The First Big Victim of the AI Bubble — Transcript
Full transcript
- 0:00July 2026, this 24-year-old loses $35
- 0:04billion in a single month betting on AI.
- 0:07The largest trading loss in history.
- 0:10But he didn't lose because his
- 0:12prediction was wrong.
- 0:13>> [music]
- 0:13>> It was coming true.
- 0:14So how do you see the future correctly
- 0:17and still lose $35 billion?
- 0:20To answer that, we need to start with
- 0:22the biggest problem in 2024.
- 0:27The world has an AI problem. Not the AI
- 0:30is coming for your job problem.
- 0:32>> [screaming]
- 0:32>> AI was getting good, scary good. And
- 0:36whoever ended up controlling it would
- 0:37hold more power than any company in
- 0:39history.
- 0:40So the solution seems obvious, right? If
- 0:43you can't beat them, join them.
- 0:45Invest in the winner. Problem solved.
- 0:48Except nobody could tell you who the
- 0:49winner was. We're first. We're the best.
- 0:53>> [music]
- 0:53>> We invented this. Cute. We have your
- 0:55entire internet and a nuclear reactor.
- 0:59Is everyone using him now?
- 1:01Every few months, the leaderboard
- 1:03flipped. A company that was dead last in
- 1:05March was untouchable by September.
- 1:08Then somebody else leapfrogged them
- 1:10>> [music]
- 1:10>> and the whole thing reset. For years,
- 1:12Wall Street tried to analyze it.
- 1:14Nothing. No pattern. It's too
- 1:16unpredictable.
- 1:18Then a 22-year-old claimed he'd figured
- 1:20it out. Not just who would win,
- 1:22but how the entire decade would unfold.
- 1:25Meet the boy wonder, Leopold
- 1:27Aschenbrenner. [music]
- 1:29Germany, 2001.
- 1:32Leopold Aschenbrenner is born to two
- 1:34doctors. Normal childhood except for
- 1:36[music] one detail. The kid is early to
- 1:38everything. He enrolled at Columbia
- 1:40University at 15, graduated
- 1:42valedictorian with a degree in economics
- 1:44and math, and a nearly perfect test
- 1:46score he was so proud of, he put it on
- 1:48LinkedIn.
- 1:49Wow. Okay. And your greatest weakness?
- 1:53I finish things too early.
- 1:56So in 2022, he did what every brilliant,
- 1:59idealistic 21-year-old did that year.
- 2:01Went to work for the smartest guy in the
- 2:03room.
- 2:04That guy was Sam Bankman-Fried.
- 2:06Leopold joined the FTX Future Fund, the
- 2:09charity arm of the crypto exchange,
- 2:11staffed by young people who genuinely
- 2:12believed they were about to give away
- 2:14billions to save the world.
- 2:16For a few months, it felt like the
- 2:18center of the universe. Everyone was
- 2:20brilliant. Everyone was certain. The
- 2:23money seemed infinite. And then,
- 2:25November 2022 happened. FTX collapsed,
- 2:28the billions vanished, and Leopold got a
- 2:30front-row seat to what happens when a
- 2:32brilliant person, a compelling story,
- 2:34and borrowed money collide.
- 2:37After FTX, Leopold moved to the company
- 2:40building the future he was most worried
- 2:41about, OpenAI.
- 2:44He joined its superalignment team,
- 2:45>> [music]
- 2:45>> whose job was essentially to make sure a
- 2:47superhuman AI did not destroy
- 2:49civilization.
- 2:50Small assignment, but less than a year
- 2:52later, OpenAI fired him for allegedly
- 2:55leaking internal information.
- 2:57Leopold disputed that description.
- 2:59He said he had shared an AI safety
- 3:00document with outside researchers.
- 3:03Either way, he was out.
- 3:04So, picture him now.
- 3:0622,
- 3:07no job, no money,
- 3:10fired from the most important company on
- 3:11the planet. He had exactly one thing
- 3:13left, an opinion.
- 3:15In June 2024, Leopold published 165
- 3:19pages explaining exactly what he
- 3:21believed was about to happen.
- 3:23He called it Situational Awareness: The
- 3:25Decade Ahead.
- 3:27Its argument was simple enough to sound
- 3:28insane. AI models were getting smarter
- 3:31on a straight, [music] predictable line.
- 3:33Draw that line forward, he said, and
- 3:35machines hit human-level intelligence
- 3:37around 2027. Shortly after that, [music]
- 3:40smarter than us. Game over.
- 3:42Most people read that and started
- 3:44arguing about robots stealing their
- 3:45jobs. Leopold saw a shopping list.
- 3:48Because smarter AI needs bigger models.
- 3:51Bigger models [music] need more chips.
- 3:53Those chips need memory.
- 3:55All of it needs vast humming data
- 3:57[music] centers.
- 3:59And those data centers need so much
- 4:00electricity that the power companies
- 4:02hadn't even started planning for it.
- 4:04To build all of this, the world might
- 4:06[music] need to spend a trillion
- 4:07dollars.
- 4:08And whoever supplied that construction
- 4:09boom was about to get very, very rich.
- 4:13Okay, so you might be wondering why call
- 4:15it situational awareness. Well, it is
- 4:17exactly because screw-ups like you
- 4:19[music] are not aware of this situation.
- 4:21Only a few hundred people saw what was
- 4:23coming.
- 4:24And obviously,
- 4:25>> [music]
- 4:25>> I was one of them.
- 4:26But you should be aware of this quick ad
- 4:28break, though.
- 4:30How many AI subscriptions are you paying
- 4:31for right now?
- 4:33Chat GPT for research, Claude for
- 4:35coding, something else for images, and
- 4:37every one of them wants another $20 a
- 4:39month.
- 4:40Chat LLM by Abacus AI fixes [music]
- 4:42that. For $10 a month, you get GPT-6
- 4:45Astra, Claude Fable 5.1, Gemini,
- 4:48>> [music]
- 4:49>> Grok, Deep Seek, and more in one
- 4:51dashboard.
- 4:52New models appear the day they [music]
- 4:53launch, and if you don't know which one
- 4:55to use, Root LLM picks the best model
- 4:57for your prompt. Use it to research and
- 4:59write documents, create presentations,
- 5:01and generate images [music] and videos.
- 5:04You can also describe an idea and have
- 5:06its AI agent build the whole website or
- 5:08app. Front end, back end, hosting,
- 5:10[music]
- 5:11even payments. You can also create AI
- 5:13agents that work like personal
- 5:14assistants 24/7. [music]
- 5:18Head over to chat.llm.abacus.ai
- 5:21or [music] click the link in my
- 5:22description to get started. Trust me,
- 5:24it's the only AI bet that won't cost you
- 5:26billions.
- 5:28Okay, back to the video.
- 5:30The essay spread through Silicon Valley.
- 5:32Investors passed it around. Podcasters
- 5:35read [music] it. People started calling
- 5:37Leopold the Nostradamus of AI.
- 5:40Now, most writers would have taken the
- 5:41fame, launched a podcast, and sold a
- 5:43course for $499.
- 5:46Leopold did something far more
- 5:48dangerous.
- 5:49He decided to bet real money that his
- 5:50prophecy was true.
- 5:52With other people's money.
- 5:53There was just one problem. In his
- 5:55entire life, he had never professionally
- 5:58managed a single dollar.
- 6:02July 2024.
- 6:04Leopold launches a hedge fund. He names
- 6:06it what else?
- 6:08>> [music]
- 6:08>> Situational Awareness.
- 6:10By every rule of Wall Street, raising
- 6:12money should have been impossible.
- 6:14>> [music]
- 6:14>> I am 22. I had no track record, and my
- 6:17resume included one collapsed crypto
- 6:19empire
- 6:19>> [music]
- 6:19>> and one recent firing.
- 6:22But I want to manage your money.
- 6:24Right.
- 6:25You've read the essay.
- 6:26Do you think it's wrong?
- 6:28No.
- 6:29I think it's right.
- 6:31Then trust your own conclusion.
- 6:33I'm early at everything. Give me my
- 6:35money.
- 6:36It worked. His backers reportedly
- 6:38included some of the most respected
- 6:39names in tech, the founders of Stripe, a
- 6:42former CEO of GitHub, big-name investors
- 6:44who'd sat through a thousand pitches and
- 6:46seen every trick in the book.
- 6:48Then Leopold built the trade. And
- 6:50honestly,
- 6:52it was gorgeous.
- 6:54See, all of Wall Street was obsessed
- 6:56with the same question.
- 6:58Which company wins AI? OpenAI? Google?
- 7:01[music] Anthropic? Some startup that was
- 7:04currently two guys and a Discord server?
- 7:07Leopold thought that was the amateur's
- 7:08question.
- 7:09During a gold rush, he figured, you
- 7:11don't need to guess which prospector
- 7:12strikes gold. Half of them go broke.
- 7:15Half of them go crazy.
- 7:17You don't bet on the prospectors at all.
- 7:19You sell every single one of them a
- 7:21shovel, and you get paid whether they
- 7:22strike gold or die of dysentery.
- 7:25So, step one,
- 7:26buy the picks and shovels. Memory. Data
- 7:29centers. Power.
- 7:31Names like SK Hynix, Micron, CoreWeave,
- 7:34Nebius, Bloom Energy.
- 7:36The unglamorous companies that simply
- 7:38had to exist, no matter who won the
- 7:40actual race.
- 7:42Step two, and this is the clever half.
- 7:45Short the victims.
- 7:47Bet against the software companies that
- 7:49are going to be killed by AI, such as
- 7:51Adobe. Why pay a monthly subscription
- 7:53for the tool when the robot does the
- 7:55same job for free?
- 7:57That was the brilliance of the
- 7:58portfolio.
- 7:59Owning both sides was supposed to
- 8:01protect him from a market crash. If the
- 8:03whole market fell, he would lose money
- 8:05on the stocks he bought, but make money
- 8:07on the stocks he shorted. Ideally, one
- 8:09would cancel out the other.
- 8:11He was simply betting that the AI
- 8:13winners would do better than the AI
- 8:14victims.
- 8:15But it was also the trap.
- 8:17Leopold ran a few dozen positions, but
- 8:19almost every one of them depended on the
- 8:21same sentence being true.
- 8:23AI will transform the world, and it will
- 8:25happen on schedule.
- 8:27For a while, that sentence looked
- 8:28correct. First half of 2025, up 47%
- 8:32after fees, managing about 1.5 billion
- 8:36dollars.
- 8:37Solid.
- 8:38>> [music]
- 8:38>> Then the returns became absurd. By late
- 8:40June 2026, the fund was reportedly up
- 8:43around 439% for the year, and more than
- 8:451,000% since launch.
- 8:48Assets passed to 20 billion dollars. At
- 8:50its peak, [music] the firm controlled
- 8:52roughly 45 billion dollars. In less than
- 8:54two years, a man who had never traded
- 8:56professionally had built one of the most
- 8:58watched funds on Wall Street. Other
- 9:00investors [music] started admiring him,
- 9:02and
- 9:03the filing is out.
- 9:05What's he holding?
- 9:07Memory? Power? Data centers? Same as
- 9:10last quarter, but bigger. Buy all of it.
- 9:12You don't want to look at the
- 9:13fundamentals first?
- 9:15He read the future. I'm reading him.
- 9:17It's faster.
- 9:18Every new buyer pushed Leopold's
- 9:20holdings higher. Those higher prices
- 9:22improved his returns. Better returns
- 9:24attracted more investors.
- 9:26>> [music]
- 9:26>> More investor money allowed the fund to
- 9:28buy even more. So, success created more
- 9:32success. But, underneath it was a the
- 9:35engine.
- 9:36Leverage.
- 9:37Leopold wasn't just investing his
- 9:39investors' money, he was borrowing
- 9:41against it. Roughly three to four
- 9:43dollars of debt for every dollar he
- 9:44actually had.
- 9:46Here is why that feels wonderful. You
- 9:47have $100. The bank lends you $400. Now
- 9:51you own $500 of stock.
- 9:54If those stocks rise 20%, you make $100
- 9:57and double your original money.
- 9:59That sounds incredible. It is.
- 10:02But please do not ask what happens if
- 10:04the stocks fall 20% because leverage
- 10:06does not simply make the good days
- 10:08better. It puts a clock on the bad days.
- 10:11If the portfolio falls, the bank asks
- 10:13for more collateral. It's called a
- 10:15margin call.
- 10:16If you cannot provide cash, you must
- 10:18sell stocks.
- 10:19Not next year when your prediction comes
- 10:21true.
- 10:22Today. [music]
- 10:24Mr. Leopold, history is littered with
- 10:26investors who saw the future correctly,
- 10:28but still went broke because their
- 10:29timing was catastrophically wrong.
- 10:32So what happens if your stocks go down
- 10:34before your thesis plays out?
- 10:36>> [music]
- 10:36>> Yeah, great question. I think the only
- 10:38thing going down is my estimate of your
- 10:41situational awareness.
- 10:43I'm serious. So am I. [music] Obviously,
- 10:46not blowing up is task number one and
- 10:48two.
- 10:49Then on July 10th, 2026, [music]
- 10:52the company at the center of Leopold's
- 10:53entire thesis rang the opening bell in
- 10:56New York. It was supposed to be his
- 10:58victory lap.
- 10:58>> [music]
- 10:59>> Instead, it started the clock.
- 11:03The company was SK Hynix, the South
- 11:06Korean memory chip giant. If Leopold's
- 11:08AI prediction was correct, the world
- 11:10would need an extraordinary [music]
- 11:12amount of high-performance memory.
- 11:14That company made it. And Leopold had
- 11:16reportedly put 6.5% of his money there.
- 11:19This was the purest version of his
- 11:21trade.
- 11:22SK Hynix crossed onto the American
- 11:24market with enormous [music] hype.
- 11:26It rose by 13% on the first day, but it
- 11:29didn't last.
- 11:30The stock fell.
- 11:32At first, that should not have mattered.
- 11:34Nothing about the actual future had
- 11:36changed.
- 11:36>> [music]
- 11:37>> No data center got canceled. No AI
- 11:39company stopped buying chips. It was
- 11:41just the ordinary weather of the stock
- 11:42market. On a 10-year chart, pure noise.
- 11:45[music]
- 11:46But Leopold didn't own SK Hynix with
- 11:4810-year money. He owned it with money
- 11:50that could demand to be repaid today.
- 11:53Leveraged bets on Korean stocks began
- 11:54unwinding. Regulators in Seoul rushed to
- 11:57make traders [music] put up more cash.
- 11:59Anyone who could not provide it was
- 12:01forced to sell.
- 12:02That selling pushed prices lower, which
- 12:04triggered more margin calls,
- 12:06>> [music]
- 12:06>> and created more sellers.
- 12:08SK Hynix stock eventually fell more
- 12:10[music] than 30% from their peak.
- 12:13Then the fire spread to the rest of
- 12:14Leopold's portfolio.
- 12:16Micron fell. Coreweave, Nebius.
- 12:19Some of the AI infrastructure stocks
- 12:21dropped by half or more.
- 12:24And then his supposedly protective short
- 12:25positions turned against him, too.
- 12:28Remember the clever half?
- 12:29As terrified investors fled the flashy
- 12:31AI names, they ran back into the boring,
- 12:34safe software companies.
- 12:36The exact companies Leopold had bet
- 12:37against. So his longs fell while his
- 12:40shorts rose.
- 12:41Now the leverage took control.
- 12:43Introducing the doom loop.
- 12:46Your stocks fall.
- 12:48Your cushion of real money shrinks.
- 12:50The bank calls, [music] asking for more
- 12:52cash to cover the loan. But you don't
- 12:55have cash.
- 12:56So you [music] sell stocks. And selling
- 12:58that much, that fast, pushes the price
- 13:00down, which shrinks your cushion again,
- 13:02and triggers the next call, [music] and
- 13:04the next sale.
- 13:06Fall, call, sell. Repeat. Faster every
- 13:10turn.
- 13:11By Friday, July 24th, the damage was
- 13:13severe enough that Leopold [music] sent
- 13:15a letter warning investors that the fund
- 13:16had suffered heavy losses. And also
- 13:19asked them for more money.
- 13:20In his mind, nothing has changed about
- 13:22the thesis. Not one data center got
- 13:25canceled this week.
- 13:26Only the price had. Everything he wanted
- 13:28to own had just gone on sale. [music]
- 13:31So, he called it the best buying
- 13:32opportunity since early 2025. But, the
- 13:36letter actually exposed how vulnerable
- 13:37he was. His rivals already knew what he
- 13:39owned.
- 13:40Now, they knew he might be forced to
- 13:42sell.
- 13:43So, they sold first, pushing prices down
- 13:45before he could get out.
- 13:47Over the next three trading days,
- 13:48several of his largest positions dropped
- 13:51another 30% or more.
- 13:53The prophecy had become a countdown.
- 13:55And Leopold [music] had somewhere else
- 13:57he was supposed to be that week.
- 13:58Guests were flying into Carmel, a
- 14:00seaside town on the California coast,
- 14:02[music] for his wedding.
- 14:03A multi-day celebration, a villa
- 14:05ceremony, a forest spa honeymoon
- 14:07send-off, and
- 14:08>> [music]
- 14:09>> uh because this was Leopold, a
- 14:10pre-wedding colloquium with panels and
- 14:12breakout sessions. So, while his guests
- 14:14packed, the groom worked the phones.
- 14:17"Hey, Jim Bob. [music] I'm offering 10
- 14:19billion in positions at a discount."
- 14:21"How big a discount?" Uh "Name it."
- 14:24"Nah."
- 14:25"I'll wait."
- 14:26"Tomorrow, it's cheaper."
- 14:28That was the answer everywhere he
- 14:30called. Without a deal, he'd have to
- 14:32dump billions in stocks onto the market,
- 14:34pushing [music] prices down even
- 14:35further.
- 14:36He needed someone who could take the
- 14:37entire portfolio off his hands in one
- 14:39transaction.
- 14:41Longs, shorts, everything.
- 14:43Someone with enough money and the balls
- 14:45to buy what the rest of Wall Street
- 14:47wouldn't touch.
- 14:48And one man was willing to do it.
- 14:52Introducing
- 14:53Ken Griffin, founder of Citadel.
- 14:56Griffin and his lieutenants [music]
- 14:57pulled an all-nighter tearing through
- 14:59Leopold's book, position by position,
- 15:01working out what could actually be sold
- 15:04and what couldn't. By morning, [music]
- 15:06he had an offer.
- 15:07"Okay, my offer is
- 15:09I would take the entire [music] $16
- 15:11billion position, but I want a 10%
- 15:13discount." "10%? The prices are already
- 15:16down a lot."
- 15:18"Correct. That's what makes them worth
- 15:20owning."
- 15:21"What if they keep falling?
- 15:23Then I hold them.
- 15:25That's the difference between us.
- 15:26You're selling time.
- 15:28I'm buying it.
- 15:29And Griffin knew exactly what time was
- 15:31worth because 18 years earlier he'd been
- 15:34the one who couldn't afford it. Back in
- 15:362008, Citadel's biggest funds lost
- 15:38[music] 55%.
- 15:40Griffin had spread his bets across
- 15:42stocks, bonds, credit. And when the
- 15:44system broke, everything broke together.
- 15:47Then his investors asked for their money
- 15:49back. Paying them meant selling into the
- 15:51panic, locking in every loss before
- 15:53anything could [music] recover.
- 15:55So Griffin didn't pay them.
- 15:57He locked the doors and refused
- 15:58withdrawals for months.
- 16:00Investors were furious. Investors were
- 16:02stuck.
- 16:03And Citadel got the one thing Leopold no
- 16:05longer had, the ability to choose when
- 16:08to sell.
- 16:09Now in 2026, [music]
- 16:11he was buying from a man who had lost
- 16:13that choice completely.
- 16:14The trade cleared overnight.
- 16:16And the moment it did, the doom loop had
- 16:18nothing left to eat.
- 16:20Coreweave, SK Hynix, Bloom Energy.
- 16:23The same stocks that had been falling
- 16:25for three straight weeks started
- 16:26climbing.
- 16:27Some of them finished the day up more
- 16:29than 25%. [music]
- 16:31Citadel had reportedly resold most of
- 16:33the portfolio at a substantial profit.
- 16:35Its stock picking fund
- 16:36>> [music]
- 16:36>> finished July up 14.2%
- 16:39the best month in the firm's history.
- 16:42Meanwhile, for Leopold,
- 16:44his fund had fallen from roughly $45
- 16:46billion at its peak
- 16:47>> [music]
- 16:47>> to around $10 billion.
- 16:49A private stake in Anthropic remained,
- 16:51helping preserve what was left.
- 16:53$35 billion [music] gone.
- 16:56He wrote to his investors and said,
- 16:58"Dear partners,
- 17:00we let you down this month.
- 17:02We came closer [music] to permanent
- 17:03capital impairment than is acceptable to
- 17:05us.
- 17:06Over the past 2 years, we have delivered
- 17:08outstanding [music] results despite
- 17:10occasional sharp pullbacks. But our fund
- 17:13must always be structured [music] such
- 17:15that we can take a loss and fight
- 17:17another day.
- 17:18I take full responsibility for these
- 17:20events.
- 17:21We are continuing [music] to operate.
- 17:23However, we will manage our public book
- 17:25on a fully paid for basis.
- 17:27These were very expensive scars,
- 17:29>> [music]
- 17:29>> but my core promise to you is that we
- 17:31will not waste the opportunity to learn
- 17:33from these events.
- 17:35On the portfolio itself, we are very
- 17:37[music] optimistic. The underlying
- 17:39fundamentals are accelerating at the
- 17:41very same time that prices have declined
- 17:43significantly.
- 17:45Thank you for your patience and your
- 17:46partnership.
- 17:47>> [music]
- 17:48>> I am fully invested alongside you.
- 17:50Virtually all of my capital is in the
- 17:52fund. I'm available next week for
- 17:54one-to-one [music] calls with each of
- 17:55you to discuss all of the above.
- 17:59As an interim update, [music] our
- 18:00current net month-to-date performance is
- 18:02minus 67%
- 18:04and of net year-to-date [music]
- 18:05performance is plus 80%.
- 18:08We will follow up to share further
- 18:09details.
- 18:11All the best, Leopold.
- 18:14That is the story of Leopold
- 18:16Aschenbrenner,
- 18:17>> [music]
- 18:17>> the kid who called the decade,
- 18:19outperformed almost everyone on Wall
- 18:21Street, and still said $35 billion on
- 18:24fire in a single month.
- 18:26>> [music]
- 18:26>> Early to college, early to the future,
- 18:29and almost finished his story
- 18:31early.
About this transcript
This page contains the full transcript of The First Big Victim of the AI Bubble by Crayon Capital, generated from the public captions YouTube serves with the video. The transcript has 3,018 words across 575 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.