From LCD to AI: Taiwan Display Industry Transformation in the Semiconductor Era
從 LCD 到 AI：台灣面板雙雄如何在半導體時代重新定位自己
- 前半段為文章的英文版本 (The first half is the English version)
- 後半段為中文版本 (The second half is the Mandarin version)
====
From LCD to AI: Taiwan Display Industry Transformation in the Semiconductor Era
AUO and Innolux once fought for leadership in global displays. Now, Taiwan’s two display giants are turning decades of panel-making expertise into a surprising advantage in advanced packaging and AI infrastructure.
Go back two decades, and AUO and Innolux represented one of the most ambitious industrial bets in Taiwan’s technology history.
It was the era when LCD panels were considered one of Taiwan’s “Two Trillion, Two Star” industries.
Taiwan poured enormous amounts of capital, land, equipment, and engineering talent into building a complete display ecosystem — from glass substrates and thin-film transistors to color filters, modules, materials, and production equipment.
AUO and Innolux emerged from that ecosystem as two of the world’s major TFT-LCD manufacturers.
Then the economics of the industry changed.
Backed by massive investment, policy support, and a huge domestic market, Chinese panel makers expanded aggressively. BOE and TCL CSOT gained scale, while the global LCD industry entered a prolonged cycle of oversupply, price pressure, and capacity restructuring.
Scale — once one of Taiwan’s greatest advantages — became increasingly difficult to defend.
But that is only half of the story.
The Taiwan display industry transformation now underway suggests that the industrial capabilities built during the LCD era may not simply disappear. Instead, some of them are being redirected toward semiconductors, advanced packaging, and AI infrastructure.
That leaves AUO and Innolux with a very different question.
For years, the question had been:
How do we manufacture more panels?
Today, it is becoming:
What else can we do with the capabilities we built to manufacture those panels?
Increasingly, the answer is pointing toward the semiconductor era.
The LCD Scale Race Is Fading
Display manufacturing has always been an intensely capital-intensive business.
Larger fabs, newer-generation production lines, and larger glass substrates generally meant better cutting efficiency and lower unit costs.
For years, the industry followed a straightforward logic:
More scale → lower cost → greater market share.
But as Chinese manufacturers entered the market with even larger investments and rapidly expanding capacity, Taiwanese manufacturers gradually lost their advantage in the scale race.
That helps explain why both AUO and Innolux have spent recent years rationalizing capacity, restructuring assets, and moving toward higher-value businesses.
This is not simply about becoming smaller.
It reflects a more fundamental shift in strategy.
The future is no longer about who can manufacture the most LCD panels. It is about who can extract more value from the capabilities they already possess.
And that shift is giving some technologies once associated with a mature industry an unexpected second life.
Innolux: From Display Manufacturing to Semiconductor Packaging
For Innolux, the clearest example is FOPLP — Fan-Out Panel-Level Packaging.
The basic idea is relatively straightforward.
Traditional semiconductor packaging has largely evolved around circular wafers. Panel-level packaging attempts to move parts of that process onto much larger, rectangular substrates.
In theory, rectangular panels can improve area utilization and potentially offer advantages for larger package formats.
But what makes this particularly interesting for Innolux is that the underlying manufacturing challenge is not entirely unfamiliar.
For more than two decades, panel manufacturers have specialized in doing something remarkably difficult:
creating extremely precise structures across very large substrates, consistently and at mass-production scale.
That requires capabilities in thin-film processing, lithography, etching, metallization, automation, and — perhaps most importantly — yield management.
From an industrial-capability perspective, display manufacturing and panel-level semiconductor packaging are therefore not as far apart as they might initially appear.
Innolux is trying to recombine those capabilities.
Instead of using large-area manufacturing expertise only to produce displays, it is applying parts of that knowledge to semiconductor packaging.
By 2026, Innolux’s FOPLP business had begun securing commercial orders.
The scale is not yet large enough to transform the economics of the entire company.
But the significance goes beyond near-term revenue.
It provides an early demonstration that:
capabilities developed for panel manufacturing can cross an industry boundary and become semiconductor manufacturing capabilities.
Innolux is not simply abandoning its past.
It is trying to redefine what that past is worth.
AUO: A Different Route into AI Infrastructure
AUO is taking a somewhat different path.
Rather than placing its primary semiconductor bet on mass-production FOPLP, the company is extending its existing capabilities in glass, optoelectronics, and Micro LED into a broader set of AI infrastructure opportunities.
Two areas are particularly interesting:
glass core substrates and Micro LED optical communication.
They may look like separate technologies.
But strategically, they share the same underlying idea:
take capabilities developed for the display industry and move them into semiconductors and data-center infrastructure.
This is another important dimension of the broader Taiwan display industry transformation.
Instead of competing with Chinese manufacturers primarily through LCD scale, AUO is attempting to move parts of its technology base into markets where its accumulated expertise in materials, glass processing, and optoelectronics can create new value.
When Glass Becomes a Semiconductor Material
Glass is one of AUO’s most deeply rooted industrial capabilities.
For decades, flat-panel displays have relied on glass substrates, giving companies such as AUO extensive experience in large-area glass processing and precision manufacturing.
Now, glass is finding a potentially important new role in semiconductors.
As AI accelerators, HBM, chiplets, and advanced packaging continue to evolve, semiconductor packages are becoming larger and more complex.
And as packages grow, issues such as dimensional stability, warpage, and high-speed signal integrity become increasingly important.
That has made glass an interesting candidate for next-generation semiconductor substrates.
For AUO, this creates an unusual opportunity.
The company can take capabilities originally built around display glass and combine them with technologies such as redistribution layers, or RDL, for advanced semiconductor applications.
What makes this development important is not simply that AUO is working on glass core substrates.
It is that:
a core technology built for the display industry is being redeployed into the semiconductor industry.
When Micro LED Is No Longer Just About Displays
AUO’s other direction may be even more unexpected: Micro LED.
Micro LED has traditionally been discussed as a next-generation display technology.
Higher brightness.
Lower power consumption.
Better contrast.
But AUO is exploring another use entirely:
using Micro LED to move data rather than display images.
That matters because AI data centers are running into a growing problem.
Compute performance can continue to increase, but if GPUs, memory, switches, and other computing nodes cannot exchange massive amounts of data quickly and efficiently, system performance will eventually hit another bottleneck.
That is why optical communication — including technologies associated with CPO, or Co-Packaged Optics — is becoming increasingly important to the future of AI infrastructure.
AUO is exploring how Micro LED technology could participate in this world.
The transformation is striking:
A component designed to produce a display pixel
↓
becomes an optical transmitter
↓
and enters the high-speed interconnect architecture of an AI data center.
The same technological building block enters an entirely different value chain.
This also highlights the difference between the transformation strategies of Taiwan’s two display giants.
Innolux is effectively asking:
Can our panel-manufacturing capabilities become semiconductor-packaging capabilities?
AUO appears to be asking a broader question:
Can our expertise in glass, Micro LED, and optoelectronics become part of the AI infrastructure stack?
Taiwan Display Industry Transformation: Two Companies, Two Strategies
Put the two companies side by side, and two different approaches begin to emerge.
Innolux is primarily pursuing a reinvention of manufacturing capability.
AUO is pursuing something closer to an extension of its technology platform.
One is moving from panels toward packaging.
The other is moving from glass and optoelectronics toward packaging, optical communication, and data-center infrastructure.
Yet beneath those differences, both companies are doing essentially the same thing:
redefining what their core capabilities actually are.
That is what makes the Taiwan display industry transformation more interesting than a simple story about two LCD companies entering semiconductors.
When an industry matures, the obvious question is often:
Does this industry still have a future?
But there may be a better question:
Do the capabilities created by that industry still have another use?
Taiwan’s Underappreciated Industrial Assets
Technology tends to make us think in generations.
PCs.
Smartphones.
Cloud computing.
AI.
Every new technology cycle seems to replace the one that came before it.
But industrial capabilities do not work that way.
Twenty years of process engineering, yield management, materials expertise, precision manufacturing, equipment integration, and mass-production discipline do not disappear simply because demand for a particular end product slows.
The real question is whether those capabilities can be recombined.
Taiwan’s display industry is a good example.
Over three decades, Taiwan built an enormous industrial system around LCD manufacturing.
Glass.
Thin films.
Optics.
Chemicals.
Automation.
Precision equipment.
Large-area manufacturing.
High-yield mass production.
Those capabilities were originally built to manufacture televisions, notebook displays, and monitors.
But the AI era is creating demand for something different:
larger packages,
more complex substrates,
higher-speed optical links,
and new materials.
Suddenly, some capabilities associated with yesterday’s technology cycle are becoming relevant again.
From Legacy Capability to Strategic Asset
If we looked only at the LCD market over the past several years, it would be easy to reach a pessimistic conclusion about Taiwan’s display industry.
Chinese capacity is larger.
Price competition is intense.
Older production lines have lost value.
LCD is no longer a high-growth industry.
Under that interpretation, much of the industrial capability accumulated during Taiwan’s display boom might look like a legacy burden.
AI could change that calculation.
Because AI needs much more than GPUs.
It requires packaging.
Substrates.
Optical communication.
Power.
Cooling.
Networking.
And an enormous amount of physical infrastructure.
Whenever new bottlenecks emerge, industries begin searching for capabilities that may already exist somewhere else.
That creates the possibility that some of Taiwan’s display expertise could shift from being viewed as a mature-industry capability to becoming a new strategic asset.
That does not mean every LCD fab can become a semiconductor fab.
Nor does it mean every display technology will find a second life in AI.
But it does illustrate something important:
Industrial capabilities built for one technology cycle do not necessarily disappear when that cycle ends.
Sometimes, they are simply waiting for another application.
Why Taiwan’s Display Industry Transformation Matters Beyond Displays
The story is bigger than AUO and Innolux.
It reflects something fundamental about Taiwan’s technology ecosystem.
Taiwan’s greatest strength has never been only a single company or a single blockbuster product.
It is the industrial density accumulated over decades of manufacturing.
Semiconductors.
PCBs.
Optics.
Displays.
Electronic components.
Precision machinery.
Materials.
Thermal management.
Power electronics.
Servers.
These industries contain enormous amounts of overlapping technological and manufacturing knowledge.
When a new technology cycle emerges, Taiwan does not always need to build an industrial base from scratch.
It can often recombine capabilities that already exist.
The AI era may make this especially important.
NVIDIA GPUs and TSMC’s advanced semiconductor processes are the most visible pieces of the AI revolution.
But the physical infrastructure supporting AI extends far beyond them.
Advanced packaging.
Cooling.
Power.
Optical communication.
Testing.
Materials.
Substrates.
That means the next group of companies to benefit from AI will not necessarily look like traditional AI companies.
Some may simply discover that:
the capabilities they spent twenty years building have suddenly become more valuable in the AI era.
The Reinvention Has Only Begun
It is still far too early to declare the transformations of AUO and Innolux a success.
Innolux has crossed an initial commercialization threshold in FOPLP, but whether it can scale the business and move toward higher-value applications remains to be proven.
AUO’s glass core substrate and Micro LED optical communication initiatives are also still progressing through technology validation, ecosystem development, and commercialization.
So the most useful question is not:
Can AUO or Innolux become the next TSMC?
That was never the right comparison.
The more interesting question is:
Can companies that built enormous industrial capabilities for the LCD era successfully redeploy those capabilities for the AI era?
Innolux is trying to answer that question through FOPLP.
AUO is answering it through glass, RDL, Micro LED, and optical communication.
We do not yet know how the story will end.
But the Taiwan display industry transformation already points to something important about the country’s technology economy.
Taiwan’s next generation of industrial advantage may not need to be built from scratch.
Some of it is already here.
The AI era may simply be giving us a reason to see its value again.
If you enjoyed this analysis
Follow and Subscribe Taiwan Tech Dispatch on LinkedIn:
👉 https://www.linkedin.com/newsletters/taiwan-tech-dispatch-7355532535025586178
—
Or explore more articles on the blog:
https://whitehsu.blog/
Further Reading
- Taiwan’s AUO showcases glass substrates for AI chip boom
https://t.media/3282622 - Innolux targets advanced packaging with debut of Chip Last tech
https://www.digitimes.com/news/a20260901PD242/packaging-innolux-technology-growth-hpc.html - The Hsinchu Semiconductor Ecosystem: Why Building a Fab Is Much Easier Than Replicating Taiwan’s Technology Miracle
https://whitehsu.blog/2026/07/20/hsinchu-semiconductor-ecosystem/ - From “Computex Is Dying” to the Center of the AI World: How the AI Era Redefined Taiwan’s Value
https://whitehsu.blog/2026/06/22/computex-ai-era-taiwan-transformation/
====
從 LCD 到 AI：台灣面板雙雄如何在半導體時代重新定位自己
友達與群創曾在全球顯示器市場爭奪領先地位。如今，兩家公司正把數十年累積的面板製造能力，轉化為進軍先進封裝與 AI 基礎設施的意外優勢。
如果把時間拉回二十年前，友達與群創代表的是台灣科技產業最具野心的一場戰爭。
那是一個 LCD 面板被稱為「兩兆雙星」的年代。
台灣投入龐大的資本、土地、設備與工程人才，建立從玻璃基板、薄膜電晶體、彩色濾光片，到模組組裝與設備材料的完整顯示器供應鏈。友達與群創，也因此成為全球 TFT-LCD 產業最具代表性的台灣企業。
但後來，產業競爭的邏輯改變了。
中國面板廠在政策支持、資本投入與龐大內需市場支撐下快速擴產，京東方與 TCL 華星光電等企業逐漸取得規模優勢。全球 LCD 產業進入長期供過於求、價格競爭與產能重整。
台灣面板產業曾經最引以為傲的「規模」，開始變成沉重的負擔。
於是，友達與群創面臨的問題也變了。
過去，它們思考的是：
如何製造更多面板？
現在，它們開始思考的是：
過去為了製造面板而建立的能力，還能拿來做什麼更有價值的事情？
而答案，正逐漸指向半導體與 AI 基礎設施。
LCD 規模競賽正在退場
面板製造長期是一個典型的資本密集產業。
更大的廠、更高世代的產線、更大的玻璃基板，通常代表更好的切割效率與更低的單位成本。
這讓面板產業長期遵循一個非常直接的邏輯：
規模越大，成本越低；成本越低，越能搶下市占。
但當中國面板廠以更大的資本投入加入競爭之後，台灣廠商逐漸失去這場規模戰的優勢。
這也是為什麼，近年友達與群創都開始進行產能合理化、資產重整與業務轉型。
這背後不只是「縮小」。
而是兩家公司逐漸接受了一個現實：
未來不再是比誰能做最多面板，而是比誰能把既有能力轉化成更高價值的技術與產品。
這個轉折，非常關鍵。
因為它讓一批過去被視為「舊產業能力」的技術，重新進入新的成長週期。
群創：從面板製造走向半導體封裝
群創目前最明確的方向，是 FOPLP，也就是扇出型面板級封裝。
它的核心概念並不難理解。
傳統半導體封裝，多半圍繞圓形晶圓建立；面板級封裝則試圖把封裝製程搬到更大、而且是方形的基板上。
理論上，方形面板可以提高面積利用率，也更適合處理大尺寸封裝。
但更重要的是，這件事情對群創來說並不陌生。
因為面板廠過去二十多年一直在做的事情，本質上就是：
在大型基板上，以高度精密、高度一致，而且可以量產的方式製作微細結構。
這些能力包括薄膜、曝光、蝕刻、金屬化、自動化，以及最重要的——良率管理。
換句話說，顯示面板製造與面板級半導體封裝，看起來是兩個不同產業，但底層其實共享了相當多製造能力。
這也是群創轉型最有意思的地方。
它並不是把過去二十多年的能力全部丟掉，再重新開始。
而是把原本用來製造顯示器的工業能力，重新組合成半導體封裝能力。
2026 年，群創的 FOPLP 已經開始取得實際商業訂單。
這個規模目前還不足以改變整家公司。
但它至少證明了一件事：
面板製造能力，真的有可能跨越產業邊界。
友達：另一條通往 AI 基礎設施的路
友達走的則是另一條路。
友達並沒有把重心完全押在 FOPLP，而是把既有的玻璃、光電與 Micro LED 技術，延伸到更廣泛的 AI 基礎設施。
其中最值得注意的，是兩個方向：
玻璃核心基板，以及 Micro LED 光通訊。
這兩項技術表面上看起來並不相同，但背後其實有同一個邏輯：
把原本屬於顯示產業的核心能力，帶進半導體與資料中心。
當玻璃從面板材料變成半導體材料
友達最具代表性的能力之一，就是玻璃。
幾十年來，顯示面板一直大量使用玻璃基板，而友達也因此累積了非常深厚的大尺寸玻璃加工與製程經驗。
現在，玻璃正在半導體世界裡出現新的可能性。
隨著 AI 晶片、HBM、Chiplet 與先進封裝持續發展，封裝本身變得越來越大，也越來越複雜。
尺寸越大，材料穩定性、翹曲與高速訊號傳輸，就越成為問題。
這也是為什麼玻璃開始被視為下一世代封裝基板的潛在選項。
友達正在把自己的玻璃加工與 RDL 製程能力帶進這個領域。
這件事情真正有意思的，不只是「友達開始做玻璃核心基板」。
而是：
過去為顯示器建立的核心技術，開始在半導體裡找到新的用途。
當 Micro LED 不再只是用來顯示畫面
另一個更有想像力的方向，是 Micro LED。
過去談 Micro LED，大家想到的通常是下一代顯示器。
更高亮度。
更低功耗。
更高對比。
但友達現在正在嘗試把 Micro LED 用在另一件事上：
傳輸資料。
AI 資料中心正在面臨一個愈來愈明顯的瓶頸。
晶片運算速度可以持續提高，但如果 GPU、記憶體、交換器與其他運算節點之間，無法快速而且低功耗地傳輸大量資料，整個系統仍然會受到限制。
因此，光通訊與 CPO，也就是共封裝光學，逐漸成為 AI 基礎設施的重要方向。
而友達正在嘗試把 Micro LED 帶進這個領域。
這是一個非常有代表性的技術轉換：
原本用來顯示影像的元件
↓
變成光學發射元件
↓
進入 AI 資料中心高速互連
同一項技術，因此進入完全不同的價值鏈。
這也讓友達的轉型路線，和群創呈現出非常不同的風格。
群創比較像是在問：
我們能不能把面板製造能力，轉化成半導體封裝能力？
友達則像是在問：
我們能不能把玻璃、光電與 Micro LED 技術，轉化成 AI 基礎設施的一部分？
兩家公司，兩種轉型方式
如果把兩家公司放在一起看，就會發現它們其實是在做兩種不同的轉型。
群創比較偏向：
製造能力的再利用。
友達則比較偏向：
技術平台的再延伸。
一個從面板走向封裝。
一個從玻璃與光電走向封裝、光通訊與資料中心。
但兩家公司背後，其實都在做同一件事：
重新定義自己的核心能力。
這也正是這場轉型最值得觀察的地方。
因為很多人看到產業衰退時，第一個反應是：
這個產業是不是已經沒有未來？
但真正重要的問題其實是：
這個產業累積的能力，還有沒有下一個用途？
台灣被低估的工業資產
科技產業很容易讓人產生一種錯覺。
我們習慣把世界分成一個又一個世代：
PC。
智慧型手機。
雲端。
AI。
好像每一次新的科技週期出現，上一個時代就會被完全取代。
但真正的工業能力，往往不是這樣運作的。
一家公司花二十年累積的製程能力、良率管理、材料技術、精密製造、設備整合與量產紀律，不會因為某個終端市場開始衰退就突然失去價值。
真正的問題，是這些能力能不能被重新組合。
而台灣面板產業，正是一個很好的例子。
過去三十年，台灣為 LCD 建立了一整套龐大的工業體系：
玻璃。
薄膜。
光學。
化學材料。
自動化。
大面積製造。
精密設備。
高良率量產。
這些能力原本服務的是電視、筆電與顯示器。
但 AI 時代開始需要：
更大的封裝。
更複雜的基板。
更高速的光通訊。
更多新的材料。
於是，一些原本看起來屬於「上一個產業週期」的能力，突然又開始變得重要。
從舊產業能力，變成新的戰略資產
過去幾年，如果只從 LCD 本身來看友達與群創，很容易得到一個悲觀的結論。
中國產能更大。
價格競爭更激烈。
舊世代產線價值下降。
面板不再是高成長產業。
因此，很多人自然會把這些能力視為上一個時代留下來的負擔。
但 AI 可能正在改變這個判斷。
因為 AI 不只需要 GPU。
它還需要封裝、基板、光通訊、電源、散熱、網路與大量實體基礎設施。
而每當一個新的瓶頸出現，產業就會開始尋找可以被重新利用的既有能力。
這時候，台灣過去三十年累積的面板技術，就可能從傳統產業能力，轉變成新的戰略資產。
當然，這不代表所有 LCD 產線都可以改造成半導體產線。
也不代表每一項面板技術，都會在 AI 時代重新獲得價值。
但它至少證明了一件事：
上一個科技週期建立的工業能力，不一定會隨著產業退潮而消失。
有時候，它只是還沒找到下一個應用場景。
台灣的下一個優勢，可能早已存在
這個故事其實不只屬於友達與群創。
它更像是台灣科技產業的一個縮影。
台灣真正強大的地方，從來不只是某一家公司，或某一項明星產品。
而是數十年製造業發展所形成的產業密度。
半導體。
PCB。
光學。
面板。
電子零組件。
精密機械。
材料。
散熱。
電源。
伺服器。
這些產業彼此之間，存在大量重疊的技術與製造能力。
當新的科技週期出現時，台灣的優勢很多時候不是從零開始。
而是把原本已經存在的能力，重新組合。
AI 時代尤其如此。
NVIDIA GPU 與台積電先進製程，當然是最容易被看見的部分。
但真正支撐 AI 基礎設施擴張的，是整個實體科技堆疊。
封裝。
散熱。
電源。
光通訊。
測試。
材料。
基板。
也因此，下一批受益於 AI 的公司，不一定都是傳統定義上的 AI 公司。
它們可能只是突然發現：
自己過去二十年建立的能力，在 AI 時代變得更有價值了。
這場轉型才剛開始
目前還不能說友達與群創的轉型已經成功。
群創的 FOPLP 已經跨過第一個商業化門檻，但能否進一步放大規模、取得更高價值訂單，還需要時間驗證。
友達的玻璃核心基板與 Micro LED 光通訊，也仍然處於技術驗證、生態系建立與商品化推進的階段。
所以，真正值得觀察的問題並不是：
友達與群創會不會變成下一個台積電？
這從來都不是正確的比較。
真正的問題是：
一家曾經為 LCD 時代建立龐大工業能力的公司，能不能成功把這些能力重新部署到 AI 時代？
群創正在用 FOPLP 回答這個問題。
友達正在用玻璃、RDL 與 Micro LED 回答這個問題。
答案還沒有揭曉。
但這場轉型本身，已經透露出一個值得台灣科技產業記住的訊號。
台灣下一個世代的競爭優勢，未必需要從零開始建立。
有些能力，其實早就存在。
只是到了 AI 時代，我們才重新看見它們的價值。
如果你喜歡這篇分析
歡迎訂閱以及追蹤我的 LinkedIn Newsletter – Taiwan Tech Dispatch，掌握即時觀點與產業動態：
👉 https://www.linkedin.com/newsletters/taiwan-tech-dispatch-7355532535025586178
—
更多文章請見：
https://whitehsu.blog/
延伸閱讀
- 友達 2409 : 發展不一樣的CPO與FOPLP
https://vocus.cc/article/6a0abac3fd89780001f8cca6 - 群創成功開發Chip Last、異質整合技術平台 搶攻AI高效運算商機
https://udn.com/news/story/7240/9727450 - 全世界都想複製新竹，但真正困難的從來不是晶圓廠
https://whitehsu.blog/2026/07/20/hsinchu-semiconductor-ecosystem/ - 從「Computex 已經過時」到全球科技朝聖地：AI 時代如何重新定義台灣的價值
https://whitehsu.blog/2026/06/22/computex-ai-era-taiwan-transformation/
