名字寫錯了😅
新「台灣之光」 效法張忠謀！55歲台人新創AI企業 美媒跨海專訪
美國財經媒體《Business Insider》昨（6）日報導，曾任職蘋果(Apple)、亞馬遜(Amazon)的台灣資深工程師黃錫仁，55歲那年放下美國的高薪穩定生活，回台創立AI晶片新創公司「千逢科技」(Tranxform AI)。他用親身經歷證明，創業這條路，年齡從來不是問題。
ChatGPT一爆紅 他決定「不能再等了」
黃錫仁在矽谷做晶片超過數十年，資歷相當扎實——參與過聯發科（MediaTek）的GPU開發、蘋果Face ID技術，也待過亞馬遜的AI晶片團隊。「創立AI晶片公司」這個念頭，其實在他心裡放了很久，只是一直沒下定決心。
轉捩點出現在2022年底，ChatGPT一推出就掀起全球熱潮。黃錫仁看準時機成熟，認為市場已經準備好了。55歲那年，他決定放手一搏，在2024年創立千逢科技，總部設在台灣，主攻省電處理器，讓AI模型不必依賴大型資料中心也能順利運作。公司目前約40名員工，首款晶片預計最快明年就能問世。
張忠謀年過半百創台積電 他說經驗才是本錢
跟一票靠大型語言模型起家的年輕創辦人不同，黃錫仁創業前早已在晶片業磨了數十年。但他從沒把年紀當包袱，反倒覺得這是自己的強項。他常拿台積電創辦人張忠謀舉例——55歲創立台積電，一手打造出全球最大晶圓代工廠。在他看來，硬體新創特別吃經驗，跟軟體業講求年輕、衝勁的玩法完全是兩回事。
黃錫仁解釋，系統單晶片（SoC）是手機、AI系統等裝置的核心處理器，設計時得在軟硬體之間反覆權衡，這種功夫急不來。他說：「要做出一顆好的SoC，你需要經驗，不然根本不知道怎麼拿捏各種運作之間的平衡。」
創業這條路，黃錫仁走得並不輕鬆。他在美國原本有穩定職涯、收入優渥，說走就走並不容易，加上大半時間得留在台灣，家人一開始其實很難接受，好在公司站穩腳步後態度也慢慢轉變，黃錫仁笑說現在家人都以他們的成果為榮。時機點也幫了他一把，創業當時他的兩個兒子都已成年，一個在科技業上班，一個剛畢業，他提到孩子能獨立生活，自己才能把時間和精力全部投入公司。
回台灣新竹紮根 靠老同學救人才荒
黃錫仁的信心不是憑空而來。市場研究機構PitchBook統計，2025年AI與機器學習晶片新創的創投募資金額比前一年暴增逾70%，來到162億美元；交易件數卻從266件降到232件，顯示投資人愈來愈願意把大筆資金押在少數潛力股身上。今年截至6月22日，募資金額已達99億美元，共87筆交易。
AI熱潮推著黃錫仁走上創業路，但矽谷愈演愈烈的搶人大戰，讓他重新盤算基地該設在哪裡。多年矽谷經驗讓他清楚，新創公司要跟Google、蘋果、輝達（Nvidia）這些巨頭搶工程師有多吃力，他無奈地說，公司不斷訓練人才，結果人才都被挖走。最後他選擇落腳新竹，希望組出一支穩定的工程團隊。大學同學李維興就是他找來的重要戰友，這位高通（Qualcomm）退休老將，去年正式加入千逢科技擔任技術長。
創業維艱 樣樣都得自己來
如今黃錫仁的日子早已不是過去那種按部就班的上班生活，天天都在忙募資、找人、跑客戶會議，還得隨時解決各種技術難題。他直言創業真的很辛苦，商業夥伴要自己找，故事要自己講，錢也得自己想辦法籌。
黃錫仁透露，千逢科技正在籌備下一輪募資，細節暫不方便透露，也坦言公司仍處於授權初期，營收還不多。不過他對AI的未來依然樂觀，強調這個產業可能才剛剛起步。
Technology
After decades in Silicon Valley, a former Apple and Amazon engineer started an AI chip company in his mid-50s
Stephen Huang/Tranxform
- After decades in Silicon Valley, Stephen Huang, 55, launched AI chip startup Tranxform.
- Tranxform is trying to build more efficient AI processors and recruit top talent in Taiwan.
- Huang says AI hardware favors deep experience over youthful speed.
Stephen Huang had spent decades building chips in Silicon Valley when ChatGPT's launch convinced him the market was ready for the AI chip company he had long considered building.
By then, Huang had worked on MediaTek's GPUs, Apple'sFace ID technology, and at an Amazon AI chip team.
When ChatGPT burst onto the scene in late 2022, Huang became convinced the industry had reached a turning point. "I felt the market had arrived," he said.
So at 55, Huang decided to start over.
In 2024, he founded Tranxform AI, a Taiwan-based AI chip startup developing power-efficient processors designed to run AI models outside large data centers.
The company now employs about 40 people and is preparing its first chip, which Huang expects to be ready next year.
The CEO is among a growing number of entrepreneurs chasing opportunities created by the AI boom. Unlike many founders building companies around large language models, however, he spent decades designing chips before launching his own startup.
But Huang never viewed age as a disadvantage.
"Morris Chang started TSMC in his 50s," he said, referring to the founder of Taiwan Semiconductor Manufacturing Company,who founded the world's largest contract chipmaker at 55.
In fact, Huang believes age may be on his side, arguing that hardware startups often favor experience in ways software startupsdo not.
Semiconductor design is a long game, he said. Building a system-on-a-chip — the integrated processor that powers devices ranging from smartphones to AI systems — requires balancing countless tradeoffs across hardware and software, a skill that can take decades to develop.
"To build a good SoC, you need experience," Huang said. "Otherwise, you would not know how to balance different operations."
Huang's startup adventure was not an obvious one.
Before founding Tranxform, he had a stable career and a comfortable income in the US. Building the company meant taking a significant risk and spending most of his time in Taiwan, a decision his family initially struggled to accept.
As Tranxform grew and reached key milestones, his family's attitude changed. "Today, they are proud of what we have accomplished," he said.
The timing also helped. Huang's two sons were adults when he founded the company. One works in the technology industry, while the other recently graduated from university.
"Having them become independent has made it easier for me to dedicate the time and energy needed to build Tranxform," he said.
Huangsaidhebelieved the risk was worth taking because demand for specialized AI hardware would continue growing as companies looked for faster and more energy-efficient ways to run increasingly complex models.
Huang's optimism comes as investors pour fresh capital into AI hardware.
Venture funding for AI and machine-learning chip startups rose over 70% to $16.2 billion in 2025 from the year before, according to PitchBook. However, the number of deals fell to 232 from 266 in the same period as investors increasingly wrote bigger checks to a smaller cohort of potential breakout companies.
This year, funding stood at $9.9 billion as of June 22, with deal counts at 87, per PitchBook.
Returning home
While AI drew Huang toward entrepreneurship, Silicon Valley's intensifying talent wars made him rethink where to build his company.
His years in Silicon Valley taught him how difficult it was for startups to compete with tech giants such as Google, Apple, and Nvidia for engineering talent.
"We kept training people, and they got poached," he said.
Huang built his company in the Taiwanese chip hub of Hsinchu, where he believed he could assemble a more stable engineering team.
One of those recruits was his college classmate Way-Shing Lee, who joined Tranxform as chief technology officer last year after retiring from the US chip giant Qualcomm.
Now, startup life has replaced the stability of Huang's previous career with fundraising, recruiting, customer meetings, and constant technical problem-solving.
"Starting a company is very hard," Huang said. "You have to find business partners. You have to sell your story. You have to find funding."
Huang said Tranxform is preparing for its next fundraising round, declining to disclose details.
The company is still in the early stages of licensing and generates little revenue, he said.
Still, Huang believes the biggest opportunities in AI remain ahead.
The AI industry is "probably just getting started," he said.
