligence對iPhone出貨的貢獻。舉一篇最近的例子 (見下方的連結），我在2個月前，提出市場共識對2025年iPhone出貨過於樂觀的看法，並重申Apple Intelligence對出貨幫助低於預期。 近期市場共識開始保守看待iPhone出貨量與Apple Intelligence (Siri)，驗證了我先前的預測。此外，很顯然的，Apple在年初對關鍵供應商提出iPhone出貨保守預期的同時，在當時肯定也考慮了Apple Intelligence不如預期這件事。

Three months ago, I predicted that mass production of the new display-equipped HomePod would slip from 1Q25 to 3Q25 (after WWDC) due to software development issues. Beyond Apple Intelligence, the issues also involve ensuring the HomePod’s interface aligns with new OS updates (like iOS 19) set to roll out across other devices in 2H25. This alignment is essential for seamlessly integrating new products into Apple’s ecosystem.

我在3個月前預測新款配備螢幕HomePod量產由1Q25延後到3Q25 (WWDC之後)，原因是因為軟體開發。除Apple Intelligence外，軟體開發問題也包括要確保新款HomePod的介面設計，要與2H25其他裝置的新作業系統 (如iOS 19) 一致，這對整合新產品至Apple生態至關緊要。

我最新的產業調查指出，小米預計在2025年底發售的新款旗艦機型Xiaomi 16 Pro，也將採用鉑力特用3D列印製造的金屬中框，意味著用3D列印生產手機中框的趨勢可能正加速發生中。用3D列印生產手機中框的好處在於，利用鏤空的設計，可以在不犧牲中框強度下，同時減輕重量並提升散熱效能。 生產效率一直都是3D列印的挑戰，但在此問題持續改善下，當採用3D列印製造的中框所帶來的好處，大於因生產效率帶來的成本時，手機品牌業者就會開始採用，無需等到生產效率改善至與主流技術相當。這樣的道理並非無先例可循：在Apple用CNC製造MacBook一體成型金屬機殼前，沒有人相信CNC可以應用在消費電子產品。

My latest industry survey shows that Xiaomi’s upcoming flagship, the Xiaomi 16 Pro, set to launch in late 2025, is also adopting a metal mid-frame produced with Bright Laser Technologies (BLT)’s 3D printing technology. This points to an accelerating trend of using 3D printing for smartphone mid-frames. The advantage of producing smartphone mid-frames with 3D printing lies in its ability to leverage hollow designs, which reduce weight and enhance thermal performance without compromising structural strength. Production efficiency has historically been a hurdle for 3D printing, but as this bottleneck continues to ease, phone manufacturers will …

SiTime: Exclusive Supplier of MEMS Oscillators for Apple’s In-House Modem, Set for Robust Growth in Coming Years Driven by New Orders Click to read https://t.co/O06xIVo2fN https://t.co/zDFjGSQ5Ng

SiTime為Apple自家數據機MEMS振盪器獨家供應商，新訂單為未來數年強勁成長驅動 點擊閱讀 https://t.co/6I7f5amYvi https://t.co/ZPfqpFGdTO

Apple's C1 modem process technology: - Baseband: 4/5nm (both technologies are similar) - Low-frequency/Sub-6 TRx (Transceivers): 7nm - Intermediate Frequency (IF) TRx: 7nm - PMIC: 55nm The C1 refreshed version is under development for mass production next year, aiming to improve power consumption and transmission speed and support for mmWave (including TRx and front-end components, built on a 28nm process). While supporting mmWave isn't particularly challenging, achieving stable performance with low power consumption remains a key hurdle. Unlike processors/GPUs, baseband chips don't aggressively adopt the latest advanced node because the re…

Apple C1數據機的製程技術： - 基帶：4/5nm (兩者技術相似) - 低頻/Sub-6 TRx (收發器)：7nm - 中頻 (Intermediate Frequency；IF) TRx：7nm - PMIC：55nm 預計明年量產的C1升級版開發中，目標是改善耗電跟傳輸速度，以及支援毫米波 (TRx，包括前端，採用28nm)。支援毫米波不難，但要能穩定運作且維持低耗電並非易事。 數據機與處理器/GPU不同，不會積極採用最新的製程技術，因為投資報酬率不高。故預期至少明年還不會見到Apple基帶採用3nm製程。 - 採用先進製程對數據機的傳輸速度無太大幫助。 - 先進製程可改善數據機耗電，但需注意基帶並非手機無線系統中最耗電的零組件。

我最新的產業調查指出，鉑力特為折疊iPhone的領先與關鍵受益者。鉑力特為折疊iPhone的軸承蓋與中框NPI (兩者都是鈦合金) 供應商。因軸承蓋要求薄，且中框要求鏤空故設計複雜，故兩者都需要用到3D列印技術。 目前鉑力特製造的軸承蓋與中框樣品的單價共約800-900人民幣，量產後因良率改善預計會降至400-600人民幣。若假設初代iPhone出貨量為2,000萬部，且鉑力特供應比重達80%，則初代折疊iPhone在2026–2027年將貢獻鉑力特營收約80億人民幣。鉑力特2024年營收約12億人民幣，故預期鉑力特自2026年開始，營收將強勁成長。

My latest industry survey indicates that Bright Laser Technologies (BLT) is a leading supplier and key beneficiary in producing the foldable iPhone’s hinge cover and middle frame NPI, both crafted from titanium alloy. The hinge cover is required to be very thin, while the middle frame requires hollow-out, resulting in a complex design that requires 3D printing technology for both components. Currently, the combined unit price for the hinge cover and middle frame samples produced by BLT is around 800-900 RMB. Once mass production ramps up and yields improve, this price is estimated to drop to between 400 and 600 RMB. Assuming first-generation…

Intel's Panther Lake (PTL) Production Delayed, Unfavorable for 2H25 Sales; Patience Key to the Foundry Business Click to read https://t.co/PHmxOvUt2u https://t.co/s0Pwb0aPWg

Intel Panther Lake量產時間延後，不利2H25銷售動能；晶圓代工業務推廣可能需要更多耐心等待 點擊閱讀 https://t.co/t4uuSpLuTI https://t.co/XPILYPQdsb

My understanding is that TSMC always takes a highly cautious approach to capacity expansion. Unless there's a dramatic industry reversal in the short term, it's not easy to see significant order reductions (variations in the low to mid-single digits are considered reasonable, as no one can predict future demand with pinpoint accuracy.) Last year, at the height of AI and Nvidia market optimism, Nvidia pushed TSMC to expand its monthly CoWoS capacity to 100,000 wafers—a request that TSMC rejected. Incidentally, using this as an example, if outsiders had obtained Nvidia's optimistic request—a common practice when customers push for expansion du…

我的理解是，台積電對產能擴張一直非常謹慎，除非在短期內產業趨勢急劇反轉，否則要看到顯著砍單並非易事 (低至中個位數的變化都算合理範圍，沒人能那麼精準預估未來需求)。去年AI/Nvidia市場氣氛最熱時，Nvidia曾提出樂觀要求，希望台積電每個月CoWoS產能能擴張到10萬片，被台積電拒絕了。 順帶一提，以此為例，如果上述Nvidia的樂觀要求流出被外人知道 (實務上，客戶常常在市場氣氛熱絡時提出擴產要求），可能就會有部分市場參與者，以此流出訊息樂觀估算台積電產能或Nvidia的訂單，並成為市場傳言的素材。

Yes, my take on 18A mostly comes from upstream semiconductor survey & research, not PC/NB ODM/EMS.

是的，我對18A的理解，主要是來自對半導體上游的調查與研究，而非PC/NB ODM/EMS。

In 2026, Apple's new Houston, USA factory will begin producing AI servers powered by the high-end M5 processor. Apple announced on Feb 24, 2025, that it will invest over $500 billion in the U.S. over the next four years, including opening a Houston factory in 2026 to build AI servers for Apple Intelligence Private Cloud Compute (PCC). In my earlier Apple M5 predictions, I wrote, "Apple's PCC infrastructure build-out will accelerate after the mass production of the high-end M5 chips." The mass production of the high-end M5 chip is slated for 2H25-2026, aligning with Apple's timeline for expanding AI server capacity.

Apple在美國Houston於2026年開張的新工廠將生產配備M5高階處理器的AI伺服器。 Apple於2025年2月24日公告，預計未來四年將在美國花費超過5,000億美元。當中提到2026年將會在Houston開設新的工廠，生產用於Apple Intelligence Private Cloud Compute (PCC) 的AI伺服器。 我先前關於Apple M5的預測中，提到「Apple PCC基礎建設建置，將隨高階M5晶片量產後加速」，M5高階晶片的量產時間為2H25-2026，這與Apple公告加速AI伺服器產能擴張的時間點一致。

The first Panther Lake engineering samples, made with Intel/IFS’s 18A, are currently being tested by major PC ODM/EMS makers. My early 2025 industry survey showed 18A yields below 20-30%, so there’s still a lot of room to step up—which doesn’t bode well for Intel’s goal of hitting mass production in 2H25. On top of the tech challenges, IFS faces a big obstacle in winning outside orders due to its org setup, supply chain mgt, and culture. That’s where TSMC totally stands out.

首批Intel/IFS 18A先進製程生產的Panther Lake工程樣品正由主要PC ODM/EMS廠測試中。我的產業調查指出，Panther Lake的生產良率仍有很大的改善空間 （2025年初不到20-30%)，這對下半年要量產的目標而言並非易事。與生產技術相較，IFS欲取得外部客戶訂單的另一大挑戰是組織/供應鏈管理與文化，這也是台積電獨特的競爭優勢之一。

Following Qualcomm, Broadcom's Wi-Fi chips will also be replaced by Apple's in-house chips at a faster pace. My latest industry survey indicates that all new 2H25 iPhone 17 models will feature Apple's in-house Wi-Fi chips (vs. only the slim iPhone 17 will adopt Apple's C1 modem chip). Beyond cost reduction, the switch to in-house Wi-Fi chips will enhance connectivity across Apple devices.

繼Qualcomm後，Broadcom的Wi-Fi晶片也將被Apple自製晶片取代，且被取代速度更快。我最新的產業調查顯示，2H25所有新款iPhone 17機型將採用Apple自製的Wi-Fi晶片 (相較於2H25新款iPhone 17中，僅超薄iPhone 17會採用Apple C1數據機)。Apple改用自家Wi-Fi晶片，除可降低成本外，未來有助於強化自家產品互聯的用戶體驗。