广达鸿海斗了二十年,为什么就只服黄仁勋?黄老板被扫地出门,AI 产业链还稳吗?兆元宴合影藏着整个台湾代工江湖|艾森 Essen — Transcript
Full transcript
- 0:00On the evening of January 31st this year
- 0:02, in front of a Taiwanese restaurant
- 0:05called Zhuan Yao Gu Zao Wei in Taipei,
- 0:07about 40 people crowded together for a
- 0:09group photo. The combined net worth of
- 0:12the chairmen and CEOs in this photo
- 0:14exceeds one trillion New Taiwan Dollars
- 0:16.
- 0:17The person who hosted this meal, dubbed
- 0:19by the outside world as the "
- 0:21Trillion-Dollar Banquet," is Jensen
- 0:23Huang.
- 0:24He sat in the center of the front row,
- 0:27with TSMC Chairman C.C. Wei on one side
- 0:29and Quanta Chairman Barry Lam on the
- 0:31other.
- 0:33However, if you look closely at the
- 0:35second row, you will notice a very
- 0:37strange positioning.
- 0:39Foxconn Chairman Young Liu stood
- 0:41directly behind Jensen Huang.
- 0:44Foxconn's revenue last year was 8.1
- 0:47trillion NTD, nearly four times that of
- 0:49Quanta; it is the largest company in
- 0:52Taiwan and Nvidia's top AI server
- 0:54contract manufacturer.
- 0:57The one with the largest market share is
- 0:59standing, while the one with the
- 1:00second-largest is sitting. Seniority is
- 1:03certainly one reason, as Barry Lam is a
- 1:05veteran in the industry. But everyone in
- 1:07Taiwan's tech circles knows that these
- 1:09two companies have been bitter rivals
- 1:11for 20 years.
- 1:13These Taiwanese bosses, who fought tooth
- 1:15and nail under Apple years ago, stood
- 1:18obediently in the same photo today.
- 1:21Among those present was even the only
- 1:23mainland Chinese boss in the room.
- 1:25Four months later, at the same
- 1:27restaurant, against the same wall, the
- 1:29same group of people took another photo
- 1:31, with the seating in the first two rows
- 1:33almost unchanged. Half a year later, on
- 1:36November 12, Nvidia announced that the
- 1:38person who had been sitting in the
- 1:39middle had resigned.
- 1:43This video tells a story about how
- 1:45commercial interests triumph over
- 1:46personal friendships and how a new order
- 1:49is established.
- 1:52We want to answer what it was that
- 1:53filled this table in the first place.
- 1:56Now that the person sitting in the
- 1:58middle has left, can they still sit at
- 2:00that table?
- 2:01To answer this, we must go back to a
- 2:03plate of dumplings. Years ago, Foxconn
- 2:06started by making connectors.
- 2:08Terry Gou's wife, Lin Shu-ju, wanting to
- 2:11help win business from Quanta, found out
- 2:13Barry Lam loved dumplings, so she
- 2:15personally cooked a plate for him. This
- 2:18plate of dumplings earned the friendship
- 2:20of the two families, and Quanta
- 2:22subsequently became Foxconn's largest
- 2:24client for laptop connectors.
- 2:27At the end of 2004, at the retirement
- 2:29dinner for the founder of a red-flag
- 2:32company, according to Taiwanese
- 2:34magazines, Terry Gou hugged Barry Lam
- 2:36from behind, called him "big brother" in
- 2:38front of the media, and promised that as
- 2:41long as Quanta was around, Foxconn would
- 2:43not touch laptop manufacturing.
- 2:46This promise only lasted three years. In
- 2:492007, the global laptop market exploded
- 2:51, and the subsequent financial crisis
- 2:54pushed Foxconn's revenue toward a
- 2:56recession.
- 2:57Terry Gou needed growth, and the biggest
- 3:00prize right in front of him was laptops
- 3:02.
- 3:04In March 2009, a new face appeared at
- 3:06Foxconn’s Neihu office in Taipei: a
- 3:08senior vice president from Quanta whom
- 3:11colleagues called Max.
- 3:15He managed the entire Dell product line
- 3:17, and Dell had even declared that if Max
- 3:19wasn’t the one handling their account,
- 3:21they wouldn't give the business to
- 3:23Quanta.
- 3:24Terry Gou didn’t just poach a vice
- 3:26president; he poached a client
- 3:28relationship, along with a whole team of
- 3:30people who would follow him.
- 3:33Nearly 60 engineers followed him to
- 3:35Foxconn.
- 3:38Barry Lam’s reaction was brutal; the
- 3:41vice president had submitted his
- 3:42resignation through proper channels.
- 3:47After learning he had been poached by
- 3:49Foxconn, Lam changed the status to "
- 3:52dismissal" and ordered that Quanta never
- 3:54hire him again.
- 3:56"Resignation" means leaving on your own
- 3:59; "dismissal" means being kicked out.
- 4:02This was a message for all the remaining
- 4:04Quanta engineers.
- 4:07Three months later, Quanta forcefully
- 4:09snatched back an Apple MacBook order
- 4:11worth billions of NT dollars that
- 4:13Foxconn had already secured.
- 4:16At the investor conference in August,
- 4:18someone asked about the group that had
- 4:20been poached. Barry Lam simply said, "
- 4:23They weren't our best employees anyway,
- 4:25and they haven't achieved anything since
- 4:27moving over."
- 4:30He didn’t say a single word against
- 4:32Terry Gou; he attacked the 60 engineers
- 4:34, essentially saying: "You only poached
- 4:36the people I didn't want."
- 4:40Many people think this feud was resolved
- 4:42by Jensen Huang, but in reality, he had
- 4:45nothing to do with it.
- 4:47On the evening of November 25, 2009, the
- 4:50two attended the same banquet and went
- 4:52to the restroom together.
- 4:55When they came out, they were holding
- 4:57hands, fingers interlaced.
- 5:00Terry Gou held their joined hands up for
- 5:02the reporters to see, saying, "Look, our
- 5:05fingers are locked.
- 5:07He is my big brother, my big brother, my
- 5:10big brother forever."
- 5:12That year, NVIDIA was still just a
- 5:14company making graphics cards for gaming
- 5:16.
- 5:18Yet, even after the handshake, the
- 5:20battle continued.
- 5:22In 2015, Apple’s iPad Pro was originally
- 5:24slated for Quanta, but Terry Gou
- 5:27snatched it back at the last minute with
- 5:29a better price.
- 5:32By 2017, Foxconn had replaced Quanta as
- 5:35the largest manufacturer of MacBooks.
- 5:39People often compare these two scenes.
- 5:41In 2004, Terry Gou called Barry Lam "
- 5:43big brother" in front of hundreds of
- 5:46people, but three years later, for the
- 5:48sake of company growth, he tore up that
- 5:50promise. In 2009, the two interlaced
- 5:53their fingers before reporters, yet six
- 5:55years later, they were still fighting
- 5:58tooth and nail over Apple orders.
- 6:00It’s not that these two men didn't value
- 6:03their friendship; it’s just that they
- 6:05knew all too well that in the Apple
- 6:07supply chain, friendship cannot secure
- 6:09orders, and promises cannot stop
- 6:10business.
- 6:12Why is that?
- 6:14The reason they were forced to stab each
- 6:16other in the back was their shared major
- 6:18client, Apple.
- 6:21Apple's method for managing suppliers is
- 6:23simple: for many key products, they
- 6:26insist on cultivating more than one
- 6:28supplier.
- 6:30iPhone assembly was originally dominated
- 6:32by Foxconn, so Apple propped up Pegatron
- 6:35as a second supplier. In 2014, for the
- 6:38iPhone 6, Foxconn took 70%and Pegatron
- 6:41took 30%.
- 6:43This approach means you can be replaced
- 6:45at any time, so no company dares refuse
- 6:47to lower their prices.
- 6:50This is clearly visible on the balance
- 6:52sheet.
- 6:53In 2017, the gross margins of all four
- 6:56companies—Quanta, Foxconn, Compal, and
- 7:00Pegatron—dropped together.
- 7:03The worst hit was Pegatron, the second
- 7:06supplier specifically propped up to
- 7:09compete with Foxconn; their gross margin
- 7:12was slashed in half from 6.2%in 2015 to
- 7:153%in 2018.
- 7:16In the Taiwanese industry, they call
- 7:18this "gross margin of three to four"—
- 7:20working hard for a year, earning only
- 7:22three or four dollars in profit for
- 7:24every hundred dollars of revenue. Since
- 7:26all four dropped together, it shows the
- 7:28problem wasn't the management of any
- 7:30single company.
- 7:32Rather, the same client tightened its
- 7:34grip on the entire supply chain during
- 7:36those two years.
- 7:39That’s how Quanta’s Apple Watch business
- 7:41went. When the first generation launched
- 7:44in 2015, Quanta was the exclusive
- 7:46assembler; they didn't make a profit for
- 7:49two years, only seeing returns by 2017.
- 7:51In 2019, Quanta decided not to take on
- 7:54new models. Barry Lam reasoned that they
- 7:57would only do what was profitable, as
- 7:59profit is paramount. The new orders
- 8:01eventually went to Foxconn and Compal.
- 8:05In the Apple supply chain, what everyone
- 8:07fights over is often a money-losing
- 8:09contract.
- 8:12That is the rule of the game: you can be
- 8:14replaced at any time, so no one’s
- 8:16promises are worth anything. Back then,
- 8:18no one could have imagined that ten
- 8:20years later, someone would appear who
- 8:22could make that same group of incredibly
- 8:24shrewd Taiwanese businessmen willing to
- 8:26bet billions of New Taiwan dollars on a
- 8:28long-term promise.
- 8:30Investing tens of billions of NT dollars
- 8:32.
- 8:34The other half of this story goes back
- 8:36to Quanta’s annual year-end party in
- 8:382010. That year, Barry Lam, who had just
- 8:41lost 60 people to Foxconn, took the
- 8:43stage dressed as the Monkey King in
- 8:45front of thousands of employees. He said
- 8:47, "I won't go to the 'Blue Ocean,'
- 8:49because one day it will turn into a '
- 8:51Red Ocean'—I’m going to the cloud."
- 8:53Everyone in the audience understood. "
- 8:55Red Ocean" referred to Foxconn. Barry
- 8:57Lam meant: "If you want to fight for
- 9:00this laptop business, go ahead; I’m done
- 9:02playing with you. I’m going to make
- 9:05servers." At that time, Quanta’s primary
- 9:07business was still laptops. He didn't
- 9:10even know if the cloud could support the
- 9:12company, or how many years it would take
- 9:14.
- 9:17Across the Pacific, someone else was
- 9:19doing something even harder to fathom:
- 9:21he was pushing a company that made money
- 9:23on gaming graphics cards into a
- 9:25computing market that almost no one was
- 9:27willing to pay for at the time.
- 9:32On October 28, 2009, in Beijing, NVIDIA
- 9:34presented a plaque for a "CUDA Center of
- 9:37Excellence" to Tsinghua University.
- 9:41The person who personally handed the
- 9:43plaque to the director of Tsinghua’s
- 9:45Computer Department was Jensen Huang
- 9:47himself.
- 9:48What exactly was it? To put it simply,
- 9:50it was a set of tools that allowed
- 9:52graphics cards to be used not just for
- 9:54gaming, but also for scientific
- 9:55computing. Back then, people doing
- 9:58scientific computing only used CPUs. If
- 10:00you applied to buy a graphics card in
- 10:02the lab, your advisor would suspect you
- 10:04just wanted to play games on the sly. A
- 10:06year later, the National University of
- 10:08Defense Technology’s Tianhe-1A was
- 10:11equipped with 7,168
- 10:13NVIDIA graphics cards, reaching the
- 10:15number one spot on the TOP500 list of
- 10:18the world's supercomputers.
- 10:23Putting CUDA on every GPU, that was a
- 10:25strategic decision that was very, very
- 10:27hard to do because the company consumed
- 10:29enormous amounts of our profits and we
- 10:32couldn't afford it at the time.
- 10:35For 10 years, that made NVIDIA what it
- 10:38is today.
- 10:40>> However, to Wall Street, these were all
- 10:43losing propositions.
- 10:47Huang later recalled that putting CUDA
- 10:49into every gaming graphics card was an
- 10:51incredibly difficult decision that
- 10:53drained huge profits from the company—
- 10:55something NVIDIA couldn't afford at the
- 10:57time.
- 11:01He said at Stanford that NVIDIA spent
- 11:03about ten years investing in a future
- 11:05that didn't even exist yet.
- 11:09For 10 to 15 years, the markets that
- 11:12sustain NVIDIA today didn't exist at all
- 11:14.
- 11:17When asked how he reacted when the
- 11:19market cap dropped by 80%, he said his
- 11:22pulse was exactly the same as usual.
- 11:26The stock price changed, but did the
- 11:28laws of physics change?
- 11:30Did gravity change? Nothing changed, so
- 11:33we just kept moving forward.
- 11:35One was waiting for the cloud, and the
- 11:37other was building the computing tools
- 11:39needed for the cloud.
- 11:42At the time, neither could guarantee
- 11:44they would win; their paths converged on
- 11:47the AI server.
- 11:51According to Quanta Vice Chairman C.C.
- 11:53Leung, the two companies’ AI
- 11:55collaboration likely began around 2015
- 11:58or 2016.
- 12:00It was still a small business back then
- 12:02.
- 12:03Seven or eight years later, AI really
- 12:05took off.
- 12:07"Nice to see you, ladies and gentlemen.
- 12:09Barry, my hero."
- 12:15>> On the evening of June 5, 2024, it was
- 12:18raining heavily in Neihu, Taipei. Barry
- 12:21Lam hosted a dinner for Jensen Huang at
- 12:23a restaurant called Shih Hsiang Fong. He
- 12:26stood in the rain waiting for Jensen
- 12:28Huang to get out of the car, first
- 12:30apologizing in English, then adding in
- 12:33Taiwanese, "phái-sè, phái-sè," and
- 12:35following up in Mandarin, "sorry." Barry
- 12:38Lam replied in Taiwanese, "bô-iàu-kín"(
- 12:40it doesn't matter). Then Jensen Huang
- 12:42turned to the camera and said, "This is
- 12:45my hero.
- 12:46Thank you for waiting for me in the rain
- 12:48."
- 12:49Barry Lam gave a thumbs up and replied
- 12:51in English, "Now you are our King."
- 12:55Fourteen years ago, the man who was like
- 12:57the Monkey King dodging the Red Sea to
- 12:59reach the cloud is now standing in the
- 13:02rain ten years later, calling another
- 13:04gambler "King."
- 13:06However, Barry Lam doesn't follow Jensen
- 13:08Huang blindly on everything.
- 13:10That very week, 12 days after Jensen
- 13:13Huang declared at the NTU Sports Center
- 13:15that the era of humanoid robots was
- 13:17coming, Barry Lam said at the Quanta
- 13:19shareholder meeting that humanoid robots
- 13:21are expensive and have no customers, and
- 13:24their only advantage might be that they
- 13:26don't complain or go on strike.
- 13:30The person sitting at Jensen Huang's
- 13:32side is not his follower, but his
- 13:34partner.
- 13:36From May 2024 to May of this year,
- 13:39Jensen Huang hosted five dinners at that
- 13:41same Taiwanese restaurant; as Pegatron's
- 13:44Jason Chang entered, he remarked that he
- 13:46was here to attend the year-end party
- 13:48for "On" and everyone else.
- 13:52In January 2025, to squeeze all 35 CEOs
- 13:55into the frame, there was quite a
- 13:57scramble over who would occupy the front
- 14:00row.
- 14:01In the end, Jensen Huang himself
- 14:03crouched down in the front row.
- 14:06He joked that those crouching in the
- 14:08front were the ones who exercised, while
- 14:11those in the back did not, but don't
- 14:12forget, these people are still fighting
- 14:15over orders today, just as fiercely as
- 14:17they fought for Apple's business back
- 14:19then.
- 14:21Although they are still competitors, in
- 14:23the Apple era, bosses treated each other
- 14:25like enemies, but in the NVIDIA era,
- 14:27they can sit at the same table and toast
- 14:30one another.
- 14:31Jensen Huang being more charismatic than
- 14:33Tim Cook is only a small part of the
- 14:35reason; the bigger answer is written in
- 14:37the financial statements of these
- 14:39companies.
- 14:42Who could have imagined that Barry Lam,
- 14:44who had 60 employees poached and hid
- 14:46away in the cloud like Sun Wukong, would
- 14:49trade places with Terry Gou on Taiwan's
- 14:51rich list over a decade later?
- 14:54On the 2018 Forbes Taiwan Rich List,
- 14:57Terry Gou ranked first, with a net worth
- 15:00twice that of Barry Lam.
- 15:03By 2025, Barry Lam had overtaken Terry
- 15:06Gou, but the more interesting part of
- 15:08this story is yet to come.
- 15:12The path Barry Lam chose back then to
- 15:14avoid Terry Gou eventually became the
- 15:16new battlefield where both of them are
- 15:19now fighting tooth and nail.
- 15:21Foxconn got its start by assembling
- 15:23iPhones for Apple. By the second quarter
- 15:26of 2025, Foxconn's cloud business,
- 15:28driven primarily by AI servers,
- 15:30surpassed its consumer electronics
- 15:32business—including the iPhone—in revenue
- 15:35for the first time.
- 15:39The world's largest Apple contract
- 15:41manufacturer has seen its biggest
- 15:43business change hands; Foxconn remains a
- 15:45giant in this new battlefield, still
- 15:47holding the number one market share.
- 15:54On this year's list, Terry Gou is ranked
- 15:56ahead of Barry Lam again, with only a
- 15:58slight difference between them; Barry
- 16:00Lam didn't defeat Terry Gou, but the
- 16:02battlefield and the price the capital
- 16:05market puts on this war have changed.
- 16:11Many might assume the high valuations
- 16:13are because Jensen Huang pays
- 16:15manufacturers better than Apple,
- 16:16allowing them to earn higher gross
- 16:18margins.
- 16:20But the numbers don't say that; Quanta's
- 16:23gross margin in 2018 was just over 4%,
- 16:26which is a standard report card for
- 16:28laptop contract manufacturing.
- 16:31After AI servers took off, it once
- 16:33climbed to nearly 8%in 2024.
- 16:36However, Quanta's revenue in the first
- 16:38few months of this year is already more
- 16:41than double that of the entire year of
- 16:432024.
- 16:44Yet the margin has fallen back to under
- 16:475%.
- 16:48Similar to when they made laptops for
- 16:51Apple back then, some might say that if
- 16:53the gross margin has dropped below 5%,
- 16:55isn't that just being exploited by
- 16:57NVIDIA?
- 17:01Quanta's CFO explained at the May
- 17:03earnings call this year that the decline
- 17:05in gross margin is mainly because the
- 17:07product has become a $ 3 million AI
- 17:09server rack, where chips and memory
- 17:12account for over 80%of the cost.
- 17:17The contract manufacturer only earns
- 17:19money from assembly and testing; revenue
- 17:21is inflated by expensive components, so
- 17:23the gross margin percentage is naturally
- 17:26diluted.
- 17:28Profit margins are falling, but profits
- 17:30are rising.
- 17:32Quanta's operating profit has grown
- 17:35sixfold from 2018 to 2025, but what has
- 17:38truly turned things upside down is the
- 17:40market cap.
- 17:43From the end of 2018 to November 12 of
- 17:46this year, Quanta's market cap has grown
- 17:495.6 times.
- 17:50Wistron and Foxconn have also multiplied
- 17:53several times over.
- 17:54Of course, this money didn't all come
- 17:56from Jensen Huang alone.
- 17:59No one can stop this wave of AI.
- 18:03But Jensen Huang knows exactly what he
- 18:05has given these people.
- 18:09After his fifth dinner party in May, he
- 18:11told reporters that the stock prices of
- 18:13all companies partnering with NVIDIA
- 18:15have tripled within a year.
- 18:20I am very happy for them; they deserve
- 18:22it.
- 18:25He didn't say how much profit he shared
- 18:27with them; he said your stock prices
- 18:29have gone up.
- 18:31In the Apple era, what contract
- 18:33manufacturers cared about most was how
- 18:35much money an order could make.
- 18:38In the NVIDIA era, they have to
- 18:39calculate something else: will there
- 18:41still be a place for them in the next
- 18:43generation?
- 18:45As long as the capital market believes
- 18:48you will continue to produce generation
- 18:50after generation, even if you only make
- 18:52a few dollars on every 100, the stock
- 18:54will be revalued based on growth over
- 18:57the next decade.
- 19:00I think the most valuable thing Jensen
- 19:03Huang has given this industry is not
- 19:05orders or profits, but a future that can
- 19:07be valued in advance by the capital
- 19:09market.
- 19:11But a future is only valuable if someone
- 19:13believes in it.
- 19:15Why should Taiwanese bosses believe him
- 19:17?
- 19:18The answer lies in the balance sheets of
- 19:20both companies.
- 19:23According to Apple's latest annual
- 19:25report, after suppliers deliver goods to
- 19:27Apple, they wait about four months on
- 19:29average to get paid.
- 19:32Yet Apple's own inventory turns over
- 19:34extremely fast.
- 19:37Apple has taken cash turnover to the
- 19:39extreme with a cash conversion cycle of
- 19:41negative 67 days, meaning Apple always
- 19:43has over two months of cash on hand that
- 19:46suppliers are essentially carrying for
- 19:48them.
- 19:50NVIDIA is the exact opposite. According
- 19:53to their annual report, they hold a
- 19:55large amount of inventory themselves and
- 19:57pay suppliers quickly, resulting in a
- 20:00cash conversion cycle of positive 133
- 20:02days.
- 20:04Even more staggering are Nvidia's
- 20:06procurement and capacity commitments,
- 20:09which hit $ 279 billion by the end of
- 20:11July—nearly tripling in just six months
- 20:14.
- 20:16The bulk of this increase comes from
- 20:18prepurchasing memory.
- 20:22Most of this money hasn't been paid yet
- 20:24, and some can still be adjusted before
- 20:26formal orders, but the intent is clear:
- 20:29Nvidia is locking in future capacity for
- 20:31the entire supply chain.
- 20:37One makes suppliers bear the cost, while
- 20:39the other steps up to secure capacity;
- 20:41both are dominant, but they are at
- 20:43different stages.
- 20:48Apple faces a mature market with a fixed
- 20:50pie, so it pushes to minimize every cost
- 20:53.
- 20:56AI is a rapidly growing market where
- 20:58chips and memory are both scarce; Nvidia
- 21:00needs to snatch up tomorrow's supply
- 21:03today.
- 21:07Apple forces suppliers to fight for a
- 21:09visible market, while Nvidia forces them
- 21:11to compete for a future not yet realized
- 21:13.
- 21:16Apple relies on the fear that you are
- 21:18replaceable.
- 21:20Jensen Huang relies on the expectation
- 21:22that you are needed for the next
- 21:24generation.
- 21:27An order can get a factory running, but
- 21:30to make a boss sink billions into a line
- 21:32only for the next generation, orders
- 21:34aren't enough; they must believe that
- 21:36line will still be busy in two years.
- 21:40Fear can force price cuts, but capacity
- 21:43expansion relies on expectations.
- 21:46This brings up something that doesn't
- 21:48quite add up.
- 21:49At the 2024 Computex, Jensen Huang
- 21:52called Barry Lam a hero, and Barry Lam
- 21:55called Jensen Huang a king.
- 21:59Yet, less than two weeks after Computex
- 22:02, Taiwanese media reported that for the
- 22:05first batch of Nvidia's next-gen AI
- 22:08GB200, Foxconn took 40%and Quanta took
- 22:1130%.
- 22:16Back in March of that year, analyst
- 22:18Ming-Chi Kuo warned the market that
- 22:20Nvidia was putting more effort into
- 22:22building its own systems to reduce
- 22:24reliance on Foxconn.
- 22:28If Jensen Huang and Barry Lam are so
- 22:30close, why isn't Quanta the biggest
- 22:33winner? And if Foxconn's reliance was
- 22:35being reduced, why did they still take
- 22:37the top spot? Because Jensen Huang knows
- 22:39better than anyone how Apple managed
- 22:41these people in the past.
- 22:44He can make everyone believe the future
- 22:46pie is huge, but he will never let any
- 22:48single one of them hog the whole thing.
- 22:52Friendship is for the dinner table, but
- 22:54orders are business; exactly how orders
- 22:56are allocated isn't decided over dinner
- 22:58.
- 22:59Once a batch of the tightest parts
- 23:01arrives, which factory should they be
- 23:03sent to?
- 23:06Nvidia wrote in their own tech blog that
- 23:08they built a system with an American
- 23:10data analytics firm, Perplexity, to
- 23:12recalculate every week, with the
- 23:14resulting plan handed to supply chain
- 23:16planners to finalize.
- 23:20Clients also choose which foundry gets
- 23:22their orders.
- 23:25Industry reports state that Wistron has
- 23:27been the exclusive foundry for Dell from
- 23:30the B200 to the GB200.
- 23:33Quanta lost a portion of Meta's orders
- 23:35over the past two years; it was Meta
- 23:37that changed its supplier.
- 23:39So, what does Jensen Huang hold in his
- 23:42hands? At the end of 2024, Taiwanese
- 23:44media reported that NVIDIA had finalized
- 23:47the preliminary order allocation for the
- 23:49GB200, with Foxconn as the largest
- 23:51supplier and Quanta as the second.
- 23:55Note the term "finalized," because
- 23:57NVIDIA’s headquarters sets the framework
- 24:00: whose technology to use, who to
- 24:02certify, who receives platform support,
- 24:04who is eligible to sit at the table this
- 24:07generation, and the approximate capacity
- 24:09each company will handle.
- 24:12Simply put, Jensen Huang issues licenses
- 24:15, not orders.
- 24:18Getting a license doesn't guarantee your
- 24:20share, but without it, you don't even
- 24:21have the qualification to sit at the
- 24:23table.
- 24:25How do you get this license?
- 24:27It depends on whether you can do what
- 24:29others cannot; personal relationships
- 24:31come second. High-bandwidth memory from
- 24:34South Korea is exactly like this. This
- 24:36is a type of high-speed memory
- 24:38specifically used alongside AI chips. SK
- 24:40Hynix, Micron, and Samsung have all
- 24:43received NVIDIA's certification. Samsung
- 24:46once failed certification, causing its
- 24:48share to drop from 40%to less than 20%,
- 24:51but it was pulled back up after passing
- 24:54.
- 24:54The only entrepreneur from mainland
- 24:57China present that day was Chen Tao,
- 24:59chairman of Huizhou Shenghong Technology
- 25:01.
- 25:02His factory is one of the few in the
- 25:04world capable of mass-producing the
- 25:06high-end multi-layer circuit boards
- 25:08NVIDIA requires.
- 25:11Taiwanese experts say without this
- 25:12capability, he wouldn't be able to sit
- 25:14at this table.
- 25:16The company with the highest gross
- 25:18margin in this supply chain is actually
- 25:20Chuan Hsin, a company from Kaohsiung
- 25:22that started out making cabinets.
- 25:24NVIDIA's new platform's server rails,
- 25:26those with seven layers or more, are
- 25:29made by Chuan Hsin, and their gross
- 25:31margin is even higher than NVIDIA's. So
- 25:34, do you understand this group photo now
- 25:36?
- 25:37Whether one is sitting or standing isn't
- 25:40that important; what matters is who is
- 25:42qualified to be in the photo at all.
- 25:46Jensen Huang didn't abolish Apple's rule
- 25:48of having suppliers compete against each
- 25:51other; he simply moved that competition
- 25:53into a high-growth market.
- 25:59Facing a cake that keeps getting bigger
- 26:01, everyone still fights for it, but even
- 26:03those who lose the fight get to eat, so
- 26:05no one flips the table.
- 26:10Jensen Huang didn't make the Taiwanese
- 26:12bosses stop competing; he just made them
- 26:15believe that even while competing,
- 26:16everyone can get rich together.
- 26:21And this belief relies on one person's
- 26:23credibility.
- 26:24On November 11, NVIDIA's board accepted
- 26:27Jensen Huang's resignation. The next day
- 26:30, it was announced that he would step
- 26:33down as President, CEO, and Director,
- 26:35with CFO Colette Kress serving as acting
- 26:38CEO. For that entire week, almost none
- 26:41of these Taiwanese bosses said a word.
- 26:43On November 10th, Barry Lam attended an
- 26:45awards ceremony.
- 26:47Reporters hounded him about whether his
- 26:49supply chain would be affected, but he
- 26:51wouldn't comment.
- 26:52When asked if his mood was affected, he
- 26:55said "not bad" twice.
- 26:57Executives like Terry Gou and Jason
- 26:59Chang remained silent that week.
- 27:04The only person to stand up for Jensen
- 27:06Huang was Quanta’s number two, C.C.
- 27:08Leung.
- 27:10He stated that Jensen Huang would
- 27:12certainly not betray anyone, and added
- 27:15that everyone still deeply needs AI.
- 27:18Current orders remain untouched.
- 27:21On November 13th, Taiwan's Central News
- 27:23Agency interviewed four semiconductor
- 27:26supply chain executives; they said
- 27:28orders through 2027 are unchanged, and
- 27:30Nvidia's technical roadmap has not been
- 27:33adjusted.
- 27:35One testing executive noted that
- 27:37Nvidia’s platform barriers are getting
- 27:39higher, so it isn't easy to just switch
- 27:41.
- 27:44Nvidia’s new chairman also stated in an
- 27:46announcement that company operations,
- 27:48customer commitments, and product
- 27:50roadmaps would not be interrupted.
- 27:54However, behind these public statements
- 27:56, some have already started calculating
- 27:58a different bottom line.
- 28:01The Koreans have been the most blunt
- 28:02about it. On November 13th, both Samsung
- 28:05and SK Hynix refused to comment, but a
- 28:08vice chairman of the Korea Semiconductor
- 28:10Industry Association spoke up.
- 28:13He said this might not necessarily be a
- 28:15bad thing for Korea.
- 28:17Even if other companies fill the market
- 28:20share left by Nvidia, those companies
- 28:22will still require Korea's
- 28:23high-bandwidth memory.
- 28:27He added that they don't need to
- 28:29concentrate production on a single
- 28:31client; they can diversify among
- 28:33multiple clients, which is actually more
- 28:35favorable for them.
- 28:38If Nvidia wants to change suppliers, he
- 28:41countered: where are the alternative
- 28:43suppliers?
- 28:45The Koreans have the confidence to say
- 28:47this because only a few companies
- 28:49worldwide can supply advanced
- 28:51high-bandwidth memory at scale.
- 28:54Today, it isn't just that the Koreans
- 28:56can't live without Nvidia; Nvidia cannot
- 28:59live without Korea.
- 29:01A U.S. semiconductor research firm
- 29:03published an article on the day of
- 29:04Jensen Huang's resignation.
- 29:09It estimated that in 2027, the price per
- 29:11unit of SK Hynix's next-gen
- 29:13high-bandwidth memory for Nvidia could
- 29:16be nearly 20%cheaper than for clients
- 29:19like AMD.
- 29:22How does Nvidia secure such terms?
- 29:25Because Nvidia is one of their largest
- 29:28clients, the two parties have
- 29:29collaborated for many years, possibly
- 29:32even exchanging benefits like priority
- 29:34GPU supply.
- 29:36How many of these terms are in contracts
- 29:39versus based on mutual trust for future
- 29:41cooperation is unclear to outsiders.
- 29:45What is certain is that after the new
- 29:47management takes over, holding only past
- 29:49contracts won't be enough when sitting
- 29:51down to negotiate prices and capacity.
- 29:56They still have to convince the Koreans
- 29:58that Nvidia remains the client most
- 30:00worthy of priority.
- 30:03When Jensen Huang was there, everyone
- 30:05knew how these interest exchanges were
- 30:07coordinated.
- 30:10If Google and Amazon come knocking with
- 30:12major orders in the future, why should
- 30:14Nvidia remain at the front of the line?
- 30:18This isn't a problem that can be solved
- 30:20with a meal or a photo op.
- 30:23More troublingly, even if Jensen Huang
- 30:26hadn't resigned today, Nvidia's
- 30:27competitors have long been carving into
- 30:30their market share.
- 30:33Google has its own AI chips, Amazon does
- 30:36too, and AMD has been catching up
- 30:38steadily.
- 30:42Morgan Stanley predicts that by 2027,
- 30:45Nvidia's share of global high-bandwidth
- 30:48memory demand will drop to around 37%,
- 30:50while Google's self-developed chips will
- 30:53account for 36%, putting them neck and
- 30:56neck.
- 31:00In that CNA report, a high-level
- 31:02executive at a substrate manufacturer
- 31:04also mentioned that the landscape
- 31:06between GPUs and application-specific
- 31:09chips might shift toward the latter.
- 31:13So the competition itself began long ago
- 31:15, regardless of whether Jensen Huang
- 31:17leaves or not.
- 31:20The real issue is that at the peak of
- 31:22this intense competition, Nvidia
- 31:24suddenly lacks the person who was best
- 31:26at binding technology suppliers and
- 31:28capital together.
- 31:30At a time like this, will others seize
- 31:32the opportunity to snatch away the next
- 31:34orders that originally belonged to
- 31:36Nvidia?
- 31:38A signed order won't become void just
- 31:40because one person resigns.
- 31:43But what about the next unsigned order,
- 31:45the next factory that hasn't broken
- 31:47ground, or the next year’s capital
- 31:49expenditure yet to be approved?
- 31:52A former Nvidia Korea representative
- 31:54stated clearly that the contracts signed
- 31:57by Samsung and SK Hynix with Nvidia
- 31:59already cover next year, so there won't
- 32:01be much impact in the short term.
- 32:05However, variables may emerge after 2028
- 32:08.
- 32:10Looking at it from another angle, if
- 32:12this supply chain really cannot function
- 32:14without Jensen Huang, it would suggest
- 32:16that in his 30-plus years of running
- 32:18Nvidia, he failed to turn his most
- 32:20important capabilities into company
- 32:22systems.
- 32:23But Nvidia's technical standards,
- 32:25customer network, and supply chain
- 32:27systems are all still in place. TSMC’s
- 32:30production capacity won't simply vanish
- 32:32into thin air just because he resigned.
- 32:35Can Nvidia keep spinning without Jensen
- 32:37Huang?
- 32:38The answer is already clear.
- 32:41It certainly can.
- 32:44The real test of this transition is
- 32:46whether Nvidia can continue to convince
- 32:49the entire industry chain that, even
- 32:51without Jensen Huang, betting the next
- 32:53round of investment here remains the
- 32:55best business move.
- 32:59TSMC has already provided a hint through
- 33:01its actions.
- 33:05On November 10, the day before Nvidia's
- 33:07board accepted his resignation, TSMC's
- 33:10board approved three investment quotas
- 33:13of up to $ 500 million each: subscribing
- 33:15to shares of AI company Anthropic upon
- 33:18its IPO, buying OpenAI preferred stock,
- 33:20and establishing a strategic investment
- 33:23firm.
- 33:28This certainly doesn't prove that C.C.
- 33:30Wei knew anything in advance, but it at
- 33:32least shows that TSMC has never bet on
- 33:35any one individual at NVIDIA, but rather
- 33:37on the future demand for AI as a whole.
- 33:42Some of these companies are now making
- 33:44their own chips.
- 33:46As mentioned in my video a couple of
- 33:48days ago, production lines have inertia
- 33:51, but bridges don't.
- 33:53To be more specific now: orders have
- 33:55inertia, but trust does not.
- 33:59When the next round of discussions
- 34:01regarding pricing, capacity, and
- 34:02next-gen platforms begins, the person
- 34:05sitting across the table will no longer
- 34:07be the one known in Taipei’s industry
- 34:09circles as "The King."
- 34:12The November 17th earnings call will be
- 34:14Kress's first time addressing the market
- 34:17as acting CEO.
- 34:19As of November 13th, those bosses in
- 34:21Taiwan and South Korea are still waiting
- 34:23for him to speak.
- 34:25This year, on May 28th, was the fifth "
- 34:28AI Banquet."
- 34:30That day, Jensen Huang's son Spencer and
- 34:32daughter Madison were also there.
- 34:35C.C. Wei was still on one side of him,
- 34:38Barry Lam on the other, and Young Liu
- 34:40was still standing in the second row.
- 34:44That was the last banquet he hosted in
- 34:47his capacity as NVIDIA's CEO.
- 34:51In 2009, Terry Gou and Barry Lam locked
- 34:53fingers in a restroom, but once they
- 34:55came out, they continued their rivalry
- 34:58just the same.
- 34:59A few years later, this group finally
- 35:01managed to sit at the same table.
- 35:05In truth, they never stopped competing
- 35:07for a single day, but someone appeared
- 35:09in their midst who made everyone believe
- 35:11the future pie would keep growing—large
- 35:14enough for everyone to get a slice.
- 35:17Now that person is gone. By this time
- 35:19next year, there might still be a group
- 35:21photo taken in front of that Taiwanese
- 35:23restaurant.
- 35:25Who sits in the first row or stands in
- 35:27the second row really doesn't matter as
- 35:29much anymore.
- 35:31What truly matters is whether these
- 35:33people are still willing to bet billions
- 35:35upon billions of capital on the next
- 35:37generation of chips once they go back.
- 35:46A photo can keep everyone in the same
- 35:48frame, but only a future that everyone
- 35:50believes in can keep them sitting at the
- 35:52same table.
- 35:59Thank you for watching today's video, we
- 36:01...
About this transcript
This page contains the full transcript of 广达鸿海斗了二十年,为什么就只服黄仁勋?黄老板被扫地出门,AI 产业链还稳吗?兆元宴合影藏着整个台湾代工江湖|艾森 Essen by 艾森 Essen, generated from the public captions YouTube serves with the video. The transcript has 5,183 words across 900 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.