YouTube2Text

NVIDIA Explained Like You're 5 — Transcript

by Crayon Capital · 2,199 words · 421 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This dishwasher boy built a $5 empire
  2. 0:03from the table of a Denny's diner.
  3. 0:05He pitched the business plan over cheap
  4. 0:06coffee in 1992, and 34 years later, his
  5. 0:10company makes $20
  6. 0:11every single hour.
  7. 0:13Not from the safe bet, but from a gamble
  8. 0:15nobody believed in that crushed entire
  9. 0:17industries and took over the AI space,
  10. 0:20the gaming industry, and even all of
  11. 0:22your app algorithms.
  12. 0:23And just as he started winning, the
  13. 0:25government moved to [music] bring him
  14. 0:27down.
  15. 0:29Jensen Huang was born in Taiwan in 1963.
  16. 0:33When he was nine, [music] his parents
  17. 0:35sent him and his brother to America for
  18. 0:36a better education.
  19. 0:38The plan was simple. Live with relatives
  20. 0:40in Kentucky, study hard, build a future.
  21. 0:45This doesn't look like a school. It
  22. 0:47looks more like somewhere people get
  23. 0:48sent after they've done something very,
  24. 0:50very wrong. That's because
  25. 0:52>> [music]
  26. 0:52>> it kind of is. Welcome to Oneida Baptist
  27. 0:54Institute. You'll learn English,
  28. 0:56discipline, and how not to complain
  29. 0:58while cleaning bathrooms.
  30. 1:00This is not how most tech CEO origin
  31. 1:03stories start, but it didn't break him.
  32. 1:05It taught him something no lecture hall
  33. 1:07could.
  34. 1:08When the situation is terrible, you
  35. 1:10adapt [music]
  36. 1:10or you disappear.
  37. 1:12He eventually made it to Oregon State,
  38. 1:14then Stanford, spent years at chip
  39. 1:16companies,
  40. 1:17AMD, then LSI Logic, studying
  41. 1:21semiconductors with a level of obsession
  42. 1:22that was, frankly, a little alarming.
  43. 1:25By 1992, Jensen had an idea.
  44. 1:28Graphics are going to change computing.
  45. 1:31Games are getting bigger. 3D is getting
  46. 1:33harder. CPUs can't carry this forever.
  47. 1:37Someone is going to build a chip for
  48. 1:38this new world.
  49. 1:40I think it should be us.
  50. 1:41Jensen, I have a mortgage, a stable job,
  51. 1:44and a family that currently believes I
  52. 1:46make responsible decisions.
  53. 1:48Good.
  54. 1:49Then this will be exciting. So, the plan
  55. 1:51is, quit our jobs, start from zero, and
  56. 1:54build a chip for a market that barely
  57. 1:56exists?
  58. 1:57Correct. I hate that this sounds insane
  59. 2:00and also kind of logical.
  60. 2:02They called the company Nvidia.
  61. 2:04The idea was ambitious. Build one
  62. 2:06powerful [music] graphics chip that
  63. 2:07could handle everything. A clean
  64. 2:10universal chip for the future of gaming
  65. 2:12hardware.
  66. 2:13For 2 years they worked. They spent
  67. 2:15every dollar they had. And in 1995,
  68. 2:18Nvidia launched its first product, the
  69. 2:20NV1.
  70. 2:22Jensen, we have a problem. Microsoft
  71. 2:25just [music] published the Direct X
  72. 2:26spec, full triangle based rendering
  73. 2:28pipeline. Holy crap, the NV1 runs
  74. 2:31quadratic mapping?
  75. 2:33Yes.
  76. 2:34We are the only ones going in the
  77. 2:35opposite direction. Games are about to
  78. 2:37be built from tiny flat triangles?
  79. 2:40Our chip is built for curved surfaces.
  80. 2:42We built the wrong chip with tremendous
  81. 2:44confidence. 2 years, every dollar. A
  82. 2:47chip built for a world that no longer
  83. 2:48existed. The NV1 [music] flopped.
  84. 2:51But Nvidia still had a lifeline. Sega
  85. 2:54had contracted them to build the
  86. 2:55graphics processor for the company's
  87. 2:57next generation console,
  88. 2:59the Dreamcast.
  89. 3:01Real money, real chance to start over.
  90. 3:04There was only one problem.
  91. 3:06Sega's chip used the same dead-end
  92. 3:08technology. [music] You're doing the
  93. 3:09Sega math again, aren't you?
  94. 3:11We'll spend the next year building
  95. 3:12something we already know is dead. And
  96. 3:14if we cancel, we lose the only customer
  97. 3:17still paying us.
  98. 3:18Uh, Pretty much.
  99. 3:21I miss normal employment.
  100. 3:23If we're going to die, let's die
  101. 3:25building the right chip.
  102. 3:26>> [music]
  103. 3:26>> So Jensen flew to Japan.
  104. 3:29The chip is a dead end. You should
  105. 3:31cancel our contract and find someone
  106. 3:32else.
  107. 3:33We shouldn't finish the contract. It
  108. 3:35would be a waste of your money.
  109. 3:36>> [music]
  110. 3:38>> There's one more thing. I still need the
  111. 3:40money, the $5 million remaining on our
  112. 3:42contract. Please put it into Nvidia as
  113. 3:44an investment instead. Otherwise, we'll
  114. 3:46vaporize overnight.
  115. 3:47>> [music]
  116. 3:48>> I have nothing to offer you. This money
  117. 3:50will most likely be lost, but I'm asking
  118. 3:52anyway.
  119. 3:55I'll need a few days.
  120. 3:56Jensen [music] bowed, walked out, and
  121. 3:58got on a plane. Nothing to do now but
  122. 4:01sit, and wait, and watch the Pacific
  123. 4:03stretch out beneath him.
  124. 4:05Jensen, you might want to see this.
  125. 4:08He opens [music] it, eyes wide. It's a
  126. 4:10massive 10% offer on the value of
  127. 4:13BetterHelp.
  128. 4:15Just kidding.
  129. 4:16Let's be real for a second. When the
  130. 4:18outcome is completely out of your hands,
  131. 4:20the anxiety Did I make the right call?
  132. 4:23That noise won't switch off. It doesn't
  133. 4:25go away on its own.
  134. 4:27The hardest part for me was noticing the
  135. 4:29pattern.
  136. 4:31I was pushing through instead of
  137. 4:33actually dealing with anything.
  138. 4:35My therapist asked me one question.
  139. 4:37Are you solving something, or are you
  140. 4:39rehearsing a disaster?
  141. 4:41I didn't have an answer. Just having a
  142. 4:43name for what I was doing,
  143. 4:45that single reframe broke the loop.
  144. 4:48BetterHelp, the paid [music] partner of
  145. 4:49today's video, makes starting therapy
  146. 4:51easier.
  147. 4:52Take a short quiz, get matched with a
  148. 4:54credential therapist, and meet by phone,
  149. 4:56[music] video, or text. Whatever works
  150. 4:59for your schedule.
  151. 5:01If it's not the right fit, switching
  152. 5:02costs nothing extra.
  153. 5:05And it's working.
  154. 5:06Over 6 million people have gotten help
  155. 5:08[music] through BetterHelp, rated 4.8
  156. 5:10out of five on the App Store.
  157. 5:13Click the link in the description,
  158. 5:13[music]
  159. 5:14or go to betterhelp.com/crowncapital
  160. 5:16for 10% off your first month of therapy.
  161. 5:19Every great ship [music] needs
  162. 5:20maintenance. So do you.
  163. 5:25Hello.
  164. 5:26The money is coming.
  165. 5:28The reason the Sega CEO said yes had
  166. 5:31nothing to do with business logic. He
  167. 5:33just liked Jensen.
  168. 5:35That $5 million bought Nvidia 6 months.
  169. 5:38And now they had one last shot to build
  170. 5:40the right chip, or
  171. 5:41>> [music]
  172. 5:41>> go bankrupt.
  173. 5:44They called it the Riva 128,
  174. 5:46triangle-based, [music]
  175. 5:47DirectX compatible, built from scratch.
  176. 5:51Weeks of late nights, bad coffee, and
  177. 5:53absolutely zero plan B. They had exactly
  178. 5:561 month of payroll [music] left in the
  179. 5:58bank to find out if it worked.
  180. 6:01The chip launched in 1997, and this time
  181. 6:04Nvidia did not miss. Game developers
  182. 6:06could actually use it. Customers
  183. 6:08actually bought it. From the brink of
  184. 6:09bankruptcy to a million units sold in 4
  185. 6:13months.
  186. 6:14Most CEOs would have taken a long
  187. 6:16vacation and slow down a bit.
  188. 6:18He walked back into the office and wrote
  189. 6:20one word on the whiteboard.
  190. 6:22More.
  191. 6:24So, by 1999, Nvidia launched the GeForce
  192. 6:27256 and gave the world a term it didn't
  193. 6:29really have yet, GPU.
  194. 6:32Graphics processing unit.
  195. 6:34It was a genuine leap. A chip that
  196. 6:36handled complex 3D calculations that
  197. 6:38previously required an entirely separate
  198. 6:40processor. For gamers,
  199. 6:42>> [music]
  200. 6:43>> this meant smoother worlds, better
  201. 6:45lighting, and monsters that looked
  202. 6:46slightly less like wet cardboard. Gamers
  203. 6:49loved it. Developers wept [music] with
  204. 6:51joy.
  205. 6:52Revenue exploded. And by the early
  206. 6:542000s, Nvidia had become the king of PC
  207. 6:57graphics. Every serious gamer, every
  208. 7:00visual effects studio, every game
  209. 7:02developer on the planet. The obvious
  210. 7:04move was simple. Keep making better
  211. 7:06gaming chips, sell them to gamers,
  212. 7:08become rich, buy a very large leather
  213. 7:10jacket museum.
  214. 7:11But Jensen started looking at something
  215. 7:13other people missed.
  216. 7:15A CPU is like one very smart employee
  217. 7:18doing tasks one by one extremely fast.
  218. 7:21Meanwhile, a GPU is like 10,000 interns
  219. 7:24doing [music] tiny calculations at once.
  220. 7:26Are the interns smart?
  221. 7:28No.
  222. 7:29But there are a lot of them, and if your
  223. 7:31problem can be broken into thousands of
  224. 7:32small calculations, physics, chemistry,
  225. 7:35weather, biology, artificial
  226. 7:37intelligence,
  227. 7:39suddenly, the chip built for video games
  228. 7:41[music] starts looking like something
  229. 7:42else, a cheap supercomputer.
  230. 7:45Jensen looked at this and thought, "What
  231. 7:47if the same chip that rendered
  232. 7:48explosions in video games could help
  233. 7:50simulate molecules, predict weather,
  234. 7:53model fluids, and maybe one day train
  235. 7:55machines to think?"
  236. 7:57So, in 2006, Nvidia launched CUDA,
  237. 8:00Compute Unified Device Architecture, a
  238. 8:03platform that let scientists,
  239. 8:04researchers, and engineers use Nvidia
  240. 8:07GPUs [music] for general-purpose
  241. 8:08computing.
  242. 8:10And then, he committed roughly $500
  243. 8:12million to build it.
  244. 8:14Jensen,
  245. 8:16the gaming business is printing money.
  246. 8:18Walk me through who CUDA is actually
  247. 8:20for.
  248. 8:21Researchers, scientists, people running
  249. 8:23fluid dynamics simulations, climate
  250. 8:26models, drug discovery?
  251. 8:28So, not our customers.
  252. 8:30Not yet.
  253. 8:32I'm starting to hate that phrase. This
  254. 8:34is our CUDA revenue after 2 years. Do
  255. 8:36you see a number?
  256. 8:37Not today.
  257. 8:39Jensen, this is not a business. We are
  258. 8:42subsidizing a hobby for PhD students.
  259. 8:45PhD students are where the future
  260. 8:47starts. For the first time, they can use
  261. 8:49a gaming chip like a supercomputer.
  262. 8:51We don't know what they'll build with
  263. 8:53it. But when they build [music]
  264. 8:54something important, it will run on us.
  265. 8:56If they build nothing,
  266. 8:58we still have the best gaming cards in
  267. 9:00the world and a $500 million science
  268. 9:03project. We've survived worse.
  269. 9:06>> [music]
  270. 9:06>> Six years passed. The gaming business
  271. 9:08kept printing money. CUDA kept printing
  272. 9:11almost nothing.
  273. 9:12And Jensen kept waiting for the problem
  274. 9:14big enough to prove him right.
  275. 9:16Then, in Toronto, someone found it.
  276. 9:20It's 2012. Every year, the world's top
  277. 9:23AI researchers compete in a challenge
  278. 9:24[music] called ImageNet. Tens of
  279. 9:27millions of images, and your software
  280. 9:29has to identify what's in them.
  281. 9:31You see the Toronto submission? Some PhD
  282. 9:34student [music] who running a a network
  283. 9:35on gaming cards. We've been hand-coding
  284. 9:37classifiers for a decade. Gaming cards
  285. 9:40aren't going to crack this.
  286. 9:41Ladies and gentlemen, thank you for
  287. 9:43waiting.
  288. 9:44The ImageNet [music] 2012 results are
  289. 9:46final. Second place, University of Tokyo
  290. 9:50with a 26.2% error rate. And this year's
  291. 9:53winner,
  292. 9:54University of Toronto, AlexNet, 15.3%.
  293. 9:59I'm sorry, 15?
  294. 10:00That's not a win. That's a murder scene.
  295. 10:03For years, researchers have been trying
  296. 10:05to teach computers how to see by
  297. 10:07manually designing the features they
  298. 10:09should look for.
  299. 10:10Edges, textures, shapes, patterns.
  300. 10:14AlexNet took a different path. Give the
  301. 10:16neural network enough data, enough
  302. 10:18layers, and enough computing power,
  303. 10:21and let it learn the features itself.
  304. 10:24Training a neural network requires
  305. 10:26millions of small calculations happening
  306. 10:28at the same time.
  307. 10:29Exactly the kind of work that Nvidia had
  308. 10:32been quietly perfecting for 6 years.
  309. 10:35The $500 million bet
  310. 10:36>> [music]
  311. 10:36>> had finally found its customer.
  312. 10:38Suddenly, every major AI lab started
  313. 10:40paying attention. Stanford, MIT, Google,
  314. 10:45Meta, Microsoft.
  315. 10:47They weren't just buying graphics cards
  316. 10:49[music] anymore.
  317. 10:50They were buying computing power for
  318. 10:51intelligence itself.
  319. 10:53And Nvidia had the hardware, the
  320. 10:55software, and the scale to deliver it.
  321. 10:58Great news for Nvidia, until in 2022,
  322. 11:00[music]
  323. 11:01Washington made a phone call.
  324. 11:05Before Washington called, Jensen had
  325. 11:07already tried to fix Nvidia's biggest
  326. 11:08weakness. Because Nvidia designed the
  327. 11:10chips,
  328. 11:12but it didn't make them.
  329. 11:13That job belonged mostly to TSMC in
  330. 11:16Taiwan.
  331. 11:17And he didn't own the underlying
  332. 11:19architecture, the fundamental
  333. 11:20instruction set that tells processors
  334. 11:22how to think.
  335. 11:24That belonged to a British company
  336. 11:25called ARM.
  337. 11:27A company whose technology sat inside
  338. 11:29virtually every smartphone on Earth.
  339. 11:31Apple, Samsung, Qualcomm, your pocket.
  340. 11:35Their DNA, not Jensen's.
  341. 11:38So, for all of Nvidia's power, [music]
  342. 11:39Jensen was still renting the building.
  343. 11:42He owned the furniture.
  344. 11:43So, in 2020,
  345. 11:44>> [music]
  346. 11:44>> Jensen made his move. Nvidia offered $40
  347. 11:47to buy ARM.
  348. 11:49Own the GPU, own the software, own the
  349. 11:52architecture. Simple.
  350. 11:54Unless you were literally everyone else
  351. 11:55in tech.
  352. 11:57Nvidia cannot own ARM. Agreed. Agreed.
  353. 11:59Also agreed.
  354. 12:01Wow, I hate how united we are. If Jensen
  355. 12:04owns ARM, then every chip we build runs
  356. 12:06through our most dangerous competitor.
  357. 12:08Phones,
  358. 12:09>> [music]
  359. 12:09>> servers, AI chips, everything. So, what
  360. 12:12do we do?
  361. 12:13We complain to every regulator with an
  362. 12:15email address.
  363. 12:17And they did. The FTC sued. The European
  364. 12:20Commission investigated. The UK blocked
  365. 12:22it on national security grounds.
  366. 12:2418 months later, Nvidia walked away.
  367. 12:27Then Washington called. Mr. Huang,
  368. 12:30effective immediately, advanced AI
  369. 12:32chips, the H100,
  370. 12:34>> [music]
  371. 12:34>> can no longer be exported to China.
  372. 12:37How much revenue are we talking about?
  373. 12:39That sounds like a you problem.
  374. 12:41Just like that, some of Nvidia's biggest
  375. 12:43customers were cut off.
  376. 12:45Alibaba, Baidu, ByteDance, China's AI
  377. 12:49companies,
  378. 12:50gone.
  379. 12:51So, Nvidia did what Nvidia always did.
  380. 12:53It engineered around the problem.
  381. 12:56They built a weaker version, the H800,
  382. 12:58designed to fit inside the rules.
  383. 13:01Legal enough to ship. Powerful enough
  384. 13:03[music] to sell.
  385. 13:04For about 5 minutes, because by late
  386. 13:062023, Washington tightened the rules
  387. 13:08again.
  388. 13:09The H800 was restricted to
  389. 13:11>> [music]
  390. 13:12>> The message was clear. Nvidia was now a
  391. 13:14strategic weapon in America's technology
  392. 13:16war with China.
  393. 13:18Jensen wasn't just a CEO anymore. He was
  394. 13:21a piece on someone else's chessboard,
  395. 13:23moved by people he couldn't negotiate
  396. 13:24with.
  397. 13:25But while politics got messier, demand
  398. 13:27got insane. OpenAI, Google, [music]
  399. 13:30Anthropic, xAI, Microsoft. Everyone
  400. 13:34building large AI models needed the same
  401. 13:36thing, more Nvidia chips, more data
  402. 13:39centers, more power, more cooling, more
  403. 13:42everything.
  404. 13:44Blackwell, Nvidia's next [music]
  405. 13:45generation chip, was being ordered
  406. 13:47before factories could even finish
  407. 13:48making it.
  408. 13:50By October 2025, Nvidia [music] crossed
  409. 13:52$5 trillion in market value.
  410. 13:54The most valuable company on Earth,
  411. 13:56ahead of Google, ahead of Apple.
  412. 13:59Not bad for a company that once had 30
  413. 14:01days of cash left and one very wrong
  414. 14:03chip.
  415. 14:04None of it was supposed to work.
  416. 14:06Every single [music] bet looked insane
  417. 14:08from the outside, and every single one
  418. 14:10paid off.
  419. 14:11As what he himself [music] said,
  420. 14:13"My will to survive exceeds everybody
  421. 14:15else's will to kill me."

About this transcript

This page contains the full transcript of NVIDIA Explained Like You're 5 by Crayon Capital, generated from the public captions YouTube serves with the video. The transcript has 2,199 words across 421 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.