Name: 循環交易如何推動AI爆炸式成長！ (How Circular Deals Are Driving the AI Boom)
Uploaded: 2026-01-28T02:44:42.000Z
Duration: 10 min 3 s

From Wall street to rural America, today's AI is the economy.

從 Wall street 到 rural America，今天的 AI 就是 economy。

And markets have priced it like a miracle that can't fail.

而 markets 已經把它當成一個不會失敗的 miracle 來定價了。

程度副詞

Investors are really banking on incredible growth.

Investors 真的在 banking on 驚人的 growth。

Microsoft, meta Alphabet, they are just committing billions and billions of dollars to capital expenditures and expected to continue to raise those numbers over time.

Microsoft、meta、Alphabet，他們正在 committing billions and billions of dollars 到 capital expenditures，而且 expected to continue to raise those numbers over time。

The AI boom is more than just soft construction.

So that's building out the data centers, that's securing the energy that's needed, the water that's needed.

那是建立資料中心，確保所需的 energy 和 water。

But in this burgeoning industry, what goes around doesn't necessarily come around.

但在這個蓬勃發展的產業裡，不是所有投入都會有相對的回報。

A precarious investment strategy is emerging multibillion dollar circular deals.

一種不穩定的投資策略正在出現，就是數十億美元的循環交易。

Nvidia will invest as much as $100 billion in OpenAI.

Nvidia 將對 OpenAI 投資高達 1000 億美元。

Huge, huge sums of money are just being passed between these enormous companies that are pouring hundreds of billions of dollars into the promise of AI.

這些大型企業正將巨額資金在彼此之間流轉，投入數千億美元來實現 AI 的願景。

大約有 80% 的美國企業都在使用 AI。

So this is a structural shift like electricity or the Internet.

這就像是電力或網際網路一樣的結構性轉變。

But AI, with all its potential, remains largely unproven for profitability.

然而，AI 雖然潛力無窮，但其獲利能力仍有待驗證。

Probably the biggest question in San Francisco right now is are we in an AI bubble?

舊金山現在最大的問題，大概就是我們是否身處 AI 保麗龍之中？

And if we are, then, well, how big is this bubble and bad would it be if and when it does burst?

如果是的話，這個保麗龍有多大？一旦破裂，後果又會有多嚴重？

So is this the dawn of a new age of AI powered growth or the biggest bubble ever?

所以，這是一個由 AI 驅動的新時代的開端，還是史上最大的保麗龍？

A circular deal is when companies are basically blowing money, products, services in between two of them.

循環交易是指企業之間基本上在互相「吹捧」金錢、產品和服務。

Nvidia has said it's going to invest up to $100 billion in OpenAI.

Nvidia 表示，他們將對 OpenAI 投資高達 1000 億美元。

At the same time, OpenAI is a major customer of Nvidia services of their chips.

同時，OpenAI 也是 Nvidia 晶片服務的主要客戶。

To make it even more complicated, there are also sort of other middlemen.

更複雜的是，還有一些其他的中間商。

OpenAI sometimes will lease compute from Oracle.

OpenAI 有時會向 Oracle 租用運算資源。

And so you have Oracle being a customer of Nvidia.

所以，Oracle 也是 Nvidia 的客戶。

So you can see how there could be more sort of arrows back and forth between these companies.

你可以看到這些公司之間會有更多來回的連結。

But Nvidia, OpenAI and Oracle are just part of the web.

但 Nvidia、OpenAI 和 Oracle 只是這個網絡的一部分。

The full picture entails a who's who of the AI landscape.

整個畫麵包含了 AI 領域的重量級公司。

This money is kind of spinning around the same companies and that's why people are worried.

這些資金在同一批公司之間打轉，這就是大家擔心的原因。

I don't think there's anything inappropriate about that in principle.

Now if you start stacking these where they get to huge amounts of money, then yeah, you can overextend yourself of course, becomes kind of symbiotic.

如果你開始把這些錢疊加起來，金額變得非常龐大，那確實可能會過度擴張，變得有點共生關係。

And I think that's sort of the concern that's cropping up now is if one of those companies stumbles or doesn't do well, does the whole thing fall apart?

我認為現在大家擔心的就是，如果其中一家公司跌倒或表現不好，整個體系會不會垮掉？

A lot of this investment is going straight to the build out of data centers across the nation.

很多這類的投資都直接用於全國各地數據中心的建置。

We're really in an infrastructure build out arms race.

我們真的處於一場基礎設施建置的軍備競賽。

Have a look at construction spending in 2025.

我們來看看 2025 年的營建支出。

It's down in most sectors, but not data centers or power stations.

大部分的產業都在下滑，但數據中心或發電站卻沒有。

We're seeing a lot of companies that are sort of the picks and shovels of the AI industry.

我們看到很多公司像是 AI 產業的「鋤頭和鏟子」。

They're out there actually digging into the ground.

One recent estimate from Morgan Stanley predicts that companies will spend $3 trillion on AI data centers.

摩根士丹利最近的一項估計預測，公司將在 AI 數據中心投入 3 兆美元。

If you're in the business of selling picks and shovels, you are living it up, you are getting all the money you need, and

如果你是賣「鋤頭和鏟子」的，你現在簡直是爽翻了，賺進了所有你需要的錢，

而且需求大到你根本應付不來。

We're currently in a facility that was once upon a time a textile facility, one of the largest ones on the eastern seaboard, about 1 million square foot.

我們現在所在的這間廠房，以前曾經是東岸最大的紡織廠之一，佔地約一百萬平方英尺。

We realized we could actually repurpose it as a data center.

我們發現其實可以把這裡改建成數據中心。

There is an insatiable demand for building data centers, for making sure that we have the power to build the data centers, the megawatts, the infrastructure and the expertise.

對於興建數據中心、確保我們有足夠的電力來蓋數據中心、兆瓦級的電力、基礎設施和專業知識，需求是永無止境的。

We don't see any slowdown in that for a long while.

我們預期這種情況在很長一段時間內都不會減緩。

It comes to artificial intelligence, time is not your friend.

談到人工智慧，時間對你來說不是優勢。

So if you can get up and running in six months using a retrofit format versus a greenfield format, which is starting from scratch, which takes up to two years, that's a much better proposition.

所以，如果你能透過改造現有廠房的方式，在六個月內就啟用，而不是從零開始的全新建設，那可能需要兩年時間，這會是個好得多的選擇。

所有這些數據中心都需要電力。

The growth in utility costs is outpacing inflation.

公用事業成本的成長速度已經超過了通膨。

Utility companies have done really well, especially ones that are providing energy to data centers.

提供數據中心能源的公用事業公司表現非常好。

Building fast isn't always built to last.

快速建造的東西不一定能長久。

You don't just build a data center, switch it on, forget about it, and make money.

你不能只蓋好一個數據中心，然後開機就放著不管，等著賺錢。

You have to keep on investing to keep that technology working.

你必須持續投資，才能讓這些科技維持運作。

Otherwise it will quickly become useless to the people you want to sell it to.

否則，很快就會對你想賣東西給他們的人來說變得毫無用處。

And so far, all major AI projects are operating at a loss.

到目前為止，所有主要的AI專案都在虧損營運。

The problem is, every time someone uses ChatGPT, OpenAI likely loses money.

問題是，每次有人使用ChatGPT，OpenAI很可能都在賠錢。

Those OpenAIs and anthropics of the world are not yet profitable.

那些OpenAI和Anthropic這類的公司還沒有獲利。

Sam Altman says OpenAI should be able to break even around 2029, 2030.

Sam Altman說，OpenAI應該能在2029年、2030年左右達到損益兩平。

Given the amount of cash that company is burning through right now and the amount it still needs to spend to build the data centers and pay for the computing power to do what it wants to do.

考量到這家公司現在燒錢的速度，以及為了建立資料中心和支付運算能力所需的花費，來達成他們想做的事。

這對我來說聽起來是個艱鉅的任務。

There's some concern about whether the AI startups are actually able to pay their bills when they're racking up these huge commitments to spending on data center infrastructure.

大家有點擔心，這些AI新創公司在承諾投入巨資於資料中心基礎建設時，是否真的有能力支付帳單。

These AI data center companies, they are the kind of canary in the coal mine.

這些AI資料中心公司，就像是煤礦裡的خرى。

They are the ones where we perhaps see on their balance sheet any first sign of companies pulling back from needing as much AI data center capacity.

他們是我們或許能在資產負債表上看到，公司開始縮減對AI資料中心容量需求的最初跡象。

Right now, all the companies, they're all saying that the demand for AI products is really, really high.

目前，所有公司都說，對AI產品的需求真的、真的很高。

If that were to change, if demand were to suddenly weaken, that would become an issue.

如果情況改變，如果需求突然減弱，那就會是個問題。

To understand the stakes of today's AI boom, you don't need a crystal ball.

想了解現今AI熱潮的利害關係，你不需要水晶球。

In 2000, the dot coms promised a brave new world.

2000年時，網路保麗龍公司們承諾了一個嶄新的世界。

Savings, empty office parks and $5 trillion in vanished value.

儲蓄歸零、空蕩蕩的辦公大樓，還有五兆美元蒸發的價值。

The worst hit shares around the world have been the technology stocks, including the dot com companies.

全球受創最嚴重的股票是科技股，包括網路保麗龍公司。

Even the strongest companies in that era took years to recover.

即使是那個時代最強的公司，也花了好幾年才恢復。

Amazon, one of the big great famous survivors, its share price didn't recover for another eight years after the dot com crash happened.

亞馬遜，當時偉大著名的倖存者之一，在網路保麗龍崩盤後，股價又花了八年才恢復。

Cisco, one of the picks and shovel companies, it took them 25 years before their stock price recovered.

思科，當時的「採礦工具」公司之一，他們的股價花了二十五年才恢復。

這兩者之間確實有些相似之處。

One is that there is circular deal making in both instances.

其中一點是，這兩種情況都有循環性的交易。

The question is, is this bubble going to get to a level that goes beyond just kind of the normal ups and downs of tech booms, but

問題是，這個保麗龍會不會膨脹到一個程度，超越一般科技熱潮的起伏，但

actually has major consequences for the economy?

真的會對 economy 造成重大影響嗎？

The dot com boom was devastating for the economy, but

網路保麗龍時期對 economy 造成了毀滅性的打擊，但

I don't think it would compare to the far reaching effects of the AI boom collapsing.

我覺得這跟 AI 保麗龍破裂的深遠影響比起來，還不算什麼。

The money pouring in has been a huge contributor of growth to gdp, helping boost a US economy otherwise hampered by tariffs and inflation.

湧入的資金大幅挹注了 GDP 的 growth，幫助提振了原本受關稅和通膨拖累的美國 economy。

Everyday Americans are exposed to this risk via 401ks and other investment accounts that hold stakes in many of the big tech companies participating in the spending spree.

一般美國民眾透過 401k 和其他投資帳戶接觸到這個風險，這些帳戶持有許多參與這波大灑幣的科技巨頭的股份。

So does that mean the AI bubble is too big to pop?

那麼，這是否意味著 AI 保麗龍大到無法破裂？

There's definitely this question about are these companies becoming too big to fail?

大家都在問，這些公司是不是變得大到不能倒？

If they were to go down, that could have not just economic consequences.

如果它們倒下，影響可能不只是 economic 層面。

There is a suggestion this might be like the global financial crisis where huge financial institutions needed money to survive to prevent a wider collapse of the economy.

有人認為這可能像全球金融危機一樣，當時大型金融機構需要資金才能生存，以防止 economy 更廣泛的崩潰。

Then that's obviously a far greater problem for the US economy.

那顯然對美國 economy 來說是個更大的問題。

But despite these stakes, many remain bullish on AI because of the evolving nature of the technology itself.

不過，儘管有這些風險，許多人仍然看好 AI，因為這項技術本身不斷在演進。

In the dot com boom there were companies that were laying fiber optic cable like subsidiaries of a company were basically all spending in a circle.

在網路保麗龍時期，有些公司鋪設光纖網路，就像是某家公司的子公司，大家都在互相投資。

And it did contribute to the dot com bubble popping.

這確實促成了網路保麗龍的破裂。

But after a while that became the backbone of Internet broadband and that unused part of all the fiber optics that were built in 90s actually ended up being extremely important.

但過了一段時間，這些光纖網路成了寬頻網路的骨幹，而 90 年代鋪設的那些未使用的光纖，最後變得非常重要。

And so we think with respect to data centers, if there is potential excess capacity that's being built, those data centers will eventually be used.

所以我們認為，關於資料中心，即使現在有潛在的過剩產能，這些資料中心最終都會被使用。

There's a scenario here where AI takes longer than we think, and

這裡有一種情況是，AI 的發展比我們預期的要慢，

therefore the strongest companies, sure, they'll survive, but

是以，最強勁的公司，當然，牠們會存活下來，但

in the meantime, there might be a huge hit to their valuations.

在此同時，牠們的估值可能會受到重創。

Ultimately, this technology is not going to burst.