YouTube2Text

Nvidia's Worst Nightmare Just Came True (and it's not Jensen Huang) — Transcript

by Finance Stack · 3,438 words · 195 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Nvidia's second worst nightmare came true  last week when it lost its CEO, Jensen Huang,
  2. 0:06the man who co-founded it. But its actual worst  nightmare is coming true at the same time. And
  3. 0:11it's got nothing to do with Jensen. Their worst  nightmare is China. See, a year ago, Jensen Huang
  4. 0:18told the Financial Times, "China is going to win  the AI race." Now, Nvidia quickly scrambled to
  5. 0:23retract that statement, but the damage was done.  He basically said the quiet part out loud. Now,
  6. 0:29China's tech giants are spending about 140 billion  on AI this year. That's double last year. So,
  7. 0:36out of every top 100 AI chips that Chinese  data centers are buying right now, how many
  8. 0:41do you think are Nvidia's? Like 50%, 25%. Well, it  used to be 95%, 95 out of every 100. Apparently,
  9. 0:50now it's actually zero. And that isn't because  America's uh of America's restrictions. The truth
  10. 0:56is, China doesn't want Nvidia's chips anymore  because they've got a better alternative. Now,
  11. 1:01by the end of this video, you'll know what  their alternative is, how it affects Nvidia, and
  12. 1:05ultimately how it affects you and me. I'm Nick,  and this is Finance Stack. Now, before we dive
  13. 1:11into the video, our research team puts a lot of  work into every one of these videos. So, there's
  14. 1:16a data sheet linked below in the description  with the source of every number and claim I
  15. 1:20make in this video. Check it out for yourself. So  now back in September of 2025, China's internet
  16. 1:25regulator told ByteDance and Alibaba, two  of the biggest tech companies in China,
  17. 1:29to stop buying Nvidia's special China chip. A  China chip is a watered-down version of the chips
  18. 1:36that Nvidia makes. It makes them specifically for  China. Now, America bans Nvidia's best chips from
  19. 1:41being sold there. So Nvidia slows one down until  it's just under the limit and then it ships it
  20. 1:47to China. It's like a car maker that's banned from  selling its sports cars in a country. So, it sells
  21. 1:51the same car there, but with a crappy motor in  it. Well, America says, "This is the best chip
  22. 1:56that you guys can get from us." And China said,  "No thanks. You keep that crap." Now, of course,
  23. 2:02China hatched a plan to procure some of the best  chips through alternative means. But
  24. 2:08that's not what today's video is about. Today's  video is about China building its own chips. So
  25. 2:142 months after China stopped buying watered-down  chips from Nvidia, they went one step further and
  26. 2:18they banned foreign AI chips from new state-funded  data centers. And if one of those data centers
  27. 2:24was less than 30% built, it had to rip out any  foreign uh chips that had already put in which is
  28. 2:30insane because you know these chips cost  a lot of bloody money. And then after all this,
  29. 2:35America changed its mind. In December of 2025,  Trump said Nvidia could sell China a much better
  30. 2:40chip, the H200 chip, as long as America gets a  25% cut. So about 600,000 of them worth around $16
  31. 2:48billion at a Nvidia's own price are now sitting  in Hong Kong. And Beijing just won't let them in.
  32. 2:54It does not want them. Why would anyone turn down  chips that good? Well, put yourself in Beijing's
  33. 3:00shoes for a moment. Back in 2019, America cut  off Huawei, the Chinese tech giant,
  34. 3:06and it said, uh, you can't sell Huawei phones  anymore in the US, and Huawei can't use Android
  35. 3:12anymore. It was crazy. Then it banned Nvidia's  chips one after the other. If you're China, would
  36. 3:18you build your entire AI future on a supplier  that America can just switch off at any moment?
  37. 3:25Of course you wouldn't. They don't want the US  to have that sort of leverage over them. So,
  38. 3:29at the start of this video, I asked for every  100 top AI chips that China's data centers buy,
  39. 3:34how many are Nvidia's? Well, the answer is zero. And  it's because China has stopped bloody buying them.
  40. 3:41But this begs the question, can China actually  make its own chips? So, let's go back to that
  41. 3:47sports car analogy we were using. If you can't buy  the fast car, the fast sports car, you just build
  42. 3:52your own car, right? If you have the ability  to. So, China's answer to this is that company
  43. 3:56we've been talking about, Huawei, and a smaller  chipmaker called Cambricon. Now, Huawei's boss,
  44. 4:03Eric Xu, admitted that they cannot make chips as  good as Nvidia's. So, what did they do? Well, they
  45. 4:09make up for it by combining chips. In other words,  if your car's slower, you put four weaker engines in
  46. 4:15it, and maybe, just maybe, it beats a supercar.  But that is a pretty big maybe. The data says
  47. 4:23that Huawei's best chips do about half as good as  Nvidia's chips from 2022. Not even Nvidia's newest
  48. 4:30chips, the ones from four years ago. But also,  you got to imagine if you are China, you don't
  49. 4:37want to let the US know that your latest chips,  your cutting-edge ones, are as good as Nvidia.
  50. 4:41So maybe they're kind of getting closer. We don't  know for sure. And also back in 2025, Jensen Huang,
  51. 4:47the Nvidia's boss, laughed off the idea that China  was years behind. He said they are nanoseconds behind
  52. 4:53us. So maybe China has some top-secret chips  that are close to Nvidia's. But we also know that
  53. 4:59Huawei wires hundreds of weaker chips together.  Back in 2025, its big machine did almost twice the
  54. 5:05work of Nvidia's top one. But it burned about four  times more electricity and cost nearly three times
  55. 5:10as much. So yeah, it does get there. It just cost  a bloody fortune to run. But I don't think China's
  56. 5:17worried about the cost. First of all, China  is fairly rich. I mean, so they can afford the
  57. 5:21cost. Also, a big part of the running cost of data  centers is electricity. And China is even richer
  58. 5:27in electricity. It builds power stations like  bloody nobody else. Last year, it added about 10
  59. 5:33times as much power as America did. It's building  41 nuclear reactors right now. America is building
  60. 5:39zero. See, this is where America and democracy  can't keep up. That's that sounds crazy. I I'm
  61. 5:45a fan of democracy, to be clear. I'm happy that I  live in a democracy. I don't want a nuclear power
  62. 5:51plant popping up next door to me. But if we step  back at it and look at it purely as competitors,
  63. 5:56right? In America, building a power station or  a nuclear power plant means years of paperwork,
  64. 6:01planning approval, environmental boards, lawsuits  from angry neighbors. Then when it finally gets
  65. 6:06through, the other side wins the election and  says no more nuclear power plants and it pulls
  66. 6:10back all the plants. Uh, and half a decade  of work is thrown out the window. In China,
  67. 6:14the government just decides it wants to build  a nuclear power plant and kicks you out of your
  68. 6:17house and it kind of builds it. I'm glad I don't  live there, but it is more efficient when you're
  69. 6:22trying to win the AI race, right? And uh they do  the same with solar and wind, of course. So, yeah,
  70. 6:27it it's not great for the people, but this is why  they added 10 times more power than America last
  71. 6:31year. So, even with the crappier chips that you  have to chain together that cost you four times
  72. 6:36as much to run, it's easy when you have 10 times  as much power. But if China's chips are this bad,
  73. 6:42why would Nvidia be scared? Well, it matters  because the AI coming out of those worse chips is
  74. 6:47nearly as good. The best American AI now beats the  Chinese AI by only 3% on the main leaderboard. So,
  75. 6:55think of it as a as a league table for AI, and  the gap at the top is 3%. That's close to nothing.
  76. 7:00Going back to the car analogy, which I guess we're  going to stick with, and we're just going to keep
  77. 7:04using it. China shoved four engines into its car  and it tops out at 223 miles per hour versus the
  78. 7:10American car that does 230 mph. We're not talking  about a big difference here. Uh, and that's what
  79. 7:16Chinese AI labs are good at, getting more out  of less. And we've already had a taste of what
  80. 7:21that does to Nvidia. Back in January of 2025, a  Chinese startup called DeepSeek showed off its
  81. 7:28AI and said that it was built for a fraction of  the cost. On that day, Nvidia's stock crashed and
  82. 7:33$593 billion was wiped off its valuation in a  single day, which is the biggest one-day loss
  83. 7:39for any company in Wall Street's history.  To be clear though, it didn't lose any sales or
  84. 7:44any actual money. It's just the valuation of the  company, and it obviously recovered since then.
  85. 7:48But it still goes to show how easily confidence  can be shaken by the competition that Chinese AI
  86. 7:53models bring to the AI war. This year, DeepSeek's  latest model was partly trained on Huawei chips.
  87. 7:59another Chinese lab, uh, Zhipu, I'm probably saying  that wrong, built, uh, an image model entirely on
  88. 8:05Huawei's chips. Now, what does that even mean? You  know, they built an image model entirely on Huawei
  89. 8:12chips. Well, it all comes down to training. So,  training is how an AI model learns. It's months
  90. 8:16of chips crunching through data. And it's uh the  job that everyone said needed Nvidia, but China
  91. 8:22is proving every day that you might not need  Nvidia anymore. that if you put enough engines
  92. 8:28in the car, you get close enough to matching the  top speed and then the only difference becomes
  93. 8:33the driver's skill. Now, to be fair, this is not a  complete wipeout yet. China's best AI still leans
  94. 8:39partly on Nvidia. So, just last week, a former  OpenAI researcher found signs that DeepSeek's
  95. 8:44and Kimi's newest models, Kimi being another  Chinese AI, were partly trained on Nvidia chips,
  96. 8:50uh chips that China is not allowed to buy anymore.  what uh what they're doing is possibly uh going
  97. 8:57into a gray area by renting GPU compute from  data centers in say Singapore or maybe maybe
  98. 9:03they're even smuggling in some chips. But  you know that's just speculation allegedly,
  99. 9:08right? But what is clear is that China is starting  to move away from Nvidia. They've left 600,000
  100. 9:14Nvidia chips sitting in Hong Kong in favor of  filling their data centers with Huawei chips,
  101. 9:19chips that are theirs. And DeepSeek reportedly  plans to build its next model mainly on Huawei
  102. 9:25chips. China is cutting the cord with Nvidia on  purpose. And from China's perspective, it makes
  103. 9:31complete sense. You can't rely on a supplier whose  government can flick off the switch at any moment.
  104. 9:36Talking about flicking switches, how about you  flick on the like and subscribe button? That is
  105. 9:41so cringe. Uh but it really does help the channel  out. So, so please hit the like and subscribe
  106. 9:45button. But back to what I was actually saying. So  Nvidia has lost China, right? China is no longer
  107. 9:51wanting to buy from them. They're slowly cutting  the cord. They're pulling back. But does Nvidia
  108. 9:55even care? Well, it it does care and it's mostly  because of the money as it as it is always with
  109. 10:01mega companies run by billionaires. So in 2024,  China was about 19% of Nvidia sales. In 2025,
  110. 10:07it dropped to 9%. And the top AI chips it sells  to China's data centers are now less than 1% of
  111. 10:13its data center sales. When America blocked even  its watered-down China chip in April 2025, Nvidia
  112. 10:19took a $4.5 billion hit on chips that couldn't  sell. And look, I know this is getting confusing
  113. 10:24because earlier I said that China rejected the  watered-down chips. Now I'm saying that America
  114. 10:28blocked them. This is because America has been  about as decisive as my wife when it comes to
  115. 10:33ordering takeaway. She says to me, "Uh, I don't  care what we get. You decide." So I decide. I say,
  116. 10:38"I want pizza." She says, "Nah, I don't want  pizza." I say, "Okay, what about Greek?" She says,
  117. 10:42"No, I don't feel like Greek tonight." I say,  "What about some Chinese food?" She says, "Uh,
  118. 10:46nah." And then she's like, "Maybe we should get  pizza." That's how America treats shipping Nvidia
  119. 10:51chips to China. Here's the order of how all of  this unfolded. In April of 2025, America blocks
  120. 10:57the China chip. You know, the watered-down chip.  August, America changes its mind and lets Nvidia
  121. 11:01sell it again as long as America gets a 15%  cut. China says, "Nah, thanks. Keep keep your
  122. 11:06crappy watered-down chip." So, America offers uh a  much better chip, the H200. And China says, "Nah,
  123. 11:13you can keep that one, too." This whole ridiculous  level of indecisiveness from the US has cost
  124. 11:19Nvidia billions of dollars, which I really don't  care about if Nvidia loses money. It's just it's
  125. 11:24just crazy how indecisive a government can be. But  this gives you some insight as to why they care.
  126. 11:30Uh, and Nvidia's chips were still finding their  way into China anyway, because not every Nvidia
  127. 11:35chip stopped going to China. Uh, some have been  getting in through the back door. So, one company,
  128. 11:40Megaspeed, which we've all heard a lot about  recently, moved about $7.5 billion worth of Nvidia
  129. 11:46chips through Singapore and Malaysia, according  to Fortune. And this is where Jensen Huang comes
  130. 11:52back into this story. The reports about why he  quit point to an affair with a former Megaspeed
  131. 11:57executive. US intelligence reportedly suspects  her of spying for China. China denies it, and
  132. 12:03nobody has been charged, so we don't really know.  Since he left, American lawmakers are pushing for
  133. 12:09tougher chip rules. And America is reportedly  about to stop Chinese firms renting Nvidia chips
  134. 12:14in data centers in Singapore. Remember that gray  area we talked about that still allows China to
  135. 12:19uh train its AI? Well, it was doing it by uh  renting GPU compute from data centers abroad. Now,
  136. 12:25they're looking at blocking that, too. Uh and  they're looking at blocking it in Singapore,
  137. 12:28Malaysia, and Thailand. So, now even the back  door is closing. And I know this is a lot of
  138. 12:33information, but basically first the legitimate  routes of selling chips to China were killed and
  139. 12:38that hurt Nvidia. Now the gray area routes are  collapsing and that's going to hurt them even
  140. 12:43more. But at least Nvidia is having no trouble  selling chips in Western countries. So, you know,
  141. 12:48at least there's that. Chip sales have doubled  in the last year. So, they still they're still
  142. 12:52doing incredibly well, right? Right. Well, about  that. See, Western countries are exactly where
  143. 12:59China's AI is heading next. Which brings us to the  actual nightmare. Finally, what is Nvidia's worst
  144. 13:06nightmare? It's cheap Chinese AI. Remember, it's  China's four-engine car that tops out at 223 mph. Now,
  145. 13:14China's running it as a taxi. It gets you there  nearly as fast for about the fair. And there's one
  146. 13:21more thing that China's labs do that America's  don't. Most of their best AI is open source.
  147. 13:26That means anyone can download it for free and  run it themselves. It's like a car m car maker
  148. 13:32publishing the full design of its car. So any  garage in the world can just build one. Whereas
  149. 13:37America's top AI companies keep theirs locked up  and they charge you to use it. DeepSeek charges
  150. 13:41about a fifth of what Anthropic does. Uh Anthropic  is the American company behind Claude in case you
  151. 13:47don't know. Uh and it's uh it charges a fifth of  what Claude charges for its main models. And
  152. 13:52remember the difference in performance is only 3%.  Most people would happily pay 80% less for 97% of
  153. 13:59the performance. Of course, they would. I mean,  who bloody wouldn't? And a great example of this
  154. 14:03is OpenRouter. This is a site where developers  can use whatever AI model they need to complete
  155. 14:08tasks. The amount of work that developers are  sending to Chinese models is increasing. It's
  156. 14:13now more than half of total work being sent to  models. An American company called Lindy moved all
  157. 14:19of its AI work from Anthropic's Claude to DeepSeek.  its boss said you could see the cost crash to the
  158. 14:25ground. Airbnb's boss, Brian Chesky, said that  they rely on uh Alibaba's AI Qwen because it's
  159. 14:32fast and cheap. That work used to go to American  AI companies, and American AI companies are the
  160. 14:38ones paying for Nvidia's chips. So, the people  that were buying Nvidia's sports cars are now
  161. 14:43hopping in a cheap Chinese taxi. And cheap makes  it sound bad. It's actually really bloody good,
  162. 14:48right? It's like a sports car taxi.  Uh, and Nvidia says in its uh, own filing,
  163. 14:54"Being shut out of China helped our competitors  build larger developer and customer ecosystems to
  164. 15:00challenge us worldwide." In plain English, losing  China helped build its rivals. And we've seen
  165. 15:05China do this before. Huawei makes the tech that  phone networks run on, like 5G, etc. It went out
  166. 15:12into the world about 30% cheaper than Ericsson and  Nokia, the big European makers, with Chinese state
  167. 15:17banks lending its customers the money. By 2012,  it was bigger than Ericsson. Ericsson, by the way,
  168. 15:22is Swedish. In 2020, Sweden banned Huawei from its  5G phone networks. The next year, China Mobile,
  169. 15:30China's biggest phone network, which  uh the Chinese state owns, handed out its 5G
  170. 15:36contracts. Ericsson's share fell from about 11% to  about 2%. You can guess why. Now, outside North
  171. 15:43America, Huawei makes up 41% of all the phone  network tech that is sold worldwide, right? well,
  172. 15:49outside of America. Uh, and it's it's cheap. It's  backed by the Chinese state, and it's good enough.
  173. 15:55So, the world bloody bought it. For now, China's  AI chips have barely left China. It can't even
  174. 16:00make enough to satisfy its own demand. And America  has warned companies that using Huawei's AI chips
  175. 16:05could land them in a lot of trouble. But see,  it doesn't need to sell the chips. It just uses
  176. 16:10them to train AI that's almost as good and one  day might even be better. And then it just sells
  177. 16:15the AI instead. So basically, China is  just using the same playbook it used to dominate
  178. 16:20mobile phone network tech to dominate electric  cars and uh next in line is AI. So what does this
  179. 16:26mean for your money? So this year Amazon, Google,  Microsoft, Meta, and SpaceX are spending about
  180. 16:31$828 billion on AI data centers and chips.  And Nvidia keeps about $75 of every $100
  181. 16:38it sells after the cost of making chips. Those  companies buy Nvidia's chips so they can sell AI.
  182. 16:44The whole bit only works if nobody can do it much  cheaper. Because if Chinese AI models uh do nearly
  183. 16:51the same job for about a fifth of the price, who  keeps paying top dollar? And if fewer people pay
  184. 16:56for American AI, those companies need fewer of  Nvidia's chips. And if you've got an index fund or
  185. 17:02a workplace pension, that bet is probably in it,  right? You've probably got exposure. In the S&P
  186. 17:08500, the index of America's 500 biggest companies  uh that a lot of pensions and funds follow,
  187. 17:13about $60 of every thousand is Nvidia, just Nvidia  alone. And the seven big tech companies together
  188. 17:19hold about a third of the total. So remember DeepSeek  Day, you know, when Nvidia lost $593 billion
  189. 17:26off of its valuation in a single day. That was  just a taste. And remember what Jensen Huang said?
  190. 17:31I told you at the start, China is going to win the  AI race. Nvidia walked it back. They retracted the
  191. 17:37statement, but after everything we've just seen,  it's hard to say that he was wrong. And you know,
  192. 17:42let's be honest, he'd know more about China than  most people. Now, look, the the chips that China
  193. 17:46uses are worse and they cost more to run, but it  doesn't really matter because the AI that comes
  194. 17:52out of them is nearly as good and it cost  about a fifth of the price. So, obviously, Nvidia
  195. 17:56is scared. If you found that useful, hit the like  and subscribe button, and I'll see you in the next one.

About this transcript

This page contains the full transcript of Nvidia's Worst Nightmare Just Came True (and it's not Jensen Huang) by Finance Stack, generated from the public captions YouTube serves with the video. The transcript has 3,438 words across 195 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.