Name: 為什麼「AI」不能沒有臺灣？ (Why AI Can't Exist Without Taiwan)
Uploaded: 2024-11-08T21:01:10.000Z
Duration: 9 min 41 s

基礎被動語態

90% of all the world's most advanced semiconductors, or chips, are made right here, in Taiwan.

世界上最先進的半導體或芯片 90% 都是在臺灣製造的。

Taiwan is the most important spot in the world for artificial intelligence.

臺灣是世界上人工智能最重要的地方。

All of NVIDIA's AI chips, they're made right here on the island.

英偉達的所有人工智能芯片，都是在島上製造的。

所以，沒有臺灣，就沒有人工智能。

But Taiwan is more than just a microchip factory, it's the beating heart of the wider AI ecosystem.

但臺灣不僅僅是一個微型芯片工廠，它還是整個人工智能生態系統的心臟。

Taiwanese companies are estimated to manufacture 90% of those AI servers today.

據估計，目前 90% 的人工智能服務器都是由臺灣公司生產的。

But that dominance faces a looming threat.

但這種主導地位面臨著迫在眉睫的威脅。

China sees Taiwan as part of its territory and has become increasingly assertive.

中國將臺灣視為其領土的一部分，並變得越來越強硬。

And so the future of AI has become intertwined with the security of Taiwan.

是以，人工智能的未來與臺灣的安全息息相關。

Taiwan is really an engine that's driving AI.

臺灣確實是推動人工智能發展的引擎。

So it's important that Taiwan stay safe and stay what it is today if we want to keep AI moving forward.

是以，如果我們想讓人工智能繼續向前發展，臺灣就必須保持安全，保持現在的狀態。

So, how can Taiwan's tech industry adapt to this precarious landscape, and what does it mean for AI?

那麼，臺灣的科技產業如何才能適應這種岌岌可危的局面，這對人工智能又意味著什麼呢？

Back in the 70s, Made in Taiwan had this image of cheap toys or things that break.

上世紀 70 年代，"臺灣製造 "給人的印象是廉價玩具或容易壞的東西。

But the government really wanted to push the island forward in a more high-tech way.

但政府確實希望以更高科技的方式推動該島的發展。

It saw potential in chipmaking and technology, investing heavily in the 1980s and building a top-tier talent pool.

它看到了芯片製造和技術領域的潛力，在 20 世紀 80 年代進行了大量投資，並建立了一流的人才庫。

The STEM-related degrees are considered relatively prestigious in Taiwan as opposed to other degrees.

與其他學位相比，STEM 相關學位在臺灣被認為是比較有聲望的。

Then, with this big pool of engineers, there's also a lot of entrepreneurship.

有了大量的工程師，也就有了大量的創業者。

Taiwan became a global leader in efficient mass manufacturing, with factories both at home and in China.

臺灣成為全球高效大規模製造的領導者，在國內和中國大陸都設有工廠。

And out of that, Taiwan's Semiconductor Manufacturing Company, or TSMC, was born.

臺灣積體電路製造公司（TSMC）由此誕生。

Today, its scale and cutting-edge technology are unrivaled.

如今，其規模和尖端技術已無與倫比。

And its partner right from the beginning?

NVIDIA, whose specialized chips are at the forefront of artificial intelligence.

英偉達™（NVIDIA）公司，其專用芯片在人工智能領域處於領先地位。

Please welcome to the stage, NVIDIA founder and CEO, Jensen Wang.

有請英偉達公司創始人兼首席執行官王俊森上臺。

TSMC manufactured NVIDIA's chips from the start.

臺積電從一開始就生產英偉達的芯片。

That helped NVIDIA grow, and obviously that's helped TSMC grow.

這有助於英偉達的發展，顯然也有助於臺積電的發展。

Taiwan is the home of our treasured partners.

臺灣是我們珍視的合作伙伴的故鄉。

And our partnership has created the world's AI infrastructure.

我們的合作伙伴關係打造了全球人工智能基礎設施。

Just as TSMC has been there since the start, so too have the Taiwanese companies supplying vital hardware components.

正如臺積電從一開始就參與其中一樣，提供重要硬件組件的臺灣公司也是如此。

Once used to power PC servers, they've evolved to meet the demands of AI.

它們曾經用於為 PC 服務器提供動力，如今已發展到可以滿足人工智能的需求。

那麼，人工智能服務器到底有哪些功能呢？

Of course, you have to start with the chip, then the chip goes on to a motherboard.

當然，你必須從芯片開始，然後芯片再到主板。

There's a lot of power modules that are in there.

You'll have what we now call liquid cooling systems.

我們現在稱之為液體冷卻系統。

And then that goes into a server, and a server rack, and tons of these.

然後將其裝入服務器，再裝入服務器機架，數不勝數。

Rows and rows and rows of these will create our chat GPTs of the future.

一排又一排，一排又一排，這些將創建我們未來的哈拉 GPT。

Oh, I forgot to say, and Taiwan makes all of this.

哦，我忘了說，這一切都是臺灣創造的。

If you want to do the shopping in Taiwan, you can shop pretty much every product, every component, for that reason.

如果你想在臺灣購物，你幾乎可以買到每件產品、每個部件，原因就在於此。

There's always some supplier that can fulfill your demand.

總有一些供應商能滿足您的需求。

Asia Vital Components, or AVC, is one such company in the AI supply chain.

亞洲重要元件公司（AVC）就是人工智能供應鏈中的這樣一家公司。

It's worked with NVIDIA for more than two decades.

它與英偉達™（NVIDIA®）已經合作了二十多年。

Specializing in heat dissipation, it pivoted its R&D from traditional cooling solutions like fans to liquid cooling, which better suits the power-hungry and heat-generating AI servers.

該公司專注於散熱，其研發方向從風扇等傳統冷卻解決方案轉向液冷，因為液冷更適合耗電量大、發熱量高的人工智能服務器。

They're basically using water or coolants to use the pipe attached to a certain area, like a component, so they can conduct heat very efficiently out of that chip.

它們基本上是利用水或冷卻劑將管道連接到某個區域，比如某個元件，這樣就能非常有效地將熱量從芯片中傳導出去。

Right now, 90% of thermal-related spending for data centers is on air cooling, with liquid cooling at just 10%.

目前，數據中心與散熱相關的支出 90% 用於空氣冷卻，液體冷卻僅佔 10%。

But as AI servers grow, liquid cooling could soar to 30% by 2028.

但隨著人工智能服務器的增長，到 2028 年，液冷技術的應用比例可能會飆升至 30%。

An NVIDIA NVL72 server, which is the highest tech today, is about $3-4 million.

英偉達™（NVIDIA®）NVL72 服務器是目前技術含量最高的服務器，價格約為 300-400 萬美元。

A small component can hold back that AI server from getting shipped if you don't have it, because you have to cool it, otherwise it'll melt down.

如果沒有一個小部件，人工智能服務器就無法發貨，因為你必須冷卻它，否則它就會融化。

And so a lot of these, as Jensen calls them, unsung heroes, companies that most of us have never heard of, these are the key companies in the AI revolution.

是以，詹森稱之為 "無名英雄 "的許多公司，我們大多數人都沒聽說過，但它們卻是人工智能革命中的關鍵公司。

Driven by an innovative spirit, AVC pursued Amazon as a customer and set up shop right in front of their office in Seattle.

在創新精神的驅動下，AVC 將亞馬遜作為自己的客戶，並在其位於西雅圖的辦公室前開設了店鋪。

Their persistence paid off, and shortly after landing Amazon as a customer, Microsoft followed suit.

他們的堅持得到了回報，在亞馬遜成為其客戶後不久，微軟也緊隨其後。

With its small size and efficient high-speed rail network, Taiwan continues to lure the big tech companies.

臺灣以其小巧的面積和高效的高鐵網絡，不斷吸引著大型科技公司。

Most of the major tech companies doing AI computing today, Google, Amazon, Microsoft, they all come to Taiwan because they can meet all of the partners really quickly and easily.

如今，大多數從事人工智能計算的大型科技公司，如谷歌、亞馬遜、微軟，都會來臺灣，因為他們可以非常快速、方便地與所有合作伙伴會面。

Since 2023, Taiwan's AI-related exports have overtaken China's, a reflection not just of its growing importance in the world of artificial intelligence, but also the frosty relationship between Washington and Beijing.

自 2023 年以來，臺灣與人工智能相關的出口額已超過中國，這不僅反映了臺灣在人工智能領域日益增長的重要性，也反映了華盛頓與北京之間的冷淡關係。

The US and China are engaged in almost a tech war, and so the US has banned from 2022 a lot of AI chips from getting shipped to China.

美國和中國幾乎打起了科技戰，是以美國從 2022 年開始禁止向中國運輸大量人工智能芯片。

That means a lot of these AI servers that would have been built in factories in China are being built either in Taiwan or in other manufacturing facilities that Taiwanese companies own.

這意味著，許多原本在中國工廠生產的人工智能服務器，現在都在臺灣或臺灣公司擁有的其他製造工廠生產。

But with the majority of these AI chips being shipped to China,

但這些人工智能芯片大部分都運往中國、

But with the majority of AI technology concentrated on this island, the ever-present risk from neighbouring China concerns all those who depend on it for their computing needs.

但是，由於大部分人工智能技術都集中在這個小島上，來自鄰國中國的風險無時不在，這讓所有依賴中國提供計算服務的人都感到擔憂。

China has carried out what it says is a mock air blockade of Taiwan using jets carrying live ammunition.

中國使用攜帶實彈的噴氣式飛機對臺灣進行了所謂的模擬空中封鎖。

They are determined, whichever way possible, however long they have to wait, to bring Taiwan into the fold and be part of the People's Republic of China.

他們下定決心，不管用什麼方法，不管要等多久，都要讓臺灣加入中華人民共和國，成為中華人民共和國的一部分。

With so much at stake, TSMC and Dutch company ASML can now render their manufacturing plants inoperable should an attack occur.

由於事關重大，臺積電和荷蘭 ASML 公司現在可以在攻擊發生時使其製造工廠無法運作。

是以，從本質上講，這就是一個斷電開關。

It underscores just how concerned they are about China getting hold of these most sophisticated machines.

這凸顯了他們對中國獲得這些最先進機器的擔憂。

Although an invasion remains unlikely, a war could severely damage the global economy.

雖然入侵的可能性不大，但戰爭會嚴重損害全球經濟。

Bloomberg Economics estimates it would knock out the semiconductor supply chain, delaying any advancements in AI and have a $10 trillion impact on global GDP.

彭博經濟公司估計，這將導致半導體供應鏈癱瘓，延後人工智能的發展，並對全球 GDP 產生 10 萬億美元的影響。

That would dwarf what the war in Ukraine, COVID pandemic and global financial crisis have each cost.

這將使烏克蘭戰爭、COVID 大流行病和全球金融危機各自造成的損失相形見絀。

To further mitigate this risk, Taiwanese companies have begun diversifying their manufacturing operations abroad to places like Mexico and Southeast Asia.

為了進一步降低這種風險，臺灣企業已開始將其海外製造業務分散到墨西哥和東南亞等地。

TSMC has broken ground on new facilities in Arizona, Japan and Germany.

臺積電在亞利桑那州、日本和德國的新工廠已破土動工。

They realise that they can't only stay in Taiwan, although Taiwan is still the lion's share of production.

他們意識到不能只留在臺灣，儘管臺灣仍佔生產的絕大部分。

The US has also passed the CHIPS Act, offering incentives to bring semiconductor manufacturing back home.

美國還通過了《CHIPS 法案》，為半導體制造業迴流美國提供激勵措施。

From a risk management perspective, pretty much all the clients would like to think about how to reduce the reliance, but the reality is that it takes time.

從風險管理的角度來看，幾乎所有客戶都希望考慮如何減少依賴，但現實是這需要時間。

And so, as we set sail into an AI-powered future, the technology and global stability remains anchored to Taiwan.

是以，當我們駛向人工智能驅動的未來時，技術和全球穩定仍以臺灣為依託。

Years later, maybe manufacturing can build up in other regions of the world, but that's going to take a long time.

多年後，也許製造業可以在世界其他地區建立起來，但這需要很長的時間。

Replicating Taiwan in the short term is impossible.

在短期內複製臺灣是不可能的。

欲瞭解更多聯合國視頻，請訪問 www.un.org