AI’s Single Point of Failure | Rob Toews | TED

譯者: 麗玲 辛 審譯者: Shelley Tsang 曾雯海

The following statement is utterly ludicrous.

以下陳述荒謬至極，

It is also true.

同時也是事實。

The world's most important advanced technology

全球最重要的先進科技

is nearly all produced in a single facility.

幾乎全由一家公司生產。

What's more, that facility is located

更重要的是，

該公司位於全球地緣政治 最為敏感的地區之一，

in one of the most geopolitically fraught areas on Earth,

an area in which many analysts believe that war is inevitable within the decade.

許多分析師認為，十年內 該地區必然會發生戰爭。

The future of artificial intelligence hangs in the balance.

人工智慧的未來難以預料。

The Taiwan Semiconductor Manufacturing Company,

台灣積體電路製造公司（TSMC）

or TSMC,

makes all of the world's most-advanced AI chips.

生產全球所有最先進的人工智慧晶片。

This includes Nvidia's GPUs, Google's TPUs, AMD's GPUs,

這包括輝達的 GPU、 谷歌的 TPU、超微的 GPU、

the AI chips for Microsoft, Amazon, Tesla,

微軟、亞馬遜、特斯拉、

Cerebrus, SambaNova

Cerebrus、SambaNova

and every other credible competitor.

和每個值得一提的 競爭對手的人工智慧晶片。

Modern artificial intelligence simply would not be possible

如果沒有這些高度專業化的晶片，

without these highly specialized chips.

現代人工智慧根本不可能實現。

Little wonder, then,

難怪

that Time magazine recently described TSMC as,

《時代》雜誌最近將台積電描述為

"The world's most important company that you've probably never heard of."

「全球最重要的公司， 但你可能從未聽說過」。

Nvidia CEO Jensen Huang put it more colorfully,

輝達執行長黃仁勳說得更生動，

saying, “Basically, there is air ... and TSMC.”

他說：「基本上，有空氣… 還有台積電。」

(Laughter)

（笑聲）

TSMC's chip fabrication facilities, or fabs,

台積電的晶片製造設施，或稱晶圓廠，

the buildings where chips are physically built,

即實際製造晶片的建築物，

is located on the western coast of Taiwan,

位於台灣西海岸，

a mere 110 miles from mainland China.

距離中國大陸僅 177 公里。

In this map, Taiwan is shown in orange and China is shown in green.

在這張地圖裏，台灣以橘色顯示， 中國以綠色顯示。

Today, China and Taiwan are nearer to the brink of war

現今，中國大陸和台灣

處於幾十年來最接近戰爭的邊緣。

than they have been in decades.

Many policymakers in Washington predict that China will invade Taiwan

華盛頓的許多決策者預測 中國將在未來五年內入侵台灣。

within the next five years.

A China-Taiwan conflict would be devastating

中國和台灣的衝突會有災難性後果，

for many reasons.

原因有很多。

Aside from the heavy human toll,

除了嚴重的傷亡之外，

one underappreciated consequence

一個被低估的後果是

is that it would paralyze the global AI ecosystem.

這將使全球人工智慧 生態系統陷入癱瘓。

Put simply, the entire field of artificial intelligence

簡而言之，整個人工智慧領域

faces an astonishingly precarious single point of failure in Taiwan.

面臨台灣這個異常脆弱的 單點故障危機。

Amid all of the fervor around AI today,

在當今人工智慧的熱潮中，

this fact is not widely enough appreciated.

這一事實並未得到足夠的廣泛重視。

If you are working on or are interested in AI,

如果你正在研究人工智慧 或對人工智慧感興趣，

you need to be paying attention.

你就要密切關注。

How did we get here and what can we do about it?

我們是如何走到這一步的？ 我們能做些什麼？

Let's start with a brief, whirlwind overview of the chip industry.

我們先對晶片產業做個簡短的概述。

Semiconductors, or chips, are the most complex object in the world

半導體或晶片是世界上人類知道

that humanity knows how to mass-produce.

如何大規模生產的最複雜的物體。

Making semiconductors requires the world's purest metals,

製造半導體需要世界上最純粹的金屬、

the world's most expensive machinery,

世界上最昂貴的機械、

legions of highly specialized engineers

大批高度專業的工程師

and atom-level manufacturing precision.

和原子級的製造精度。

It is important to distinguish between two different types of chip companies.

必須知道，有兩種 不同類型的晶片公司。

First, fabless chip makers,

首先是無晶圓廠晶片製造商，

which design but do not manufacture their own chips.

他們設計但不製造自己的晶片。

And second, foundries,

其次是代工廠，

which manufacture chips designed by other companies.

生產其他公司設計的晶片。

Almost every well-known chip company today is fabless,

如今幾乎所有知名晶片公司 都沒有晶圓廠，

from Nvidia to AMD to Qualcomm.

如輝達 、超微和高通。

These companies do not produce their own chips.

這些公司不生產自己的晶片，

Instead, they design chips,

而是設計晶片，

and then they rely on foundries like TSMC

然後仰賴台積電等代工廠

to actually manufacture those chips for them.

為他們製造這些晶片。

There are only three companies in the world today

當今，世界上只有三家公司

that are capable of manufacturing chips

有能力製造半導體尖端技術的晶片：

anywhere near the leading edge of semiconductor technology:

TSMC, Samsung and Intel.

台積電、三星和英特爾。

Of those three,

在這三者中，

only one can reliably produce the world's most advanced AI chips,

只有一家能夠可靠地生產 全球最先進的人工智慧晶片，

including chips like Nvidia's H100 GPUs.

包括輝達的 H100 GPU 等晶片。

That one company is TSMC.

那家公司就是台積電。

As of this morning,

截至今天上午，

TSMC's market capitalization was 470 billion dollars,

台積電市值為 4,700 億美元，

making it the 13th-largest company in the world,

是全球第 13 大公司，

larger than ExxonMobil,

超越埃克森美孚、

JPMorgan Chase or Walmart.

摩根大通或沃爾瑪。

How has TSMC become such a dominant force?

台積電是如何成為如此主導的力量？

The short answer is that powerful economies of scale

簡而言之，晶片製造領域 受制強大的規模經濟，

exist in the world of chip fabrication,

leading inexorably to winner-take-all dynamics.

不可避免地導致贏家通吃的局面。

Making advanced semiconductors requires tremendous upfront

製造先進半導體

需要大量的前期和持續的資本支出。

and ongoing capital expenditure.

In 2021, TSMC announced that it would invest 100 billion dollars

2021 年，台積電宣布

未來三年將投資一千億美元，

over the next three years

to continue expanding its fabrication capabilities.

繼續擴大製造能力。

No other company in the world can justify that level of investment.

全球沒有其他公司 可以宣稱這樣的投資規模合理。

TSMC can, because of the sheer volume of chips that it produces,

台積電可以，因為它生產的晶片數量

far more than any other company in the world.

遠遠超過世界上任何其他公司。

A related dynamic that helps explain TSMC's unassailable position

另一個可解釋台積電 無可爭議地位的動因

is what has come to be known as the TSMC Grand Alliance.

就是所謂的台積電大聯盟。

TSMC has invested heavily over decades

幾十年來，台積電投入巨資，

to develop deep partnerships with dozens of companies

與半導體供應鏈上的數十家公司

across the semiconductor supply chain,

建立了深厚的合作夥伴關係，

from software providers like Cadence

如益華電腦等軟體供應商，

to equipment manufacturers like ASML

艾司摩爾等設備製造商，

to chip designers like Nvidia.

以及輝達等晶片設計商。

In turn, these companies have developed their own products

同樣，這些公司也依照台積電的路線， 開發了自己的產品，

in accordance with TSMC's road map,

leading to powerful lock-in.

從而形成強大的鎖定效應。

In summary, a combination of economies of scale, network effects,

總而言之，規模經濟、網絡效應

and unrivaled specialization have made TSMC irreplaceable

和無與倫比的專業化相結合， 讓台積電變得不可替代，

and have made the entire world deeply, precariously dependent upon it.

並使整​​個世界深深地、 卻不牢靠地依賴著它。

This brings us to the present, delicate geopolitical moment.

接著我們來談談當前 微妙的地緣政治狀況。

Last October,

去年 10 月，

the Biden administration took the dramatic step

拜登政府採取了戲劇性的一步，

of banning the export of all high-end AI chips to any entity in China.

禁止向中國任何實體 出口所有高階人工智慧晶片。

The rationale behind these measures was clear.

這些措施背後的理由很明確。

To leverage US control

利用美國對全球半導體供應鏈的控制，

of the global semiconductor supply chain

as a choke point to handicap China's AI capabilities.

作為阻礙中國人工智慧能力的瓶頸。

The US government is currently formulating expansions to this policy.

美國政府目前正在擬定 擴大這項政策的方案。

At the same time, the US is taking steps to reduce its reliance

同時，美國也採取措施，

減少對東亞晶片製造設施的依賴。

on chip fabrication facilities located in East Asia.

In late 2022,

2022 年底，

TSMC announced that it would invest 40 billion dollars

台積電宣布投資 400 億美元

to build two new state-of-the-art fabs

在美國亞利桑那州 新建兩座最先進的晶圓廠。

in the United States, in Arizona.

The first of these two fabs is slated to begin production in 2025.

這兩座工廠中的第一座預計 將於 2025 年開始生產。

Bringing advanced chip production to US soil

把先進的晶片生產帶到美國本土，

will help mitigate the AI industry's absolute dependence on Taiwan-based fabs.

將有助於減輕人工智慧產業 對台灣晶圓廠的完全依賴。

But the Arizona fabs will not solve everything.

但亞利桑那州的晶圓廠 並不能解決所有問題。

Their production capacity will be modest,

它們的產能不大，

representing less than five percent of TSMC's total global output.

僅佔台積電全球總產量的不到 5%。

And the most advanced semiconductor production capabilities

而最先進的半導體生產能力

and technologies will remain in Taiwan.

和技術仍將留在台灣。

So where might things go from here?

那麼，未來事態會如何發展呢？

Let's briefly consider a few possibilities

讓我們簡略考量一下

這個三維棋盤上的幾個可能性。

on this three-dimensional chessboard.

Let's start with optimistic scenario.

讓我們從樂觀的情況開始。

Taiwan's central role in the global semiconductor industry

台灣在全球半導體產業中的核心角色

is often referred to as its “silicon shield.”

通常被稱為「矽盾」。

The basic theory is this:

基本理論是這樣的：

because China depends so heavily on Taiwan for the chips that it needs

由於中國重度依賴台灣提供晶片，

to keep its own economy running,

以維持其經濟運作，

China will stop short of invading Taiwan and putting TSMC's production at risk.

因此中國不會入侵台灣， 並使台積電的生產面臨風險。

And because the rest of the world is likewise so dependent on TSMC,

而且，由於世界其他國家 同樣如此依賴台積電，

the United States and other powers will go to great lengths

美國和其他大國將不遺餘力地

to protect the island and defend its sovereignty.

保護台灣，並捍衛其主權。

Under this theory,

根據這個理論，

while China may continue to build out its military

中國也許會繼續擴增軍隊，

and engage in cross-strait saber-rattling,

並在兩岸進行武力威脅，

it will stop short of kinetic action against Taiwan.

但不會對台灣採取實質的行動。

But the silicon shield is just a theory,

但矽盾只是一個理論，

not a guarantee.

絕非保證。

What would happen if China were to move decisively to retake Taiwan?

如果中國斷然採取行動奪取台灣， 會發生什麼情況呢？

TSMC's fabs would almost certainly be rendered inoperative.

台積電的晶圓廠幾乎肯定會停產。

It is conceivable that the Taiwanese, or even the US military,

可以想像，台灣，甚至美軍，

would preemptively destroy the fabs

會先發制人地摧毀晶圓廠，

in order to prevent the CCP

以阻止中共控制這寶貴的戰略資源。

from taking control of this valuable strategic resource.

Even if the physical buildings were to remain undamaged

即使實體建築在中國入侵後完好無損，

after a Chinese invasion,

it is unrealistic that the CCP would be able to continue operating the fabs

中共也無法讓這些晶圓廠 繼續運作，生產尖端晶片。

to produce cutting-edge chips.

Keeping leading edge fabs running

要讓這些尖端晶圓廠運轉，

requires ongoing and deep partnership

需要與全球半導體生態系統的各個組織

with organizations across the global semiconductor ecosystem,

進行持續而深入的合作，

as well as a steady inflow of materials, equipment and services.

還需要穩定的材料、設備和服務供應。

These would be denied to an invading power.

以上這些，侵略者將無法取得。

Let me say this one more time:

我再說一遍，

if or when China invades Taiwan,

如果中國入侵台灣，

TSMC's fabs will, in all likelihood, go offline.

台積電的晶圓廠很有可能停擺。

This will mean that no more Nvidia H100s

這將意味著全球沒有任何公司

or any other cutting-edge AI chips

可以生產輝達 H100 或其他尖端人工智慧晶片。

will be able to be produced anywhere in the world.

What would this mean for the world of AI?

這對人工智慧世界意味著什麼？

After TSMC, the company best positioned to step up

台積電之外，

最有能力接手生產 尖端人工智慧晶片的公司是三星。

and produce cutting-edge AI chips is Samsung.

Samsung is currently the only company in the world, other than TSMC,

除台積電之外，三星是目前

that is capable of producing three-nanometer chips,

全球唯一能夠生產當今尖端技術 三奈米晶片的公司。

today's cutting-edge technology.

But Samsung's production capabilities are far inferior to TSMC's today.

但三星的生產能力 遠不如現在的台積電。

In a best-case scenario,

在最好的情况下，

it would take Samsung years to scale up

三星也還需要數年時間

才能達到台積電目前的 人工智慧晶片良率和產量。

to TSMC's current AI chip yields and volumes.

This brings us to America's former chip champion, Intel.

那，美國之前的晶片霸主英特爾呢?

It was hardly a decade ago

僅僅十年前，

that Intel's chip-manufacturing capabilities were the envy of the world.

英特爾的晶片製造能力曾令全球稱羨。

But in recent years, Intel has fallen behind.

但近年來，英特爾已經落後。

The company struggled mightily in its transition to both ten-nanometer

該公司在 10 奈米和 7 奈米 節點技術過渡的過程中

and seven-nanometer node technologies,

遇到了巨大的困難，

even resorting to outsourcing some of its leading edge production to TSMC.

甚至不得不将部分 尖端生產外包給台積電。

Under CEO Pat Gelsinger,

在執行長帕特·蓋爾辛格的領導下，

Intel aspires to regain its chipmaking supremacy

英特爾渴望重新奪回 晶片製造霸主地位，

with an ambitious plan to leapfrog TSMC

並製定了一項雄心勃勃的計劃， 計劃超越台積電，

and begin producing two-nanometer chips in 2024.

將在 2024 年開始生產二奈米晶片。

Whether this ambitious plan will actually prove achievable, however,

然而，這項雄心勃勃的計劃 是否能夠真正實現，

remains to be seen.

還有待觀察。

Before we despair too much,

在過於絕望之前，

let us note a couple encouraging points.

我們來關注幾件令人鼓舞的事。

First, keep in mind that a considerable stock of AI chips already exists

首先，請記住，全球已有 大量的人工智慧晶片庫存。

in the world.

And even in a worst-case scenario, these chips would remain in use.

即使在最壞的情況下， 這些晶片也能繼續使用。

Second, while the most advanced AI chips, like Google’s TPUs or Nvidia’s H100s,

其次，雖然最先進的人工智慧晶片， 如谷歌的 TPU 或輝達的 H100，

can only be manufactured in Taiwan,

只能在台灣製造，

there are many fabs around the world,

但世界各地有許多晶圓廠，

from the US to Europe to Israel,

從美國、歐洲、到以色列，

that are capable of producing lagging-edge logic chips at scale.

有能力大規模生產落後的邏輯晶片。

Though they are far less powerful than today's leading AI chips,

儘管它們的功能遠不如 當今尖端人工智慧晶片強大，

these previous-generation chips could be used in a pinch

但這些上一代晶片可以在緊要關頭

to support some AI computing workloads.

用於支援某些人工智慧運算負載。

Ultimately, though, it would be devastating for humanity

然而，最終，如果我們無法生產 驅動當今尖端人工智慧的晶片，

to lose its ability to produce the chips

that power today's cutting-edge artificial intelligence.

將對人類產生毀滅性影響。

Progress in AI would be profoundly disrupted.

人工智慧的進步將受到嚴重干擾。

Let us hope that diplomacy prevails.

讓我們期昐外交取得勝利。

(Applause)

（掌聲）