除了市場太小，現在多了一個理由
講到台灣軟體新創為什麼要出海，過去大家最常講的就是一句話：台灣市場太小。這句話沒錯，但講了十幾年，大家也聽膩了。
這兩年我看下來，其實多了一個新的因素，而且我覺得它比市場大小更直接影響一家新創的命運，那就是 IPO 市場。
我要先講清楚，我不是說台灣的資本市場對軟體業不友善。真正的問題是，台灣有一個更厲害的產業聚落，也就是 AI 硬體跟半導體，把所有的關注度跟錢都吸走了。
到了 A 輪，投資人開始想出場
當新創的人應該都有這種感覺。早期募資的時候，投資人覺得你做的東西夠有意思、團隊夠強，就有可能投你。
但等到你要擴張，甚至走到 A 輪，投資人的問題就變了。他們開始在腦中想像：這個案子最後要怎麼出場？誰會買？在哪裡上市？上市的時候值多少？
這不是投資人太現實，是基金的結構本來就是這樣。後期投資人進場的價格，取決於他們對出場估值的想像。如果出場的天花板很低，中間每一輪的估值就會被往下壓，最後一路傳導回最早期的投資人。早期投資人如果覺得後面無利可圖，投資意願自然就會降低。
台股的 AI 信仰，在硬體
那出場的想像從哪裡來？從 track record 來。
台灣的投資人，不管是 private market 還是 public market，這幾年已經被 AI 硬體跟半導體反覆證明過：這東西能賺錢。看數字就很清楚。2026 年 4 月中，台股總市值突破 115 兆元，前 20 大公司就占了將近 70%，其中 12 家直接跟 AI 伺服器、晶片、散熱、封測相關，台積電一家就貢獻了 44.8%（Taiwan News，2026/4/19）。
IPO 市場也是一樣的故事。我把 2025 年的數字整理一下（經濟日報，2025/12/28）：
- 上市櫃合計：70 家，籌資新台幣 1,050.67 億元
- 其中半導體：11 家，籌資 475 億元，約占 45%
- 上櫃市場合計：39 家，籌資 205.22 億元
- 其中數位雲端：7 家，是上櫃家數最多的產業，籌資額未單獨揭露
這組數字最有意思的地方是：軟體公司不是上不了市，數位雲端在上櫃市場的家數甚至是第一名。但錢呢？上櫃市場籌資前兩名都是半導體，兩家加起來就拿走 42.92%。整個上櫃市場一年的籌資額，還不到半導體全年的一半。
反過來看 AI 軟體，台灣市場對所謂「AI 原生」新創的看法跟估值，老實說還沒到非常關注的程度。軟體在台股沒有過充分的 track record，大家就不會有信仰；沒有信仰，給的估值倍數自然就上不去。
一個會自我強化的死循環
所以我相信，如果現在有一票公司同時要 IPO，多數的錢還是會被 AI 硬體吸走。資金被吸走，你的關注度就降低；關注度低，估值就不會太高。
然後這件事會一路往回傳：
- 軟體沒有 track record，投資人沒有信仰
- 沒有信仰，IPO 估值給不高
- 估值不高，好的軟體公司就不會選在台灣掛牌
- 好公司不在台灣掛牌，台灣就永遠累積不到軟體的 track record
繞了一圈又回到原點。我想強調的是，這不是哪個環節的人做錯了什麼，而是結構本身會自我強化，單一一家新創沒辦法靠自己打破它。
在台灣做軟體，其實只有兩條路
這兩年觀察下來，我認為如果你真的要做軟體類的創業，可以先問自己要走哪一條路。
第一條，在台灣做一家好的中小企業。我相信你可以賺到錢。但要講清楚，這裡的賺錢是「公司自己賺錢」，不一定是「投資人賺錢」。公司有獲利、有現金流，創辦人跟團隊過得很好，可是沒有一個好的出場，投資人的錢就出不來。所以走這條路，就不要去做需要創投資金的那種創業。這完全是一個可以考慮、也很健康的選擇。
第二條，跟資本市場接軌。一旦你拿了創投的錢，你就同時答應了投資人一個出場。這時候，去選一個更大的市場，才是比較務實的做法。
大市場也不只是客戶多。除了過去大家常講的台灣市場太小，現在連資本市場這一端也在告訴你同一件事：國際化是必要的。
美國 IPO 很難，但併購是更務實的出口
現在對 AI 軟體新創比較看得懂的，我認為還是美國市場。Carta 在 2026 年 7 月整理了一千多輪軟體新創的募資，A 輪估值中位數是 US$80M，而且其中至少一半可以算是 AI 原生公司。這種估值水位，在台灣的 private market 幾乎是看不到的。
但我也要老實講：一個國際團隊要進軍美國，一路做到 IPO，這個挑戰度是很大的。
所以我反而覺得，去挑戰大市場，真正的機會在併購。在 AI 浪潮之下，這幾年併購的數量明顯比往常多：
- 美國新創 2025 年 Q4 被併購 232 家，史上最高，比 2023 年 Q4 多 60%（Carta，2026/1/5）
- 其中 AI 新創占比超過 40%
- 其中 pre-seed／seed 階段占將近一半
- 全球 AI 併購 2025 年 Q2 單季 177 件，約為 2020 年以來季平均的兩倍（FE International）
有一點要誠實面對：被併購的公司裡，將近一半是很早期的公司，很多其實是 acquihire，大公司在搶人才。但這對我的論點反而是加分。買家要的是能直接接上美國業務的團隊，你人不在大市場、客戶不在大市場，連被併購的門票都拿不到。
所以選一個比較大的市場，然後在未來幾年內被併購，可能會是比留在台灣更好的出場選擇。
要拿創投的錢，就得出海
把這些因素放在一起看，我的結論是：需要創投資金、要跟資本市場接軌的台灣 AI 軟體新創，必然得選擇出海。
這不是不愛台灣，也不是唱衰台灣的資本市場。台灣的資本市場很好，只是它的信仰現在放在硬體上，而且放得很有道理。身為一個軟體創辦人，你要做的是看清楚這件事，然後在第一天就決定自己要走哪一條路。
我比較喜歡的版本是：市場跟資本出海，團隊留在台灣。台灣有很好的工程人才跟成本優勢，這是我們的底氣；但客戶、投資人跟出場，就去它們真正在的地方找。
一起把出海這條路走通
如果你正在做 AI 軟體、也正在評估要不要出海，或者已經在美國市場摸索卡住了，歡迎來跟我們聊聊。除了投資之外，也許可以把我們看到的坑跟門票，先講給你聽。
有興趣的話，歡迎透過 contact@addin.vc 與我們聯繫。
本文同步刊於 ntu.vc。
有興趣的話，歡迎透過 contact@ntu.vc 與我們聯繫。
📄 資料來源
- 〈時評〉台積電領軍 台灣邁向全球高科技核心｜Taiwan News，2026/4/19
- 台灣 IPO 市場今年再創高！半導體領軍 上市櫃籌資突破千億元｜經濟日報，2025/12/28
- Q4 2025 Was the Hottest Startup M&A Quarter Ever｜Carta，2026/1/5
- VC Startup Fundraising Benchmarks From 1000 Rounds｜Carta，2026/7/10
- AI M&A Trend｜FE International
Why Taiwan's AI Software Startups Have No Choice but to Go Global
Beyond "the market is too small," there's now a second reason
Ask why Taiwanese software startups should go abroad, and for over a decade the answer has been one line: the domestic market is too small. True, but everyone has heard it to death.
Over the past two years I've seen a second factor emerge, and I think it shapes a startup's fate more directly than market size does: the IPO market.
Let me be clear up front. I'm not saying Taiwan's capital markets are hostile to software. The real issue is that Taiwan has a far stronger industry cluster, AI hardware and semiconductors, that soaks up all the attention and all the money.
By Series A, investors start thinking about the exit
Anyone who has run a startup knows the feeling. In the early rounds, if what you're building is interesting enough and the team is strong enough, investors may back you.
But once you're scaling, especially at Series A, the question changes. Investors start picturing the ending: How does this company exit? Who buys it? Where does it list? What is it worth when it does?
That isn't investors being cold. It's how fund structures work. The price a later-stage investor pays depends on the exit valuation they can imagine. If the exit ceiling is low, every round in between gets pushed down, and the pressure travels all the way back to the earliest investors. If early investors see no upside downstream, their appetite drops.
Taiwan's AI faith lives in hardware
So where does an exit story come from? From track record.
Taiwanese investors, in both private and public markets, have spent the past few years being shown again and again that AI hardware and semiconductors make money. The numbers are plain. In mid-April 2026, Taiwan's total market capitalization crossed NT$115 trillion. The top 20 companies made up nearly 70% of it, 12 of them directly tied to AI servers, chips, cooling, or packaging and testing, and TSMC alone accounted for 44.8% (Taiwan News, April 19, 2026).
The IPO market tells the same story. Here are the 2025 figures (Economic Daily News, December 28, 2025):
- Total listings on TWSE and TPEx: 70 companies, raising NT$105.07 billion
- Semiconductors: 11 companies, raising NT$47.5 billion, roughly 45% of the total
- TPEx (the OTC board) overall: 39 companies, raising NT$20.52 billion
- Digital and cloud: 7 companies, the most of any sector on TPEx, with no separate fundraising figure disclosed
The most interesting part: software companies are not being kept off the market. Digital and cloud was the number one sector by listings on TPEx. But the money? The two largest TPEx raises were both semiconductor companies, and together they took 42.92% of the board's total. The entire OTC board raised less in a year than semiconductors did on their own.
Flip to AI software, and Taiwan's market honestly hasn't paid much attention to so-called AI-native startups or how to value them. Without a sufficient software track record on the exchange, there's no faith. Without faith, valuation multiples stay low.
A loop that reinforces itself
So I believe that if a wave of companies went public at the same time today, most of the money would still flow to AI hardware. When capital gets pulled away, attention drops. When attention drops, valuations stay low.
And it cascades backward:
- No software track record, so investors have no faith
- No faith, so IPO valuations stay low
- Low valuations, so good software companies don't list in Taiwan
- Good companies don't list in Taiwan, so Taiwan never builds a software track record
Full circle. My point is that nobody in this chain did anything wrong. The structure reinforces itself, and no single startup can break it alone.
Building software in Taiwan, you really have two paths
Watching this unfold, I'd say anyone starting a software company should first decide which path they're on.
Path one: build a good small or medium business in Taiwan. I believe you can make money. But to be precise, that's the company making money, not necessarily the investors. The business is profitable, cash flows, the founders and team live well, but without a real exit, investors' money never comes back out. So if this is your path, don't build the kind of company that needs venture capital. It's a perfectly valid, perfectly healthy choice.
Path two: plug into the capital markets. The moment you take VC money, you've promised your investors an exit. At that point, choosing a bigger market is simply the pragmatic move.
And "bigger market" isn't only about more customers. Beyond the old argument that Taiwan is too small, the capital-markets side is now telling you the same thing: going international is necessary.
A U.S. IPO is hard, but M&A is the realistic exit
For AI software startups, the market that understands them best is still the United States. In July 2026, Carta analyzed over a thousand software fundraising rounds: the median Series A valuation was US$80M, and at least half of those companies could be called AI-native. You almost never see valuations like that in Taiwan's private market.
But I'll be honest: an international team entering the U.S. and going all the way to an IPO is an enormous challenge.
That's why I think the real opportunity in a big market is M&A. Under the AI wave, acquisition volume has clearly outpaced normal years:
- U.S. startups acquired in Q4 2025: 232, an all-time high and 60% more than Q4 2023 (Carta, January 5, 2026)
- AI startups' share of those deals: over 40%
- Pre-seed and seed stage share: nearly half
- Global AI M&A in Q2 2025: 177 deals in a single quarter, roughly double the quarterly average since 2020 (FE International)
One thing to face honestly: nearly half of the companies acquired were very early stage, and many were acquihires, big companies buying talent. But that actually strengthens my argument. Buyers want teams that can plug straight into their U.S. business. If your people aren't in the big market and your customers aren't either, you don't even get a ticket to be acquired.
So picking a bigger market and getting acquired within a few years may well be a better exit than staying in Taiwan.
If you take VC money, you have to go global
Put it all together and my conclusion is this: a Taiwanese AI software startup that needs venture capital and has to answer to the capital markets must go abroad.
This isn't a lack of love for Taiwan, and it isn't talking down Taiwan's capital markets. Those markets are excellent. Their faith just happens to sit in hardware right now, and for good reason. As a software founder, your job is to see this clearly and decide on day one which path you're taking.
My preferred version: markets and capital go abroad, the team stays in Taiwan. Taiwan's engineering talent and cost advantage are our foundation. But customers, investors, and exits should be sought where they actually are. ### Let's walk the road abroad together
If you're building AI software and weighing whether to go global, or you've already started in the U.S. and hit a wall, come talk to us. Beyond investment, we may be able to show you the potholes and the entry tickets before you get there. Reach us at contact@addin.vc.
Also published on ntu.vc.
Raising your first round? Looking for an investor who will challenge you?