YT YouTube2Text

YouTube2Text → Transcripts → 黃仁勳認出軌不認48億美元的賬！黃樂不是間諜是白手套！570萬買48億芯片阿里出錢算力租回中國！循環融資裂了雲巨頭年缺2000億才夠本！中國收入記零兩到四成靠中國！美人計？是買賣！

黃仁勳認出軌不認48億美元的賬！黃樂不是間諜是白手套！570萬買48億芯片阿里出錢算力租回中國！循環融資裂了雲巨頭年缺2000億才夠本！中國收入記零兩到四成靠中國！美人計？是買賣！ — Transcript

by 一个狠人 · 4,523 words · 693 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Official information is what it is;
0:01don't settle for casual assumptions.
0:02It’s time to catch everyone up on the
0:03gossip about Boss Huang; this situation
0:04is the perfect opportunity to dive into
0:05the industry scoop. Let's briefly
0:07summarize what happened. Although this
0:08story has been circulating for two days
0:10, Jensen Huang wrote a letter to all
0:11employees the day before yesterday. What
0:13did he admit? He admitted to personal
0:14indiscretions, saying he let down his
0:16family and everyone else. Why does this
0:18sound like something a gambler in a
0:19mid-life crisis would say? On the same
0:21day, Nvidia's board announced the
0:22formation of a special committee to
0:24investigate one thing: assessing whether
0:26their own leader had become a target of
0:28foreign intelligence operations. Just a
0:30few days prior, the chairs of three U.S.
0:32House committees sent him a joint letter
0:34, demanding he provide a closed-door
0:36briefing on October 18 th—a date that is
0:38quite interesting, as it coincides
0:40exactly with the AI fair in Shenzhen.
0:41The whole internet is telling the same
0:43story these days: a beautiful spy using
0:45a honey trap to take down a billionaire
0:47. We won't cover that today; instead,
0:49we’ll look at the ledger. My assessment
0:51? This female protagonist is not a
0:53lone-wolf spy, and Huang is not just
0:55some leather-jacketed old lecher blinded
0:57by beauty. What is this, really? It’s a
0:59deal where both sides calculated
1:00everything clearly. The subject? $ 4.8
1:02billion worth of chips. Why did he feel
1:04compelled to make this deal? The answer
1:05lies at the peak of this AI cycle.
1:07Before we begin, let’s clearly define
1:09the boundaries of the evidence. Where
1:10did this "spy" label come from? It
1:12started with media reports citing a
1:14classified U.S. intelligence memo, still
1:15speculating that she was recruited by
1:17Chinese state security. Yet, to this day
1:19, neither the U.S. nor Singapore has
1:21charged her with espionage. Singapore’s
1:23Ministry of Home Affairs even issued a
1:24clarification. They stated they never
1:26investigated her as a spy, and the U.S.
1:28is merely investigating the company
1:30behind her. So, is she a spy? That’s not
1:32something we can definitively answer,
1:33but the path she’s walked cannot be
1:35hidden. Today, we aren't chasing spy
1:36rumors; we are following the paper trail
1:38—from Chinese court rulings and exchange
1:40announcements to business records in
1:41Singapore and Malaysia, plus a Wall
1:43Street short-seller's investigative
1:44report. Of course, short-selling
1:46agencies make money by stirring up
1:47trouble. I’ll pick out the information
1:48they’ve released to discuss separately,
1:50and you can decide for yourselves what
1:51to believe. Let’s start with the
1:52protagonist, Huang Le, who goes by the
1:53English name Ice. She’s in her early 40s
1:55and is from Hunan. What kind of status
1:56does she hold? She has permanent
1:57residency in Hong Kong. What did her
1:59early resume claim? She claimed to have
2:01been a reporter for Hunan Satellite TV
2:02and CCTV, but when the New York Times
2:03later asked Hunan Satellite TV, the
2:04answer they got was that they had never
2:05heard of this person. Her subsequent
2:07path is even more interesting. According
2:09to her own public resume, as a liberal
2:10arts master’s graduate in media, she
2:12went straight into a foreign private
2:13bank in 2008, worked her way up to Vice
2:14President, and then jumped to CITIC
2:16Securities in 2013 to be a VP in finance
2:17. Today I’m going to share some gossip
2:19with you: what exactly do these two
2:20roles do in an investment bank? Anyone
2:22who understands the rules of Chinese
2:23investment banking knows how these
2:24private banking and wealth management
2:25businesses really work. Most of the time
2:27, they just hire an outsider with
2:28connections and slap a "VP"—that is,
2:30Vice President—title on them. What do
2:32they hire them for? It’s not to have you
2:33sit in an office calculating models. The
2:35unspoken rule is to take business cards
2:36to dinner parties, accompany the big
2:37shots, and build networks. What was the
2:38Hong Kong private bank she was in really
2:39doing at the time? It was helping
2:41mainland bosses move money across
2:42borders. And the department at CITIC
2:43handling stock pledges—the one lending
2:44to big players with leverage—what kind
2:45of work was that really? It was building
2:47bridges between the powerful and the
2:48capital. So her real skill was never
2:49finance; it was making connections. To
2:51put it bluntly, what was she? She was a
2:53broker. And it was at CITIC that she met
2:55her future boss. When he was at Shanghai
2:57Hexi Investment, Kang Wei was once known
2:58as the youngest mutual fund manager in
3:00China. After the 2015 stock market crash
3:02, she left CITIC and went straight to
3:03Hexi as an Executive Director. What role
3:05did she actually play at Hexi? This
3:07requires discussing another history of
3:08litigation. Hexi later got into a bonus
3:10lawsuit with another partner. To win,
3:12they submitted internal documents to the
3:14court. What did the documents show? Kang
3:16Wei made the final call to designate
3:17that 25%of Hexi's equity be transferred
3:19to Huang Le's name for her to hold. In
3:202016, he partnered with Kang Wei’s
3:22company to register a new firm with a
3:23registered capital of 600 million, and
3:24he served as the legal representative.
3:26Just one month later, this new company
3:28took 48%of Hexi’s shares. In the
3:30evidence, what was Hexi directly called
3:31? It was called the holding platform for
3:33legal representative Huang Le. Now,
3:34here’s the issue: he had four funds
3:36under his name, but in the fund filing
3:37records, he didn't even have the
3:39qualifications to manage them. Despite
3:40lacking the credentials, he was listed
3:42as the manager, and nearly half of
3:43Hexi’s shares were held in his name.
3:45This type of arrangement is a classic "
3:46white glove" tactic—putting it bluntly,
3:48he was just a frontman to hold shares
3:49and sign documents for the real boss.
3:51There’s no need to elaborate on his
3:52relationship with Kang Wei; seeing this
3:54, I can conclude that Huang is a
3:55template—a template for someone who can
3:56go to any lengths and climb the ladder
3:58through men. This is the ecosystem of
3:59China's investment banking sector. By
4:01the way, I’d like to correct a widely
4:03circulated claim online. They say that
4:04for this 25%equity, he directly went to
4:06court and had a falling out with Kang
4:07Wei. In reality, the verdict clearly
4:09states that the one who sued Kang Wei
4:10was another partner, and he was only
4:11mentioned in the evidence as the
4:13transferee. So, what is a "white glove’s
4:15" most useful skill? It’s being able to
4:17ensure the boss behind the scenes stays
4:18clean when trouble arises. He once
4:20helped Feicui’s boss, Guo Ping, leverage
4:21a buyout of an A-share listed company,
4:22involving 1.8 billion in total, and even
4:24the shareholding was...So much so that
4:25not a single public announcement was
4:27ever made. In early 2018, an annual
4:28meeting resulted in fines; Hexi and Kang
4:30Wei were fined 18.1 million. The mess
4:31Feicui later got into was even bigger.
4:33Feicui later tied itself to a wealth
4:34management platform called something
4:35like Lin Finance, which eventually
4:37collapsed. The court determined the
4:38illegal fundraising amount was over 83
4:39billion, and Guo Ping was sentenced to
4:41life in prison. As for Huang Le, who
4:42directly participated in the project, he
4:44withdrew from the holding platform in
4:45July 2017 before the fines were issued,
4:47and he also stepped down as the legal
4:48representative. The person who took over
4:50was also surnamed Huang, named Huang
4:52Yuyu. The business registry shows that
4:54these two individuals co-founded a
4:55company over a decade ago, which is why
4:57his name never appeared on any penalty
4:59lists or in subsequent legal proceedings
5:01. He simply used his Hong Kong identity
5:03to move on. Let’s look at his first
5:05venture after going overseas, and we’ll
5:07clearly see how he pulled this off. The
5:09story begins with a small gaming company
5:11. In 2015, a listed company called
5:12Century Huatong spent 4 million to
5:14acquire a startup. At that time, the
5:16company had only been around for a month
5:17, and suddenly, good news arrived.
5:19Honestly, even Buffett would be
5:20intimidated by Huang Le's track record.
5:22After acquiring the company, its profits
5:24suddenly skyrocketed, helping Century
5:25Huatong’s subsidiary exceed its
5:27projected gambling targets. What is a "
5:29gambling agreement"? It’s like a
5:30military pledge signed during an
5:31acquisition: if you meet the target, I
5:33pay you; if you don't, you compensate me
5:34. But as soon as the gambling period
5:36ended, the profits plummeted by 70%, and
5:38it immediately became a total mess. Yet
5:39, just at that moment, a Hexi
5:41partnership suddenly stepped in with 500
5:43million to take over. Something worth 4
5:44million was resold for 500 million, and
5:46this deal alone pulled Century Huatong
5:48from a profit decline that year to a 55
5:50%surge. Of course, Hexi isn't naive;
5:52they were just waiting for the next act
5:53. What was that next act? By 2019, this
5:55asset was bundled with other gaming
5:57companies to be sold to a Hong
5:59Kong-listed company called 7th Road. The
6:01price doubled to 1 billion RMB, and the
6:02deal required a full cash settlement.
6:04The problem is that Huatong had already
6:06invested in 7th Road; the two sides were
6:07considered strategic partners. Selling
6:09it directly would easily be seen as what
6:10? Selling it directly would easily be
6:12scrutinized as a related-party
6:13transaction, so what could they do?
6:14Huang Le registered a shell company in
6:16the British Virgin Islands. The assets
6:17were first transferred to her name, and
6:19after 23 days, she signed the contract
6:21with 7th Road. In the Hong Kong Stock
6:22Exchange announcement, 7th Road stated
6:24in black and white that Ms. Huang Le was
6:25an independent third party. Yet, the
6:27same announcement also stated that the
6:29Hexi partnership was the ultimate
6:30controller of these assets and an
6:31acting-in-concert party with Huang Le.
6:33Now, that’s interesting. Since these two
6:35statements are in the same document,
6:36isn't that a classic case of
6:37transferring benefits? Starting with 4
6:39million in assets, he moved it through
6:40two transactions into a billion-dollar
6:42cash deal, and in the end, he remained
6:43the ultimate beneficial owner.
6:44Furthermore, it's said that when the
6:46gaming business hit a wall, Huang Le and
6:47his group of capital players shifted
6:49tracks, but their tactics remained
6:51exactly the same. In October '22, the
6:52U.S. began restricting high-end IC
6:54exports, and five months later, 7Road
6:56sold its Singapore-based gaming
6:58subsidiary to Huang's offshore company,
7:00renaming it MASED. They changed their
7:02main website from gaming to real-time
7:03events, positioning themselves as a
7:04cloud service provider. But there was a
7:06hitch: having a Chinese owner and a
7:07Chinese gaming background made it nearly
7:09impossible to pass NVIDIA's client
7:11background checks. So, in 2023, a
7:13Singaporean businessman based in
7:15Shanghai registered a new company. Just
7:17before the U.S. chip sanctions took
7:19effect on November 10th, the entire
7:20stake was transferred under this new
7:22entity. Huang Le then resigned as a
7:24director on the same day, and just four
7:26days later, the company secured official
7:28NVIDIA Cloud Partner status. What is
7:29this qualification for? It serves as
7:31NVIDIA’s stamp of approval on the market
7:33, allowing them to procure high-end
7:35chips. From then on, no matter who asked
7:36, NVIDIA would respond: "This company
7:38has no Chinese shareholders." But what
7:40did the short-selling firm Copper find,
7:41citing comments from current employees?
7:43They claimed that Huang Le was, in fact
7:45, still their boss. Another point of
7:46suspicion is: Where did the money for
7:48the chips come from? After all, 4.8
7:49billion is no small amount—in U.S.
7:51dollars, no less. The following data
7:52comes from public corporate records,
7:54customs logs, and financial statements
7:55unearthed by the short-selling firm. By
7:56the end of 2023, MASED had only $ 572 in
7:58cash on its books. Over the next two
8:00years, its Malaysian subsidiary imported
8:02nearly $ 4.8 billion worth of NVIDIA
8:03equipment, more than half of which were
8:05the latest restricted models. Doing $
8:064.8 billion in business with $ 570 in
8:08capital—that's a leverage of over 800
8:10times. Shouldn't we wonder who fronted
8:11the money? Well, the company’s accounts
8:13are very interesting. By the end of 2024
8:15, a line item for over $ 2.9 billion in
8:16"receivable refundable deposits"
8:18suddenly appeared. Who was it paid to?
8:19Where is the answer hidden? Hidden in
8:21Malaysian business filings, on June 18,
8:232024, this Malaysian subsidiary
8:24transferred all existing and future
8:26equipment—note that future equipment
8:28didn't even exist yet, but even if it
8:29was just thin air, it was all assigned
8:31to a Singaporean company called Apex.
8:33What do the Singaporean records show?
8:35Singaporean records show that Apex is
8:36controlled by Alibaba, and its
8:37registered business is described in a
8:39single phrase: providing procurement
8:40services. Let’s look at who supplied the
8:42goods: over $ 4 billion worth came from
8:44a California server company called IV.
8:46You’ve likely never heard the name, but
8:48you’ve definitely heard of the original
8:50: Inspur Systems. Its parent company is
8:52the famous Inspur Group, which was added
8:54to the U.S. Entity List—the blacklist—
8:56in March 2023, only to change its name
8:58two months later. Why wasn't it blocked
9:00as well? Because what is the rule for
9:01the U.S. blacklist? A subsidiary is only
9:03automatically blocked if the parent
9:05company owns 50%or more. Inspur Group
9:07used a holding company to keep its stake
9:09at exactly one-third, staying just under
9:11the 50%red line. When we piece these
9:13files together, the structure of this
9:14channel becomes clear: Alibaba provided
9:16the funds, the U.S. Inspur company
9:17provided the hardware, the Singaporean
9:19shell company acted as the front, the
9:20chips sat in a Malaysian data center,
9:22the computing power was leased to China
9:24via the network, and Huang Le was the
9:25architect of the entire process. Even
9:27though the chips didn't enter China, the
9:28computing power did. But did Huang Le
9:30violate the regulations? Actually, he
9:32didn’t. At the time, this was a gray
9:33area in U.S. export controls; there was
9:35no explicit ban. But the Americans were
9:37also scratching their heads, wondering:
9:38did the computing power get through? But
9:40didn't the chips get in? They started
9:41conducting on-site inspections, and once
9:43they did, the books didn't add up. At
9:45the end of 2024, U.S. Department of
9:46Commerce officials inspected the
9:48Malaysian data center and what did they
9:49find? They saw large quantities of
9:51equipment still sealed in boxes,
9:52completely unopened. Furthermore, the
9:53number of chips in the data center
9:55didn't match the customs records. What
9:56did the customs records show? They
9:57showed 136,000 Nvidia chips, but after
9:59Nvidia sent people to take inventory,
10:00they only found a few thousand of the
10:01latest-generation chips in the room.
10:03What was Nvidia’s explanation?
10:04Everything not yet installed is kept in
10:05the warehouse, and we’ve counted it all
10:07. There’s no evidence showing it was
10:08shipped to China, but how many units are
10:09actually in the warehouse? He doesn’t
10:11even have a job. That research firm
10:12mentioned something else: back in April
10:14, a data center owned by BN Capital
10:16liquidated MS, and the reason was
10:17laughable—they discovered the racks were
10:19filled with dummy servers, just empty
10:21shells for show, like those fake books
10:23on a shelf. The real hardware had
10:24already been moved out. However, this is
10:26information cited by the research firm,
10:28and I’ll mark it as such as usual. This
10:29brings up an unavoidable question.
10:31Nvidia’s quarterly revenue is around $
10:3396 billion, so you might wonder, is it
10:35really worth it for them to authorize a
10:37suspicious firm for $ 4.8 billion? Was
10:39it worth having the boss himself show up
10:40to endorse it? Is it really worth it?
10:42I’ll tell you, it is absolutely worth it
10:43, because the answer isn’t in Singapore
10:45, but in North America. The machine
10:46powering Nvidia is starting to show
10:48cracks this year. And what is that
10:49machine? It’s the circular capital we
10:50discussed before; I call it "stepping on
10:51your own left foot to lift your right."
10:53Look, Nvidia, venture capitalists, and
10:54private equity firms get together and
10:55pour money into startups and compute
10:57leasing companies. These companies then
10:58take that money and use it entirely to
11:00buy Nvidia chips. In other words, the
11:01invested money circles right back into
11:02their own revenue. We’ve covered this
11:04specifically before, so I won't go into
11:05too much detail today. This year, that
11:07loop is starting to break at both ends.
11:09Let’s look at the equity side first: in
11:10the first half of this year, global AI
11:12startups raised $ 40.7 billion in
11:13venture capital, which set a record. But
11:15don’t be fooled by the scale; the
11:16capital structure is highly concentrated
11:18. How concentrated? OpenAI and Anthropic
11:20alone took over $ 21.7 billion, which is
11:22more than half of it. Even though the
11:24total funding didn't decrease, most of
11:25it just went into the pockets of these
11:26two companies. For the rest of the vast
11:28number of small and medium-sized
11:29companies, financing is getting harder
11:30...Let’s look at the debt side:
11:31CoreWeave, the leader in compute leasing
11:32, had $ 2.6 billion in Q2 revenue but $
11:339.4 billion in capital expenditures.
11:35That massive gap is all covered by debt
11:36. Just look at the interest expenses—
11:38isn't it terrifying? It reached $ 640
11:39million in a single quarter. What does
11:41that even mean? That means for every
11:42four dollars, one dollar has to go
11:44toward interest payments. Then, in the
11:45middle of last month, its bond yield
11:47surged to 15%to 16%. What does this
11:49figure translate to? The market is
11:50already pricing in the possibility that
11:52this company won't be able to pay back
11:53its debt. As everyone knows, another
11:54portion of Nvidia's revenue comes from
11:56cloud giants like Amazon, Google, and
11:57Microsoft. However, the situation on
11:59that front is even more intense. Goldman
12:01Sachs released a report on September
12:0224th stating that the projected total
12:04capital expenditure for these four, plus
12:05Meta and Speedx (xAI), is expected to
12:07reach $ 86 billion, which is almost
12:08double compared to last year. Goldman
12:10Sachs then did the math on the
12:11break-even point. Look, considering the
12:13rapid depreciation of AI chips and
12:14servers, if you amortize the massive
12:16capital expenditure over their lifespans
12:17, these six companies need to earn an
12:19additional $ 308 billion a year just to
12:21cover the depreciation of their hardware
12:22. Mind you, that's just to break even;
12:24not a single cent of profit yet. So,
12:26what is the reality? Based on the
12:27second-quarter books, the annualized
12:29cloud revenue growth for these cloud
12:30providers in 2024 is only around $ 70
12:32billion. Even if we count the
12:33third-quarter reports released at the
12:34end of October, that number just barely
12:36reached around $ 100 billion. It is
12:37still over $ 200 billion short of the $
12:39308 billion break-even line, and this
12:40happens every year. So, how is this gap
12:42filled? Isn't it always by finding ways
12:44to manipulate the books first? Microsoft
12:45is very cunning. As early as July, they
12:47changed their accounting estimates,
12:48extending the useful life of their data
12:50center buildings from 15 to 25 years.
12:51The buildings are the same, but their
12:52useful life has been extended, so the
12:54annual depreciation cost is reduced.
12:55Consequently, the reported capital
12:57expenditure on the books dropped by $ 15
12:59billion. What is the situation now? It
13:00means the money to buy the shovel has
13:02already been spent, but the gold being
13:03dug up isn't even enough to cover the
13:04depreciation of that shovel. Furthermore
13:05, there is now a classic signal of a
13:07cycle peak, and that signal is a surge
13:08in mergers and acquisitions. Let me
13:10briefly explain why, generally speaking
13:11, a surge in mergers and acquisitions
13:13happens when a cycle is peaking. When an
13:15industry's own performance stops growing
13:16, the leaders will use their
13:18high-valuation stocks or cash to go out
13:19and buy growth. What kind of logic is
13:21this actually? It's the logic of
13:22performance consolidation. When is
13:23consolidation needed? The organic growth
13:25rate isn't high anymore. Then, on the
13:272nd of this month, chip company Marvell
13:29announced a $ 1.15 billion all-stock
13:31acquisition of a communications company
13:32. Did you see the market on the day of
13:34the announcement? Both companies '
13:36stocks fell, one by 12.7%and the other
13:38by 18.6%. Both the buyer and the seller
13:40dropped together; this is a clear
13:41reflection of market sentiment. Even
13:42legacy chip giant Analog Devices
13:44recently shelled out $ 7.5 billion in
13:46all cash to acquire which company?
13:48Wireless chip company Spark Microsystems
13:50. The entire semiconductor industry is
13:52now relying on these high-priced
13:53acquisitions to support their valuations
13:54. Now, if we connect these two threads,
13:56it becomes clear why that $ 4.8 billion
13:58is so valuable. Its gross margin was 75
14:00%for the past two quarters, but the Q3
14:02guidance has slipped to 74%. There are
14:04actually many reasons for this, partly
14:06due to rising memory prices, and cloud
14:07giants are also making their own chips
14:09to divert orders. The revenue
14:10contributed by major firms this quarter
14:12is only 13%higher than the last. The
14:13real growth is coming from another group
14:15: those borrowing money to buy GPUs to
14:16act as secondary lessors, or those
14:18taking on government-funded national
14:19projects. This group’s orders jumped by
14:2125%, but they have a fatal flaw: they
14:22have no money themselves. They rely
14:24entirely on borrowing, and sometimes
14:25even need Nvidia to step in and
14:27guarantee them. Do you remember that
14:28move from this past August? Nvidia
14:30rounded up six Wall Street institutions
14:31for a special project. What does that
14:33mean? It means mobilizing $ 500 billion
14:34in external capital to help customers
14:36buy chips. Jensen Huang even publicly
14:37stated that $ 125 billion of that could
14:39be backed by Nvidia. What does that mean
14:40in plain terms? It means if these chips
14:41go unused, sell them to me at a 30%
14:42discount. Yes, I'll buy them; I'll take
14:44them at a 30%discount. This is very
14:45cunning. Now, look back at the Singapore
14:47route: Alibaba pays deposits and
14:49prepayments, and even pays in full for
14:51cash-on-delivery. Nvidia doesn't need to
14:53invest a cent of equity or guarantee a
14:54cent of debt. This is the cleanest cash
14:56in the entire order pool. And what's
14:58even better? It’s the accounting method
15:00—Nvidia records this revenue based on
15:01where the customer's headquarters is
15:03registered. You sell your goods to that
15:04California company, so the revenue is
15:06counted as US income. That’s why in
15:08their financial reports, they can very
15:09confidently write: "Oh, our revenue from
15:11Chinese data centers is assumed to be
15:13near zero." But if you look at the
15:14breakdown from audit agencies, 20%to 40
15:16%of their data center revenue is
15:17actually driven by Chinese demand. Let
15:19everyone think back: what is the one
15:20thing Jensen Huang has been repeating in
15:22Washington for the past two years? He
15:23says the tighter the restrictions, the
15:25faster China will catch up. That is
15:26complete and utter. For a
15:27state-of-the-art AI chip, over 85%of the
15:29design software is controlled by just
15:31three Western companies. Look at EUV
15:32lithography machines; currently, only
15:34ASML in the Netherlands can build them.
15:36The bulk of the optical components are
15:37Japanese, the lenses are German, the
15:39metrology tech is uniquely British, and
15:40finally, it has to be sent to TSMC for
15:42production and packaging. The technical
15:44level required far exceeds the
15:45difficulty of building an atomic bomb.
15:46Look at China's C919 today; I really
15:47don't believe this stuff can just be
15:48forced out by throwing money at it.
15:50Seriously, no country—not even the US—
15:52can do this alone. Thinking they can is
15:53a total pipe dream. So, his statement
15:55isn't an industry assessment; it's a
15:56business strategy. He just wants to
15:58protect the demand from his Chinese
15:59clients. Therefore, my judgment from
16:01start to finish is that this is a mutual
16:02exploitation: Huang wants Jensen's face
16:04, and Jensen wants Huang's money. I dug
16:06up a report from the Taiwanese media two
16:07years ago that is quite interesting. It
16:09says that in June 2024, Huang was at a
16:11computer expo in Taipei. At a client
16:12gathering, he said something in front of
16:14a high-ranking official: "I bet I can
16:15get Jensen to come over." And Jensen
16:17actually showed up, sitting there
16:18drinking and chatting. Then, in that
16:20same month, the documents pledging all
16:21equipment to Alibaba's subsidiary were
16:23signed, and the massive procurement
16:24kicked off. The relationship is very
16:25obvious, because Huang's side just wants
16:27the orders. I'll add one more thing: I
16:28don't know anything about it. In my view
16:29, Jensen Huang has already covered his
16:31bases with three layers. The first is
16:32the shareholder layer: nominally a
16:33Singaporean shell company, so on paper,
16:35the client has no Chinese shareholders.
16:36The second is the delivery or sales
16:37layer to that company, but with Malaysia
16:39as the destination. Once delivered,
16:40revenue is recognized; as for who gets
16:41the profit afterward, that's the
16:42client’s own business. The third layer
16:44is the letter dated October 8th. What
16:45was Congress questioning? It was about
16:47national security and loopholes in
16:48export controls. What is he admitting to
16:49now? He’s saying, "I'm sorry, I went too
16:51far," but claiming no confidential
16:52information was compromised. You see,
16:53Jensen isn't stupid; he had everything
16:54designed from the transaction structure
16:55to the narrative path long ago. So my
16:57conclusion is: this isn't a "honey trap
16:58" story—at least not a simple one, and
17:00certainly not that cliché tale of "
17:01Jensen being scared by a CCP operative.
17:02" After all, it’s not like Nvidia didn't
17:04send people to inspect the site. The
17:05findings weren't about the chips ending
17:06up in China; they were sitting in
17:07warehouses in Malaysia. Furthermore,
17:08what is the current sentiment
17:09surrounding tech stocks? It’s that as
17:11long as your performance doesn't leap
17:12upward, even a 1%drop in revenue will
17:13cause the stock price to take a nosedive
17:15. The reason is simple: current pricing
17:16already accounts for growth expectations
17:18. Once the growth factor—the biggest
17:19contributor to valuation—can no longer
17:20push the stock price higher, what
17:21follows isn't just a correction, but a
17:23total reassessment. I don't believe in
17:24conspiracy theories, and I certainly
17:25don't believe the CCP could spend ten
17:27years cultivating someone just to trick
17:28Jensen. Don't we all know the nature of
17:29the CCP system by now? They use you when
17:31you're useful, and discard you when
17:32you're not. Someone in a role like
17:33Jensen’s is ultimately just a guest, a
17:34transient player. If you tell me he's
17:35colluding with the CCP, I’d absolutely
17:36believe it. But if you say the CCP
17:38cultivated him with a specific long-term
17:39plan, I don't buy it—after all, ten
17:41years ago, Nvidia hardly had any market
17:42status. If the CCP had the foresight to
17:44plan something like that, they wouldn't
17:45be in the pathetic mess they are today.
17:47Alright, that's all for today.
17:50If you enjoyed the show, please click
17:51the bell icon.
17:53If you’d like to support the channel,
17:55feel free to join our membership.

About this transcript

This page contains the full transcript of 黃仁勳認出軌不認48億美元的賬！黃樂不是間諜是白手套！570萬買48億芯片阿里出錢算力租回中國！循環融資裂了雲巨頭年缺2000億才夠本！中國收入記零兩到四成靠中國！美人計？是買賣！ by 一个狠人, generated from the public captions YouTube serves with the video. The transcript has 4,523 words across 693 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.