News Commentary: Jen-Hsun Huang says nVIDIA's purchase of Intel shares has nothing to do with Trump, do you believe me? Do you believe it?
Author: Mr. Lin Weizhi, Executive Vice President, Ji-Pu Industrial Trend Research Institute
nVIDIA, the leader in AI chips, announced on September 18th that it has invested $23.28 per share in Intel's common stock (it was not specified whether this is an issuance of new shares or an acquisition of existing shares). According to the announcement, Intel will focus on building custom x86 architecture CPUs for nVIDIA, and the two companies plan to co-develop multiple generations of custom data centers and PCs to accelerate hyperscale, enterprise, and consumer applications and workloads. nVIDIA also emphasized that the collaboration will focus on leveraging NVLink to seamlessly connect the two organizations, combining its strengths in AI and accelerated computing with Intel's leading edge in AI and accelerated computing. nVIDIA also emphasized that the partnership will focus on utilizing NVLink to seamlessly connect the two organizations, combining its strengths in AI and accelerated computing with Intel's leading CPU technologies and x86 ecosystems to provide customers with more complete, cutting edge solutions.
Looking back at Intel's news in the last half year, from "TSMC is rumored to be asked to take over Intel", "U.S. government invests in Intel" to market rumors of being asked to accelerate the split, and so on. My opinion
對 Intel 而言,軟銀(20 億美元)、美國政府(約 90 億美元)以及 nVIDIA(50 億美元)的資金挹注,以目前虧損規模推算,大致能為其「續命」兩年左右,撐到 1.4 奈米進入量產時程,算是解決燃眉之急。然而,真正的關鍵在於雙方合作方式。首先,在 AI Server領域,x86 CPU(Intel 或 AMD)搭配 nVIDIA GPU 是過去主流組合;近年 nVIDIA 逐步推出自家 Grace + Blackwell系統架構,並規劃未來 Vera ARM + Rubin 架構。Rubin 之後是否會重回 Intel x86 + 次世代 GPU 的組合,仍有變數但我認為機率不高。相對地,針對中國市場的降規版 B30a(暫名,H20的下一代GPU)搭配 Intel x86 CPU 的方案,則有更高的可行性,並可能基於 Intel CPU的強大IP 與 nVIDIA GPU 特性做深度客製,且由 nVIDIA主導以自家名義向 IFS 下單試產,藉此協助 Intel 製造能力爬升。在 PC 領域,由於這是 Intel 的本命區域,且市場對Intel 高階 CPU(如 Arrow Lake、Panther Lake等)仍高度期待,我推測雙方將聚焦在中低階 AI PC CPU 的合作,由 nVIDIA 協助 Intel 在設計階段導入 AI 加速特性。接著,隨邊緣運算與小型工作站需求大幅成長,Intel x86 CPU + nVIDIA RTX GPU 的 SoC 預期將能創造市場號召力,有望成為另一個重要合作場景。總結來說,無論是資料中心、AI PC 或邊緣運算的領域,這些合作很可能都會以 nVIDIA 名義在 IFS 進行投片試產。因為這次合作背後的核心命題,正是如何讓 Intel 的製造業務重新站穩腳跟,取得市場認同。
最後,也是大家最關心的問題:這次合作對台積電的影響。目前雙方的實際合作細節仍未完全明朗,但以 Intel 近幾年的製造表現,很難說它短期內就能追上台積電。然而,若以前述假設與推論來看,這次合作更像是一種「把餅做大」的策略。在未來兩年內,Intel IFS 要生產出真正具有競爭力的晶片仍相當困難(效能強、良率高且量產時程趕得上客戶產品發表),因此更合理的情境是:雙方共同設計的晶片將由 IFS 與台積電同步生產,其中部分 SoC 的Dies 仍可能全權委外給台積電代工(會有哪些,得要看當時IFS的表現)。從這角度出發,短期內這會是一個美國政府、nVIDIA、Intel 與台積電都能受惠的「四贏」局面。但如我先前報告所提,如果 IFS 未來真的壯大,並且成功搶下台積電重要客戶,迫使台積電在先進製程持續進行高額資本支出、導致報酬率下降,那才是我們真正需要警惕的時刻[可參閱:英特爾或將取消代工客戶的 18A 製程節點,使台積電幾乎無可匹敵]。






