Name: 美中AI競賽為何變得更加精彩？ | The Dip Podcast (Why the AI race between the US and China just got a lot more interesting | The Dip Podcast)
Uploaded: 2026-02-04T17:09:44.000Z
Duration: 32 min

現在進行式

We're starting with the AI race, the clash between the US and China.

我們要從 AI race 開始，也就是美國和中國之間的較勁。

So a year ago, the tech world was rocked by the debut of Deep Seat.

大約一年前，Deep Seat 的問世震撼了科技界。

It really seemed to break one of these cardinal rules of AI, that bigger is better and newer is better.

它真的打破了 AI 的一條重要規則，就是越大越好、越新越好。

And now Deep Seek says it's getting ready to launch V4, which is a coding focused model that it says, according to its own internal tests, will outperform ChatGPT.

現在 Deep Seek 表示，他們準備推出 V4，這是一個專注於程式碼的模型，根據他們內部的測試，表現會比 ChatGPT 還要好。

And get this, they're doing it at 95% cheaper than ChatGPT.

而且更厲害的是，他們的成本竟然比 ChatGPT 還要便宜 95%。

And so you remember last year that Deepseek rocked the world, not because it ran on the latest cutting edge chips, but

所以你還記得去年 Deepseek 震撼了全世界，不是因為它使用了最新的尖端晶片，

because it had top tier performance even on less powerful hardware.

而是因為它即使在效能較弱的硬體上，也能有頂尖的表現。

And China's managed to make all kinds of things much more cheaper and much more efficient and mass produce them, from solar panels to electric cars.

而且中國已經成功讓各種產品，從太陽能板到電動車，都變得更便宜、更有效率，並且大量生產。

Though in many cases with the help of massive subsidies, of course.

當然，很多情況下這是有大量補貼的幫助。

Will we soon all be using Chinese AI models?

我們是不是很快就會開始使用中國的 AI models？

To find out more about the state of this race, which is definitely feeling more like a marathon than a sprint, I'm joined by Kyle Chan.

想更瞭解這場競賽的現況，這場競賽絕對感覺起來更像一場馬拉松而不是短跑衝刺，我請到了 Kyle Chan。

Thornton China center at the Brookings Institution, and

Thornton China center 的研究員，隸屬於 Brookings Institution，還有

Alok Mehta, the director of the Wadhwani AI center at the center for Strategic International Studies.

Alok Mehta，他是 center for Strategic International Studies 的 Wadhwani AI center 主任。

Do you think that OpenAI will be bracing for impact when the next version of Deep Seat comes out?

你覺得 OpenAI 會不會在 Deep Seat 下一個版本出來的時候，準備迎接衝擊？

Deep Sea V4 is expected to come out because last year it seemed to be a bit of a shock.

Deep Sea V4 預計會推出，因為去年它似乎造成了一點震撼。

Will it be just as dramatic this time round?

I even remember that Nvidia's shares took a hit when Deep SEQ first came out.

我還記得 Deep SEQ 剛推出時，Nvidia 的股價還是以下跌。

Yeah, I think what we've seen is that over the past year or so.

是的，我認為我們看到的是，在過去一年左右的時間裡。

You know, I think probably a year ago it would have been hard for people to name a significant number of Chinese models.

說真的，我認為大概一年前，大家可能很難說出幾個重要的 Chinese models。

But since then, we've seen the competition in the space between US and China models increase pretty significantly.

但從那之後，我們看到 US 和 China models 之間的競爭顯著增加。

We've seen the gap between the best Chinese models and the best US models shrink pretty significantly.

我們看到頂尖的 Chinese models 和頂尖的 US models 之間的差距顯著縮小。

So I think certainly OpenAI is going to be worried about that.

所以我認為 OpenAI 肯定會擔心這一點。

Particularly Google's released new models that have gotten very favorable reviews.

特別是 Google 發布了新模型，獲得了非常好的評價。

And so that, I think is also an indication of just how competitive the overall market for frontier models is.

所以，我認為這也顯示了整個頂尖模型市場的競爭有多激烈。

You know, Kyle, I'm wondering, is the Sputnik like AI race back on?

Kyle，我很好奇，Sputnik 這樣的 AI race 是不是又回來了？

Is it the US versus China, where one of them will be ultimately crowned the winner.

是美國對決中國，其中一方最終會被加冕為贏家嗎？

So I think this race metaphor is interesting, but

所以，我認為這個 race 的比喻很有趣，但是

I think it kind of covers over a broader set of dynamics beyond just sort of this.

我認為它掩蓋了比這更廣泛的動態。

For example, when you look at the Chinese AI landscape, you actually see a wide range of different strategies.

舉例來說，當你看看中國的 AI 領域，你會看到各種不同的策略。

From Alibaba trying to integrate with their existing ecosystem, Deepseek kind of going all out for AGI, as it were.

從阿里巴巴試圖整合現有的生態系統，到 Deepseek 像是全力衝刺 AGI。

Some of the startups like Moonshot or Jephel that recently listed on public markets are maybe targeting specific verticals as well.

像是 Moonshot 或 Jephel 最近在公開市場上市的一些新創公司，可能也在瞄準特定的垂直領域。

So I think it's multiple races, if anything.

所以我認為，如果有的話，這其實是多場競賽。

And you mentioned artificial general intelligence, AGI there, of course we're quite far away, it seems, from that.

而你提到了通用人工智能，也就是 AGI，當然，看來我們離那還很遠。

But when I look at Deepseek and the fact that it just became the number one downloaded app in the US overnight, there was a lot of buzz around, are we soon going to get to AGI then?

但當我看到 Deepseek，而且它一夜之間就成為美國下載量第一的應用程式時，大家都在熱烈討論，我們是不是很快就要達到 AGI 了呢？

But then, even then, Deep Seek suddenly just disappeared overnight from the headlines.

但隨後，即使是 Deep Seek 也突然一夜之間從新聞頭條消失了。

Why haven't we been constantly hearing about this so called battle ever since then?

為什麼從那之後我們就沒有不斷聽到關於這場所謂的戰役？

They have focused on updating their existing models.

他們專注於更新他們現有的模型。

They are expected a major new model very soon in the run up to Chinese New Year.

他們預計在農曆新年前夕會推出一個重大的新模型。

I think it's a testament to the depth of the playing field in China that it's not just them, that you have a whole range of different AI labs in China that are posting pretty impressive benchmarks and appear to be hot on the heels of their American competitors.

我覺得這證明瞭中國的競技場深度很夠，不只是他們，還有很多不同的中國 AI 實驗室都發布了令人印象深刻的基準測試，而且似乎緊追在美國競爭對手之後。

So why was it that Deep Seek broke through?

那麼，為什麼 Deep Seek 能脫穎而出呢？

And yet certainly in the west, in Europe, in the us, no other big Chinese models have yet broken through the way that Deep Seq did.

然而，在西方、歐洲、美國，沒有其他中國的大型模型像 Deep Seq 那樣突破重圍。

I mean, here there's sort of like a psychology element and a sociological element to all this, because China's sort of depth in AI has been long established.

我的意思是，這裡有種心理和社會學的元素，因為中國在 AI 方面的深度早就建立起來了。

I mean, when you look at major international conferences like Neurips, you see a lot of representations from Chinese AI researchers.

我的意思是，當你看看像 Neurips 這樣主要的國際會議時，你會看到很多中國 AI 研究人員的代表。

But I think what DeepSeq did is they were one of the first to really show that a Chinese AI lab could nearly match the performance of their American peers and do so at a fraction of the cost, reportedly, and

但我認為 DeepSeq 所做的，是他們率先真正展示了一個中國的 AI 實驗室幾乎可以匹敵美國同行的表現，而且據說還能以更低的成本做到，而且

do this in the face of US export control.

在美國出口管制的壓力下也能做到。

So there's sort of a broader national dimension to all this.

所以這一切都有更廣泛的國家層面的意義。

You know, successfully developing is one thing, but successfully deploying is another.

你知道，成功開發是一回事，但成功部署又是另一回事。

So when we, when we look at the overall strategies that countries like the US And China are adopting, we see two very different approaches to this.

所以當我們，當我們看看像美國和中國這樣的國家採取的整體戰略時，我們會看到兩種截然不同的方法。

So in the US there's sort of an approach of, you know, US Companies will develop the most powerful models and then other people will figure out how to deploy those models very successfully in various applications.

所以美國的做法是，你知道，美國公司將開發最強大的 AI models，然後其他人將找出如何在各種應用中成功部署這些 AI models。

China is much more focused on figuring out what those use cases are.

中國則更專注於找出這些使用案例是什麼。

They have a very aggressive national strategy to boost adoption throughout their economy as a centralized government.

他們有一個非常積極的國家戰略，透過中央集權的政府來推動整個經濟的採用。

They have levers they can push that the US Government can't.

他們有美國政府無法使用的槓桿。

They're also very aggressively trying to expand adoption of Chinese technology and Chinese models throughout the world, particularly in Asia Pacific and African countries.

他們也正積極地試圖擴展中國技術和 Chinese models 在全世界的採用，尤其是在亞太和非洲國家。

Along with that, they're focusing on things like robotics and other sort of specific use cases.

另外，他們也專注於像是機器人和一些特定的使用案例。

They're working with governments to figure out how to make AI work for their specific needs.

他們正與各國政府合作，找出如何讓 AI 應用於他們特定的需求。

And so this is a very different strategy than we see in the U.S. i think U.S.

所以這跟我們在美國看到的策略很不一樣。我認為美國

companies have realized that they need to focus more on use cases.

的公司已經意識到，他們需要更專注於使用案例。

And so we see more hiring in areas in specific verticals, like in health care, AI and science, AI and financial services.

是以，我們看到在醫療保健、AI 與科學、AI 與金融服務等特定領域的招聘更多。

But we don't have the kind of concerted strategy in the US that China seems to have and trying to push adoption throughout their economy.

但是我們在美國沒有像中國那樣的協調一致策略，來推動整個經濟的採用。

So, look, we've seen obviously a massive leap in AI spending.

所以，你看，我們顯然看到了 AI 支出的巨大飛躍。

Not just the companies are developing these models, but

不僅是開發這些 AI models 的公司，

the companies who wish to adopt them as well and as well as consumer spending.

還有希望採用這些 AI models 的公司，以及消費者的支出。

Is this going to hit a wall at some point?

這會不會到某個時候就撞牆了？

First of all, there's a conversation about the bubble that's the elephant in the room at the moment.

首先，大家都在討論現在這個房間裡的大象，也就是保麗龍。

Are we in an AI bubble and is this expenditure suddenly going to collapse, or do you think that there's still plenty of room left to run?

我們是不是處在一個 AI 保麗龍裡，這種支出會不會突然崩盤，還是你覺得還有很大的成長空間？

So I think what we've seen is that the, the U.S.

所以我認為我們看到的是，美國

economy has some weak fundamentals, and that is.

經濟有一些基本面疲軟的地方，而這

And it's being propped up essentially by spending on AI and data centers in particular.

正在被 AI 和數據中心的支出所支撐。

I don't think anyone can tell you definitively whether we are in a bubble.

我覺得沒有人能肯定地告訴你我們是不是處在一個保麗龍裡。

I do think there are lots of questions about whether the level of investments we've seen, whether companies can make the returns that they need to pay off those investments.

我確實認為有很多問題，關於我們看到的投資水準，公司能不能獲得他們需要的報酬來償還這些投資。

Lots of concerns about circularity of deals and companies sort of investing in companies that then pay for services.

對於交易的循環性，以及公司投資那些又付費的公司，有很多擔憂。

If there's a bubble, there are a variety of things that could play out.

如果真的有保麗龍，會有很多種不同的情況發生。

It could be a slow bubble where we sort of see valuation sort of reset gradually over time, and

可能會是個緩慢的保麗龍，我們可以看到估值隨著時間慢慢調整，然後

we see essentially, you know, deployment catch up to the level of investment over a number of years.

我們基本上可以看到，像是，部署追上投資的水準，這需要好幾年的時間。

So there's no guarantee that a bubble will burst quickly.

所以沒有保證保麗龍會快速破裂。

But I think there are lots of questions that investors are asking about what level of returns, what level of adoption in the economy would justify the valuations we're seeing and the level of spending on data centers and the sustainability of that into the future.

不過我認為投資人有很多問題想問，關於投資報酬的水準、經濟上的採用率要多高，才能證明我們看到的估值、資料中心的支出水準，以及這些在未來是否能持續。

Yeah, Alok, you mentioned data centers and Kyle, I want to turn to you because the foundation of AI is important, right?

對，Alok，你提到了資料中心，Kyle，我想轉向你，因為 AI 的基礎很重要，對吧？

And there's all sorts of different like resources in my brain.

I kind of have this as like settlers of catan or something.

我覺得這就像是「卡坦島」那樣的東西。