00:00
🎙️
I want to pivot though to uh an Nvidia story with you Honor because we've been talking with Mandep your colleague uh in
我想轉向一個關於輝達（Nvidia）的新聞，Honor，因為我們一直在與彭博資訊（Bloomberg Intelligence）科技領域的同事 Mandep 討論，
00:07
🎙️
the tech coverage at Bloomberg Intelligence about um you know whether Nvidia can sell chips into China,
關於輝達是否能向中國銷售晶片，
00:13
🎙️
whether the Chinese will buy them.
以及中國是否會購買這些晶片。
00:18
🎙️
It seems that Beijing is trying to dissuade private companies from doing that.
北京方面似乎試圖勸阻私營企業這樣做。
00:22
🎙️
But it turns out that Deepseek um is planning on using Nvidia chips in its
但結果顯示，Deepseek 正計劃在其
00:25
🎙️
next model.
下一代模型中使用輝達晶片。
00:26
🎙️
So they clearly want to use uh this product and have some access to it.
所以他們顯然想要使用這項產品，並希望能取得一定的管道。
00:32
🎙️
>> See when it comes to training I mean I don't think there is a chip out there that's better than what Nvidia is
>> 就訓練而言，我認為市面上沒有比輝達提供的
00:36
🎙️
offering. So if you are trying to compete for the best Frontier model out there I mean you will be using those
產品更好的晶片。所以，如果你試圖競爭打造最頂尖的 Frontier 模型，無論你是否公開宣稱，
00:42
🎙️
chips one way or the other whether you publicize or or not.
你最終都會以某種方式使用那些晶片。
00:46
🎙️
Um very few companies have the access to Google TPUs for example.
很少有公司能取得 Google TPU 等資源。
00:46
🎙️
That's different.
這情況不同。
00:52
🎙️
Um, so Nvidia is the is the normal choice for them.
所以輝達對他們來說是正常的選擇。
00:55
🎙️
U especially for training.
尤其是在訓練方面。
00:56
🎙️
You know, for inference down the road, we could argue that, you know, you could use other uh
至於未來的推論（inference），我們可以說，你可以使用其他
00:59
🎙️
semiconductor devices, but you know, for now, we really need Nvidia for for training reasons.
半導體設備，但就目前而言，基於訓練的理由，我們真的需要輝達。
01:06
🎙️
Yeah, a really interesting story from the information on
是的，這來自 The Information 的一則非常有趣的新聞，
01:08
🎙️
>> President Trump decided to let Nvidia sell its H200 AI chips to China after concluding the move carried a lower
>> 總統川普決定允許輝達向中國銷售其 H200 AI 晶片，結論是此舉帶來的
01:14
🎙️
security risk because the company's Chinese arch rival Huawei already offers some AI systems with comparable
安全風險較低，因為該公司在中國的主要對手華為（Huawei）已經推出了一些性能相當的
01:22
🎙️
performance.
AI 系統。
01:23
🎙️
That's according to a Bloomberg source.
這是根據彭博社消息來源的說法。
01:25
🎙️
Bloomberg tech reporter Maggie Eastland joins us.
彭博科技記者 Maggie Eastland 加入我們的節目。
01:27
🎙️
Maggie, you and I broke this story together yesterday.
Maggie，昨天我們一起獨家報導了這則新聞。
01:30
🎙️
As we understand it, right, the president was presented with a range of options from exporting no
據我們了解，對吧，總統當時收到了一系列選項，從完全不出口技術
01:35
🎙️
technology at all to the latest technology and they landed somewhere in the middle.
到出口最新技術，最後他們選擇了中間路線。
01:40
🎙️
Take it from there and what else we reported and what we know about uh the Huawei Cloud Matrix 384 system.
請接著說明，我們還報導了什麼，以及我們對華為雲端 Matrix 384 系統的了解。
01:48
🎙️
>> Yes. One key detail we've reported is that this cloud matrix 384 system, which can link together hundreds of chips, was
>> 是的。我們報導的一個關鍵細節是，這個可以連結數百顆晶片的雲端 matrix 384 系統，
01:57
🎙️
a key rationale underpinning the White House's logic to allow these advanced chips, which are better on a perch chip
是白宮允許出口這些先進晶片背後的關鍵理由，因為這些晶片在單顆晶片的基礎上
02:03
🎙️
basis than what Chinese companies are capable of.
表現優於中國公司所能製造的產品。
02:07
🎙️
However, as was made clear in our reporting, the White House is looking at this more at a system level,
然而，正如我們的報導所清楚指出的，白宮現在是從系統層面來看待這個問題，
02:12
🎙️
and their understanding was that Huawei's systems are actually advancing quite quickly.
他們的理解是華為的系統其實進步得相當快。
02:17
🎙️
So that was the rationale here and and though this is a paradigm shift from previous administrations
這就是背後的理由，雖然這與之前的政府相比是一種典範轉移，
02:23
🎙️
during which Huawei's advancements were an impetus to actually crack down and further restrict those Chinese
在之前的政府任內，華為的進步是實際打壓並進一步限制那些中國
02:30
🎙️
companies.
公司的動力。
02:31
🎙️
Now that same evidence is being used as a reason to loosen controls and kick off this new strategy
現在，同樣的證據卻被用來作為放寬管制並啟動這項新策略的理由，
02:38
🎙️
of selling Nvidia chips to China.
也就是向中國銷售 Nvidia 晶片。
02:41
🎙️
The main focus as I understand it is that H200 is 18 months behind Blackwell generation to generation and that's the
據我了解，主要的考量點是 H200 落後 Blackwell 世代約 18 個月，而這就是
02:47
🎙️
comfort level this administration has.
目前政府感到安心的程度。
02:50
🎙️
Nvidia is down 1.3% in the session right now.
Nvidia 目前在本交易時段下跌 1.3%。
02:53
🎙️
The other big report out there is from the information which reported that deepseek smuggled in thousands of
另外一則重大報導來自 The Information，該媒體報導 deepseek 走私了數千顆
03:00
🎙️
Blackwell chips into China getting them from countries of origin where they were allowed dismantling the servers.
黑威爾晶片流入中國，是從允許拆解伺服器的原產國取得。
03:09
🎙️
Nvidia has come out with a statement, Maggie, pushing back on that report.
Nvidia 已發表聲明，Maggie，反駁該報導。
03:14
🎙️
Let's start with that Nvidia statement, please.
請先從 Nvidia 的聲明開始。
03:16
🎙️
>> Yes. So, Nvidia has said that these smuggling claims are a bit far-fetched in its own words.
>> 是的。Nvidia 表示，這些走私說法有點牽強。
03:24
🎙️
And while it investigates every tip it receives, it hasn't seen evidence of this.
雖然他們會調查收到的每一條線索，但尚未發現相關證據。
03:28
🎙️
Now, one interesting thing from the report, Jensen Wong has said before that these
這份報導中有一個有趣的點，Jensen Wong 曾說過這些
03:31
🎙️
Blackwell racks are just a bit too heavy to be feasible to smuggle.
黑威爾機架太重，不太可能走私。
03:37
🎙️
But this report from the information did say that the smuggling for deepseek took place in
但 The Information 的這份報導指出，DeepSeek 的走私是透過
03:40
🎙️
these sort of eight chip segments that could potentially fit in a suitcase.
這種八晶片模組進行，這些模組可能可以放進手提箱。
03:44
🎙️
>> You know, if in Nvidia has the potential to earn 750 per share from selling chips to China, which is a number I've seen,
>> 要知道，如果 Nvidia 透過向中國出售晶片，每股有潛力賺取 750 美元，這是我看過的數字，
03:54
🎙️
that's what the stock would earn in 2026.
那將是該股在 2026 年的獲利。
03:57
🎙️
>> Hey, how about this?
>> 嘿，這個怎麼樣？
04:01
🎙️
Um, for all the hoopla around this whole trade, which has dominated every conversation that
嗯，圍繞這整筆交易的所有喧囂，自年初以來，
04:06
🎙️
we've had for the most part since the beginning of the year, Google and or Alphabet and Nvidia are
幾乎主導了我們所有的對話，Google 和/或 Alphabet 以及 Nvidia 是
04:12
🎙️
the only two Mag 7 stocks that are outperforming the S&P this year.
今年表現超越 S&P 500 指數的僅有兩檔 Mag 7 股票。
04:16
🎙️
>> Yes. >> Surprising?
>> 是的。>> 令人驚訝嗎？
04:17
🎙️
>> It is. Um, I a little bit surprising.
>> 是的。嗯，有點驚訝。
04:22
🎙️
I think there is a dramatic bifurcation in terms of performance.
我認為在表現方面存在巨大的分歧。
04:26
🎙️
They are dramatically outperforming their MAG 7 peers.
它們的表現遠遠超過其他 Mag 7 成員。
04:28
🎙️
So you're talking about somewhere around 10 to 15% when you look at names like Microsoft and Apple and Meta and
所以當你看到 Microsoft、Apple 和 Meta 等股票時，它們大約領先 10% 到 15%，
04:35
🎙️
Tesla. So it it's clear that the marketplace has identified quote unquote the winner in the uh hyperscaler
Tesla。所以很明顯，市場已經在超大規模支出和 AI 光環交易中，選出了所謂的贏家。
04:44
🎙️
spending and the AI halo trade.
62.
04:49
🎙️
Early in the year it was absolutely Nvidia.
今年初，絕對是 Nvidia。
04:52
🎙️
Maybe that's waned slightly to the benefit of Broadcom.
也許熱度稍減，讓 Broadcom 受益。
04:54
🎙️
more recently, without a question, it is about Alphabet.
最近，毫無疑問，焦點轉向 Alphabet。
04:59
🎙️
And let's remember, Apple's had a heck of a run off of that August low.
而我們要記得，Apple 從八月低點開始，已經有一波非常強勁的漲幅。
05:02
🎙️
And it really began the year struggling.
它今年初其實表現掙扎。
05:03
🎙️
It's up about 40%.
現在漲幅大約 40%。
05:05
🎙️
So, I think that's the playbook for 2026.
所以，我認為這是 2026 年的操作策略。
05:08
🎙️
I don't think you're going to see a universal move in either direction for the totality of these max.
我不認為你會看到這些權值股整體朝同一個方向全面移動。
05:13
🎙️
So, how should investors understand that?
所以，投資人該如何理解這點？
05:15
🎙️
Because we got some news this week that sounded like relatively good news on China when it comes to uh chip
因為我們這週聽到一些關於中國的消息，在晶片限制方面，聽起來相對是好消息。
05:21
🎙️
restrictions.
73.
05:23
🎙️
And one might think, oh, well, that could be good for synapsis, but then again, the the wind has a way
有人可能會想，哦，那對 Synopsys 可能是好事，但話說回來，現在風向變得很快。
05:28
🎙️
of changing direction rather quickly these days.
75.
05:31
🎙️
>> Yes. So, you you're referring to the Nvidia H200, I'm assuming.
>> 是的。所以你指的是 Nvidia H200，我猜。
05:35
🎙️
>> Uh, yeah.
>> 嗯，對。
05:37
🎙️
Which is very good news.
這是個非常好的消息。
05:40
🎙️
And the reason is very good news because the more you can build the AI stack on the
而原因是非常好的消息是，因為你能在上面建置的 AI 堆疊越多，
05:44
🎙️
foundation silicon that is US-based like the the Nvidia or AMD or whomever it's uh good news because it has stickiness
像輝達或 AMD 等美國廠商所研發的矽晶片基礎是好消息，因為它具有黏著性
05:52
🎙️
to it because the more you develop software on the on the on the hardware layer that the US provide is good news
因為你在美國提供的硬體層上開發的軟體越多，對我們來說就越是好消息
06:00
🎙️
from a synopsis perspective.
從新思科技的角度來看。
06:04
🎙️
Uh Nvidia of course as they're designing the H200 or whatever chip they're shipping it's
輝達當然在設計 H200 或任何他們出貨的晶片時
06:07
🎙️
using synopsis technology.
都使用了新思科技的技術。
06:11
🎙️
Now inside China the race is on.
現在在中國，這場競賽已經開跑。
06:13
🎙️
There are many companies that are developing alternative chips and if these companies
有許多公司正在開發替代晶片，如果這些公司
06:17
🎙️
are not uh under uh government control or restrictions we can sell to but increasingly these companies we cannot
沒有受到政府控制或限制，我們就能銷售給他們，但越來越多這些公司我們無法
06:25
🎙️
sell to them and they're trying to find local alternatives to design these chips.
銷售給他們，而他們正試圖尋找當地的替代方案來設計這些晶片。
06:30
🎙️
Here's Gene, I want to ask you about Nvidia because you're actually very bullish, more bullish than the
Gene，這裡我想問你關於輝達的事，因為你其實非常看好，比市場
06:33
🎙️
street in fact when it comes to um sales for Nvidia next year.
更看好輝達明年的銷售表現。
06:38
🎙️
U in terms of this China, you know, the information reported that that these Blackwell chips
關於中國的部分，你知道，《資訊報》報導這些 Blackwell 晶片
06:42
🎙️
were basically being trans shipped.
基本上是被轉運。
06:45
🎙️
So shipped to countries that are allowed to have the Blackwell chips installed in
也就是運到被允許在資料中心
06:47
🎙️
data centers.
安裝 Blackwell 晶片的國家。
06:48
🎙️
They were then dismantled and shipped to China.
接著它們被拆解並運往中國。
06:51
🎙️
Um and so that's how they got around the export ban.
嗯，他們就是這樣規避出口禁令的。
06:54
🎙️
and separately the information was reporting that the Chinese government is going to
另外，《資訊報》也報導中國政府即將
06:56
🎙️
meet and decide basically whether or not to allow companies to buy the H200 chip.
開會並決定是否要允許公司購買 H200 晶片。
07:02
🎙️
So there's a lot of stuff going on sort of throwing into doubts that Nvidia will actually see any sort of a a China bump.
所以現在有很多事情發生，讓人懷疑輝達是否真的能因此受惠。
07:09
🎙️
>> Well, I think for starters, I mean the question, you know, did did uh Deep Seek train on Nvidia chips?
>> 嗯，我認為首先，我的意思是這個問題，你知道，Deep Seek 是否使用 Nvidia 晶片進行訓練？
07:16
🎙️
My sense is that yes, I would put a high probability that that happened.
我的感覺是，是的，我認為發生這種情況的可能性很高。
07:20
🎙️
there is this still disconnect.
這裡仍然存在一種脫節。
07:22
🎙️
A year ago when we saw the performance of deepseek there was this question about how did they do it and
一年前當我們看到 DeepSeek 的表現時，大家對於他們是如何做到的感到疑惑，
07:27
🎙️
and and I think that at uh the most basic level is that when you think about advances in AI the type of chips the
而我認為在最基本的層面上，當我們思考人工智慧的進步時，晶片的類型，
07:34
🎙️
GPUs are really important and more advanced GPUs are what drives uh better insights and of course deepseek had a
GPU 非常重要，更先進的 GPU 能夠驅動更好的洞察，當然 DeepSeek 擁有
07:42
🎙️
lot of insights and they they proclaimed a very low cost of training.
許多洞察力，並且他們宣稱訓練成本非常低。
07:47
🎙️
I don't think that that was representative of what was actually as part of it.
我認為那並不能完全代表實際的情況。
07:51
🎙️
And I think that there probably was Nvidia.
我認為他們很可能使用了 Nvidia。
07:51
🎙️
Nvidia is motivated to say that they didn't get those chips because of uh geopolitical reasons.
Nvidia 有動機說他們因為地緣政治原因而無法取得那些晶片。
07:57
🎙️
They don't want to say that that that that had happened.
他們不想承認那件事真的發生了。
08:01
🎙️
Um as far as this, you know, turn that we saw this week with uh Nvidia getting the
嗯，關於這週我們看到的轉折，Nvidia 獲得了
08:04
🎙️
green light to sell H20s and then all like almost instantaneously Chinese leadership said, "Well, maybe
銷售 H20 的許可，然後幾乎在瞬間中國領導層就說：「嗯，也許
08:12
🎙️
maybe we're not going to be buying as many of those chips as as you thought." I think that again it comes down to is
我們不會像你們想像的那樣購買那麼多晶片。」我認為這再次歸結為
08:19
🎙️
chips are important and the reason why that this has been held out as such a a carrot in the whole trade negotiations
晶片非常重要，而這也是為什麼在整個貿易談判中，
08:26
🎙️
and be even beyond trade is uh the reason why these H20 chips have been held out for that carrot is because they
甚至超越貿易層面，這些 H20 晶片一直被當作誘因的原因，是因為它們
08:33
🎙️
are important and so I see the commentary since Monday from China as more or less theater and I'm not sure
很重要，所以我認為從週一以來來自中國的評論或多或少是種戲碼，我不確定
08:41
🎙️
why they're doing it but more or less theater and ultimately they are going to be buying these chips And I think that
他們為什麼要這樣做，但這或多或少是種戲碼，最終他們還是會買這些晶片。我認為
08:46
🎙️
Nvidia is going to capture probably something like 30 billion in revenue.
Nvidia 的營收可能會因此增加約 300 億美元。
08:52
🎙️
So that should raise their growth rate from the streets looking for 51% to 65% or
所以這應該會將他們的成長率從分析師預期的 51% 提升到 65% 或是
08:56
🎙️
better. And um so I I think that and that by the way is a kind of a more conservative estimate.
更好。嗯，所以我想，順帶一提，這是一種比較保守的估計。
09:03
🎙️
So I think that at the end of the day is that these chips are important and Nvidia is going
所以我認為歸根結底，這些晶片很重要，而輝達明年
09:07
🎙️
to do some business in China in the next year.
在中國還是會有生意。
09:09
🎙️
Um, Jean, does the H200 Okay, let's just say Nvidia is able to sell them.
嗯，Jean，關於 H200，好吧，假設輝達能夠賣它們。
09:15
🎙️
Let's also assume that they have plenty of them, right?
我們也假設他們有充足的貨源，對吧？
09:18
🎙️
And and that's the thing we were talking about last night.
這也是我們昨晚在討論的事情。
09:19
🎙️
There is a black market for these.
這些晶片存在黑市交易。
09:21
🎙️
These are, you know, these chips are being deployed in Malaysia and Singapore.
這些晶片，你知道的，正在馬來西亞和新加坡部署。
09:24
🎙️
They're they're getting used, right?
它們正在被使用，對吧？
09:25
🎙️
And Deepseek was probably trained on them.
Deepseek 可能就是用這些晶片訓練出來的。
09:27
🎙️
But how much of an advancement is Blackwell, the 200 there to, let's say, the H200.
但是 Blackwell 相較於 H200，到底有多少進步？
09:32
🎙️
Does it really matter if we're 18 months ahead of them training our models?
如果我們在模型訓練上領先他們 18 個月，這真的重要嗎？
09:35
🎙️
And obviously, this is most important for training the models, not for inference.
顯然，這對訓練模型最重要，而不是對推論（inference）。
09:40
🎙️
And so I I guess if we had that wide of a gap, does it really matter if we're selling the H200 to
所以我想，如果我們有這麼大的差距，我們把 H200 賣給
09:45
🎙️
>> I I don't think it matters that I think there's an opportunity for them to for us to sell for Nvidia to sell H the the
>> 我認為這不重要，我認為這對輝達來說是一個機會，可以賣出降規版的 H200。
09:52
🎙️
the powered down versions.
所以，嗯。
09:55
🎙️
So uh but as far as like does it matter having the most advanced chips in the grander
但就更宏大的 AI 策略而言，擁有最先進的晶片是否重要？
09:59
🎙️
scheme of of AI?
嗯，我認為答案是肯定的，很重要。
10:01
🎙️
Um and I think the answer is yes, it does.
翻譯中...
10:04
🎙️
I mean I think that that's just uh having the most advanced chips.
我的意思是，我認為這只是呃擁有最先進的晶片。
10:07
🎙️
Think about the power on these is somewhere in a magnitude of four to eight times each generation in
想想看，每一代晶片的性能提升大約是四到八倍，
10:12
🎙️
terms of its compute and battery u improvements or not pattern improvements power improvements.
在運算能力、電池改進，或者說是功耗改進方面。
10:19
🎙️
That is a 4 tox8 improvement uh uh cycle over cycle and so uh why that matters is there is a
這意味著每一代的性能都有 4 到 8 倍的提升循環，所以呃為什麼這很重要呢？因為這是一場
10:25
🎙️
race to get to general intelligence.
邁向通用人工智慧的競賽。
10:29
🎙️
I think there is an inherent benefit to being with some of the first companies
我認為，身為最先抵達那裡的
10:31
🎙️
that get there.
幾家公司之一，具有內在的優勢。
10:33
🎙️
Uh and so I think it does matter.
呃所以我認為這確實很重要。
10:35
🎙️
And why does it matter one two years at getting there?
為什麼早個一兩年抵達那裡很重要呢？
10:38
🎙️
You know, that that can make a big difference when you think about how the world is
當你思考世界的結構時，你知道，這會造成巨大的差異。
10:41
🎙️
structured.
151.
10:42
🎙️
And and no, >> Jane, just to be really clear, I don't mean it doesn't matter for the training.
而且，>> Jane，為了讓大家更清楚，我並不是說這對於訓練不重要。
10:46
🎙️
What I mean is the the um advance that we have in them, like if you have national security concerns, that that's
我的意思是，我們在這方面擁有的優勢，如果你有國家安全方面的考量，那就是
10:51
🎙️
what I meant.
我的意思。
10:52
🎙️
So, um No, that's fine.
所以，呃沒關係。
10:55
🎙️
I just want to clarify.
我只是想澄清一下。
10:55
🎙️
>> Great answer.
>> 回答得很好。
10:56
🎙️
>> All right, Jean.
>> 好的，Jean。
10:58
🎙️
Great to see you.
很高興見到你。
10:59
🎙️
>> All right.
>> 好的。
11:00
🎙️
I hope you're all doing well today and staying calm in this market.
希望大家今天一切安好，在這波市場中保持冷靜。
11:02
🎙️
We have a lot of Nvidia news to cover.
我們今天有許多 Nvidia 的新聞要探討。
11:03
🎙️
To start things off, we got two articles from the information today.
首先，我們今天收到了來自 The Information 的兩篇文章。
11:07
🎙️
I want to preface by saying that I'm always hesitant to believe anything that the
我要先聲明，基於過往的紀錄，我對於 The Information 報導的任何關於 Nvidia 的傳聞，
11:09
🎙️
information reports having to do with Nvidia rumors given their track record.
總是抱持著懷疑的態度。
11:12
🎙️
But I'm bringing these stories to your attention so that you're aware of what's going on.
但我還是把這些消息帶給大家，好讓各位知道目前的情況。
11:15
🎙️
The first story from information claims that Chinese regulators convened emergency meetings
The Information 的第一篇報導聲稱，中國監管機構召集了緊急會議，
11:19
🎙️
with major tech firms such as Alibaba, Bite Dance, and Tencent to decide whether they should be allowed to
與阿里巴巴、字節跳動和騰訊等大型科技公司會面，以決定是否應該允許他們
11:24
🎙️
purchase Nvidia's H200.
購買 Nvidia 的 H200。
11:26
🎙️
Make no mistake, these companies absolutely want access to Nvidia's chips.
毫無疑問，這些公司絕對想取得 Nvidia 的晶片。
11:29
🎙️
Now, the question is whether the government will allow it.
現在的問題是，政府是否會允許。
11:31
🎙️
Interestingly, Reuters published an article today saying that their review of 100 tenders and academic papers show
有趣的是，路透社今天發布了一篇文章，指出他們審查了 100 份投標書和學術論文後發現，
11:36
🎙️
that Nvidia's H200 is already being supplied to Chinese customers through the gray market.
Nvidia 的 H200 已經透過灰色市場供應給中國客戶。
11:41
🎙️
Along that same vein, the information published a second article today claiming that Deep Seek is
同樣地，The Information 今天發布了第二篇文章，聲稱 Deep Seek
11:45
🎙️
using Nvidia Blackwell chips that were smuggled into China via other countries.
正在使用透過其他國家走私到中國的 Nvidia Blackwell 晶片。
11:49
🎙️
The report claims that the chips were installed in data centers in those other countries before being dismantled and
報導稱，這些晶片在被拆解並
11:53
🎙️
shipped to China.
運往中國之前，已經安裝在其他國家的數據中心裡。
11:55
🎙️
We already know that Deepseek has many Nvidia Hopper GPUs as we learned earlier this year.
我們早就知道 Deepseek 擁有許多 Nvidia Hopper GPU，正如我們今年稍早所學到的。
11:59
🎙️
But Blackwell is another question.
但 Blackwell 則是另一個問題。
12:01
🎙️
And it's important to note that a Blackwell NVL72 system is much much larger than just a
需要注意的是，Blackwell NVL72 系統的規模遠遠大於單純的
12:05
🎙️
Blackwell GPU.
Blackwell GPU。
12:06
🎙️
I'm hesitant to believe that they would be able to secretly smuggle massive Blackwell systems in
我很懷疑他們是否能夠秘密走私大量
12:10
🎙️
meaningful quantities.
的 Blackwell 系統。
12:12
🎙️
But of course, I could be wrong about that.
當然，這方面我可能也判斷錯誤。
12:14
🎙️
As I said, I'm always hesitant to believe anything the information reports regarding Nvidia
如我所言，基於過往的紀錄，我對於資訊報導中有關 Nvidia
12:16
🎙️
rumors given their track record.
的傳聞總是抱持保留態度。
12:18
🎙️
Following the information's article about Deep Seek, Nvidia responded to the report by saying, quote, "We haven't
針對資訊網站關於 Deep Seek 的文章，Nvidia 回應該報告時表示：「我們
12:22
🎙️
seen any substantiation or received tips of phantom data centers constructed to deceive us and our partners, then
尚未看到任何具體證據，也未收到任何關於設立虛假數據中心以欺騙我們及合作夥伴，之後
12:28
🎙️
deconstructed smuggled and reconstructed somewhere else.
再拆除、走私並於他處重建的線索。
12:32
🎙️
While such smuggling seems far-fetched, we pursue any tip we receive.
儘管此類走私行為看似匪夷所思，我們仍會追查收到的每條線索。」
12:34
🎙️
This comes one day after US authorities announced they shut down another China Link smuggling
此聲明發出的前一天，美國當局才宣布破獲另一個「中國連結」（China Link）走私
12:38
🎙️
network that attempted to smuggle more than $160 million worth of Nvidia chips to China.
網絡，該網絡試圖將價值超過 1.6 億美元的 Nvidia 晶片走私至中國。
12:43
🎙️
It's important to note that this story has to do with Nvidia's H100 and H200 GPUs, not Blackwell.
需要注意的是，此事件涉及的是 Nvidia 的 H100 和 H200 GPU，而非 Blackwell。
12:50
🎙️
And speaking of Blackwell, Reuters published an article today saying that Nvidia has
而說到 Blackwell，路透社（Reuters）今日刊文指出，Nvidia 已
12:52
🎙️
privately developed location verification technology.
私下開發了位置驗證技術。
12:56
🎙️
According to Reuters, this technology is a software option that customers could install.
據路透社報導，這是一項客戶可安裝的軟體選項。
12:58
🎙️
While the software was built to allow customers to track computing performance, it also provides a sense of
雖然該軟體旨在讓客戶追蹤運算效能，但它也能提供
13:03
🎙️
the chip's location.
晶片位置的相關資訊。
13:05
🎙️
According to Nvidia, this customerinstalled software agent leverages GPU telemetry to monitor
據 Nvidia 稱，這款由客戶安裝的軟體代理程式利用 GPU 遙測技術來監控
13:09
🎙️
fleet health, integrity, and inventory.
設備運行狀況、完整性與庫存管理。
13:11
🎙️
This report is very interesting given all of the backdoor and security concerns that Chinese regulators had
這份報告非常有趣，尤其考量到今年初中國監管機構對Nvidia晶片的後門與安全疑慮，
13:16
🎙️
earlier this year regarding Nvidia chips following the reversal of the H20 export control.
當時H20出口管制剛被撤銷。
13:20
🎙️
It's also worth noting that this new software service appears to be a customerinstalled agent, meaning that
同樣值得注意的是，這項新軟體服務似乎是客戶端安裝的代理程式，這意味著
13:24
🎙️
customers can choose to install it if they want to, but they may not be required to.
客戶可視需求選擇安裝，但並非強制要求。
13:28
🎙️
Reuters says this feature has not been released yet, but it will be made available on Nvidia's Blackwell
路透社指出此功能尚未正式發布，但將會應用於Nvidia的Blackwell
13:32
🎙️
chips. Regardless of what happens with the China situation, the good news is that Nvidia's business is accelerating
晶片上。無論中國市場局勢如何發展，好消息是Nvidia的業務正在加速成長，
13:37
🎙️
even without having any exposure to China.
即便完全沒有中國市場的曝險。
13:39
🎙️
And that is what we should be focused on from the perspective of long-term investors.
這正是長期投資人應該關注的重點。
13:43
🎙️
You may remember that I said repeatedly in recent weeks that if there was any opening for Nvidia
大家可能記得我過去幾週多次提到，只要Nvidia在中國市場出現任何銷售機會，
13:46
🎙️
to sell in China, we would most likely see a notable move in the stock and most analysts would promptly upgrade their
股價很可能就會出現明顯波動，且多數分析師會迅速調升
13:51
🎙️
Nvidia price targets.
Nvidia的目標價。
13:52
🎙️
So why has none of that happened?
為什麼這些情況都還沒發生？
13:54
🎙️
It's because there are still multiple unknowns about whether Nvidia will be allowed to sell in China
關鍵在於Nvidia能否獲准在中國銷售仍存在多重變數。
13:57
🎙️
or not. The US greenlighting the sale of H200 is the first step in the process.
美國批准H200銷售只是第一步。
14:02
🎙️
The next question is whether or not Chinese regulators will allow the chips to be sold.
接下來的問題是中國監管機構是否會批准這些晶片銷售。
14:06
🎙️
We also have to see what happens with Congress because a bipartisan group of senators recently
我們還得觀察國會動向，因為近期兩黨參議員共同提案立法，
14:09
🎙️
introduced legislation that would block the sales of Nvidia's Blackwell and H200 China for 30 months.
擬將Nvidia的Blackwell與H200在中國的銷售禁令延長30個月。
14:15
🎙️
And then even if both the US and Chinese governments approve of the sales, there are
即便美中兩國政府都批准銷售，
14:18
🎙️
questions about H200 capacity at TSMC.
台積電的H200產能也是問題。
14:21
🎙️
Capacity is tight and much of the H200 capacity has already been transitioned to other products and so there are still
產能吃緊，且 H200 的大部分產能已轉移至其他產品，因此短期內仍有
14:27
🎙️
a lot of unknowns in the short term.
許多不確定性。
14:28
🎙️
Now, as I said previously, if there's an opening for Nvidia to sell in China, meaning that all of these issues get
現在，正如我先前所說，如果輝達能在中國銷售，意即所有這些問題都
14:33
🎙️
resolved, then at that point, we would most likely see a notable move in the stock and most analysts would promptly
獲得解決，那麼在那個時間點，我們很可能會看到股價出現顯著波動，且大多數分析師會迅速
14:38
🎙️
upgrade their price targets.
上調他們的目標價。
14:40
🎙️
Right now, many questions remain, but again, Nvidia's business is growing strong,
目前仍有許多疑問，但再次強調，輝達的業務依然強勁增長，
14:43
🎙️
even without China sales, and that is what we should be focusing on.
即使沒有中國銷售額也是如此，而這正是我們應該關注的重點。
14:47
🎙️
In other news, Nvidia backed startup StarCloud has trained the first AI model in space
另外，輝達投資的新創公司 StarCloud 已在太空中使用輝達的 H100
14:50
🎙️
using Nvidia's H100.
訓練出首個 AI 模型。
14:52
🎙️
Data centers in space is such an interesting concept as we've briefly talked about in the past
太空資料中心是個非常有趣的概念，正如我們過去曾簡短討論過，
14:55
🎙️
and the concept has been getting a lot more attention recently.
而這個概念最近獲得了更多關注。
14:59
🎙️
In space, companies have access to near infinite solar energy and they also don't have to
在太空中，企業能獲取近乎無限的太陽能，且由於深空的真空環境，
15:02
🎙️
worry about access to water for cooling because of the vacuum of deep space.
他們也不必擔心冷卻用水的問題。
15:06
🎙️
I think we're going to hear a lot about data centers in space over the coming
我認為在未來幾年，我們將會聽到很多關於太空資料中心的消息。
15:08
🎙️
years. Also, Oracle reported earnings today with EPS beating estimates and revenue coming in slightly below
此外，甲骨文（Oracle）今日公布財報，每股盈餘（EPS）超出預期，但營收略低於
15:15
🎙️
Oracle's remaining performance obligations grew to a massive $523 billion, which is up 438%
甲骨文的剩餘履約義務（RPO）增長至驚人的 5,230 億美元，年增率達 438%。
15:20
🎙️
year-over-year.
238.
15:21
🎙️
In the earnings release, we got commentary from Oracle chairman and CTO Larry Ellison saying, quote, "We
在財報電話會議中，甲骨文董事長兼執行長賴瑞・艾利森（Larry Ellison）表示：「我們
15:26
🎙️
no longer think it is strategic for us to continue designing, manufacturing, and using our own chips in our cloud
不再認為自行設計、製造並在雲端中使用自家晶片對我們具有戰略意義。
15:31
🎙️
data centers.
資料中心。
15:32
🎙️
We are now committed to a policy of chip neutrality where we work closely with all our CPU and GPU
我們現在致力於執行晶片中立政策，與我們所有的 CPU 和 GPU
15:36
🎙️
suppliers.
供應商密切合作。
15:37
🎙️
Of course, we will continue to buy the latest GPUs from Nvidia, but we need to be prepared and able to
當然，我們會繼續向輝達（Nvidia）購買最新的 GPU，但我們必須做好準備，並有能力
15:41
🎙️
deploy whatever chips our customers want to buy.
部署客戶想要購買的任何晶片。
15:43
🎙️
There are going to be a lot of changes in AI technology over the next few years, and we must remain agile in
未來幾年 AI 技術將發生許多變化，我們必須保持敏捷以
15:47
🎙️
response to those changes." The AI buildout continues.
因應這些變化。」AI 基礎建設持續進行中。
15:51
🎙️
As I've discussed in previous videos, while there are similarities between this AI revolution
正如我在之前的影片中所討論的，雖然這場 AI 革命
15:54
🎙️
and the dotcom bubble, there are also major fundamental differences.
與網路泡沫有相似之處，但也存在重大的根本性差異。
15:58
🎙️
During the.com bubble, there was an overbuild of supply and a lack of demand.
在網路泡沫時期，供應過剩而需求不足。
16:02
🎙️
Today, it's the complete opposite.
今天的情況則完全相反。
16:03
🎙️
Today, we're seeing a lack of supply in the face of overwhelmingly strong demand.
如今，我們看到在極其強勁的需求面前，供應出現短缺。
16:08
🎙️
Now, will certain parts of the AI ecosystem eventually enter into bubble territory?
那麼，AI 生態系統的某些部分最終會進入泡沫領域嗎？
16:11
🎙️
Sure, some probably will, but that does not mean that Nvidia is a bubble, and it does not mean that AI as a whole is a
當然，有些部分可能會，但這並不意味著輝達是泡沫，也不代表 AI 整體是
16:16
🎙️
bubble. Will there eventually be an overbuild of capacity?
泡沫。產能最終會出現過剩嗎？
16:19
🎙️
Possibly at some point, but I honestly think we are at least a few years away from something
在某個時間點有可能，但我誠實地認為，距離發生這種事
16:22
🎙️
like that happening, if it happens at all.
我們至少還有幾年的時間，如果真的會發生的話。
16:23
🎙️
For right now, the world is severely compute constrained in the face of very strong AI demand, and that will
目前，在非常強勁的 AI 需求面前，全球嚴重面臨運算能力受限的問題，而這種情況
16:28
🎙️
likely continue over the next few years.
可能會在未來幾年持續下去。
16:30
🎙️
We also have to consider that new use cases will be developed in the coming years, which will increase the demand
我們還必須考慮到，未來幾年將會開發出新的應用案例，這將增加需求。
16:34
🎙️
for compute even further.
用於進一步運算。
16:36
🎙️
I don't know what the stock will do in the short term, but I continue to be bullish on
我不知道該股票短期會怎麼走，但我從長期角度繼續看好
16:38
🎙️
Nvidia from a long-term perspective.
Nvidia。
16:41
🎙️
I also continue to think that most analysts long-term Nvidia estimates are
我也持續認為，多數分析師對 Nvidia 的長期預估
16:44
🎙️
too low given what's ahead today.
在當前前景下是太低了。
16:47
🎙️
Most analysts are hyperfocused on Nvidia's data center business, especially the
多數分析師過度關注 Nvidia 的數據中心業務，尤其是
16:49
🎙️
cloud service providers.
雲端服務供應商。
16:51
🎙️
But we have to consider that Nvidia has many other large long-term growth opportunities in
但我們必須考慮到，Nvidia 還有許多其他龐大的長期成長機會
16:54
🎙️
front of it that are in addition to their massive hyperscalerdriven data center business, sovereign AI,
在眼前，這些機會是除了其大規模由超大規模客戶驅動的數據中心業務之外的，包括主權 AI、
16:58
🎙️
enterprise AI, many AI native startups, quantum GPU, supercomputing, and physical AI.
企業 AI、許多 AI 原生新創公司、量子 GPU、超級運算與實體 AI。
17:03
🎙️
While I expect Nvidia's data center business to continue growing strong, I think the long-term growth of
雖然我預期 Nvidia 的數據中心業務將持續強勁成長，但我認為 Nvidia 其他
17:07
🎙️
some of Nvidia's other business segments and customer segments are being overlooked today.
業務部門與客戶群的長期成長在當前正被忽略。
17:12
🎙️
As for physical AI, Nvidia CFO recently said that physical AI addresses quote a multi- trillion
至於實體 AI，Nvidia 財務長近期表示，實體 AI 處理的是「數兆美元」的機會，
17:16
🎙️
dollar opportunity and the next leg of growth for NVIDIA.
也是 Nvidia 下一波的成長動能。
17:19
🎙️
This is not getting enough attention, but I think it will get a lot of attention in 2026 and 2027.
這尚未獲得足夠關注，但我認為在 2026 與 2027 年將會受到大量關注。
17:23
🎙️
Nvidia sells the hardware for the data centers where the models are trained.
Nvidia 販售用於訓練模型的數據中心硬體。
17:26
🎙️
They offer omniverse where the models are taught and tested.
他們提供 Omniverse 來教導與測試模型。
17:30
🎙️
And Nvidia also sells the hardware that allows ondevice real-time inference through NVIDIA AGX
Nvidia 也販售能透過 NVIDIA AGX 進行裝置端即時推論的硬體，
17:33
🎙️
allowing robots to have intelligent interactions with the real world even when they are not connected to a data
讓機器人即使未連線至數據中心，也能與現實世界進行智慧互動。
17:38
🎙️
center. Notice that Nvidia is taking a holistic platform approach to physical AI and they're embedding themselves as
注意，Nvidia 正採取全方位的平台策略來發展實體 AI，並讓自己成為
17:43
🎙️
the underlying foundation supporting all of it.
支撐這一切的基礎。
17:46
🎙️
Over 2 million developers are already building on the Nvidia robotic stack and this is not getting enough
超過 200 萬名開發者已經在 Nvidia 機器人平台上進行開發，但這一點還沒有得到足夠的關注。
17:51
🎙️
I continue to think that most analysts long-term Nvidia estimates are too low given what's ahead.
我仍然認為，考慮到未來的發展，大多數分析師對 Nvidia 的長期預估都太低了。
17:56
🎙️
As for production ramps, Blackwell Ultra is ramping very quickly right now leading
關於生產 ramp up（產能爬坡），Blackwell Ultra 目前正在快速 ramp up，帶動
17:59
🎙️
to revenue growth acceleration.
營收成長加速。
18:02
🎙️
Reuben is on track to launch in 2026.
Reuben 將按計畫於 2026 年推出。
18:04
🎙️
Then we're expecting Reuben CPX at the end of 2026.
接著我們預期在 2026 年底推出 Reuben CPX。
18:05
🎙️
Later on, we're expecting the launch of Reuben Ultra in 2027 and Fineman after that in 2028.
之後，我們預期在 2027 年推出 Reuben Ultra，接著在 2028 年推出 Fineman。
18:11
🎙️
We have a clear data center product roadmap stretching into 2028.
我們有一條清晰的資料中心產品路線圖，一直延伸到 2028 年。
18:15
🎙️
And Jensen believes there will be three to four trillion dollars of global AI factory
而 Jensen（黃仁勳）相信，從現在到 2030 年間，全球將有 3 到 4 兆美元的 AI 工廠
18:18
🎙️
buildout between now and 2030.
建置潮。
18:22
🎙️
That means Jensen is expecting growing AI demand and an expanding total
這意味著 Jensen 預期 AI 需求將持續增長，且總體可尋址市場將不斷擴大，
18:24
🎙️
addressable market underpinning all of this.
這一切都有其支撐。
18:26
🎙️
I don't think we are anywhere near any type of bubble bursting type of event.
我認為我們離任何類型的泡沫破裂事件都還很遠。
18:29
🎙️
With all of this in mind, I seriously think that Nvidia still has plenty of runway ahead of it, and I
綜合以上所有因素，我真心認為 Nvidia 前方仍有廣闊的發展空間，而且我
18:33
🎙️
think this company will be worth substantially more in future years than it is today.
認為這家公司在未來幾年的價值將遠高於現在。
18:37
🎙️
At least that's my view of the situation.
至少這是我對目前局勢的看法。
18:39
🎙️
Remember to stay calm in this market.
記得在這個市場中保持冷靜。
18:41
🎙️
Remember to maintain a long-term perspective, and do not make any hasty or irrational decisions.
記得保持長期視角，不要做出任何草率或不理性的決定。
18:46
🎙️
With all of that being said, I hope you all have a great rest of the day, and I'm
說了這麼多，希望大家今天接下來的時間一切順利，我也
18:48
🎙️
curious to hear your thoughts about Nvidia in the comments below.
很想在下方留言區聽聽各位對輝達（Nvidia）的看法。
18:51
🎙️
Please leave a like on this video so more people will see it.
請幫這支影片按個讚，讓更多人能看到它。
18:54
🎙️
And while you're down there, please consider subscribing.
順便也請考慮訂閱我們的頻道。
18:55
🎙️
It's free, and you can always change your mind.
這是免費的，而且你隨時都可以取消訂閱。
18:58
🎙️
Thanks for watching and hopefully I'll see you in the next
感謝收看，希望下一集影片能再見到各位。
