쉐도잉 연습: The AI Take Over Has Completely Backfired (Here's the Proof) - 영상으로 영어 말하기 배우기

연습 모드 선택

각 모드는 전체 화면 연습실로 열립니다.

128 문장

문장이 너무 짧거나 길면 Edit를 눌러 조정하세요.

AI was supposed to change everything, replace millions of jobs, transform every industry, and become the biggest technological shift in history.

But there's just one problem.

The numbers don't seem to add up.

Big tech is expected to spend around $725 billion on AI this year.

Meanwhile, the entire global market for AI services generated only around $25 billion last year.

That means for every $10 these companies are pouring into AI infrastructure, they're getting back only about 25 cents in revenue.

And now the cracks are becoming harder to ignore.

Products are being shut down.

The cost of running AI keeps exploding.

Lawsuits are piling up.

And investors are starting to ask the question nobody wanted to ask two years ago.

Where's the money?

So is AI actually the next technological revolution?

Or are we watching one of the biggest tech bubbles in history start to burst?

Hi friends, I'm Maddie.

I'm a senior software engineer who previously worked Google, Internet, other bictict companies like Amazon, IBM, and Microsoft.

For two years now, the story I've been told, that you've been told, is that AI was about to replace every engineer alive.

We are not far from the world.

I think we'll be there in three to six months where AI is writing 90% of the code.

And then in 12 months, we may be in a world where AI is writing essentially all of the code.

But as of now, the money flow doesn't exactly match up to this story.

So in In this video, I'll break down why the AI hype is cracking in 2026, the spending method doesn't close, the AI products quietly getting switched off,

and the legal problem no one thought about until too late.

Let's get into it.

We're going to start with the amount of money companies are spending on AI.

The five companies building this infrastructure, Microsoft, Amazon, Alphabet, Meta, and Oracle,

are on track for roughly $725 billion of capital expenditure in 2026.

That's up 77% in a single year, and it was around $162 billion back in 2022.

Goldman Sachs predicts that the total spend will be past $7.6 trillion through 2031. To put

that into perspective, $725 billion a year is more than the entire

annual GDP of Switzerland being spent by a handful of companies on hardware

that loses most of its value in three or four years.

And so here's the problem.

That money isn't coming from spare cash anymore.

Rather, it's eating companies alive.

JP Morgan estimates AI capital expenditure went from around a third

of hyperscaler operating cash flow in 2023 to roughly 93% in 2026.

Their combined free cash flow, the actual money left over, fell to about $7 billion in the second quarter of 2026,

down from $60 billion at the end of 2025.

And analysts expect it to go negative this quarter.

Amazon is looking at deeply negative free cash flow this year.

Microsoft has told investors that its spare cash is shrinking, and it left a $60 billion buyback basically sitting there, unused.

So the richest companies on earth are borrowing hard against a payoff that hasn't arrived yet.

And the moment that made everyone realize how dangerous this could be is when DeepSeek, a Chinese startup that in early 2025 trained a competitive model for a few million dollars,

erased something like $600 billion of NVIDIA's value in one trading day, the single largest day wipeout any US stock has ever taken.

DeepSeek's success forced every investor to ask this very important question, what are we actually getting for all this spend?

And even if that money were all there, the electricity and the infrared might not be.

That part often doesn't get talked about as much because it's much less exciting than those stock charts.

But there is a hard physical limit.

Training and running these models eat staggering amounts of power.

The International Energy Agency says electricity demand from data centers jumped 17% in 2025, and the AI-specific slice grew closer to 50%.

In the US, data centers are now driving about half of all new electricity demand growth.

The grid just simply can't keep pace

because capital moves at the speed of markets and power plants move at the speed of permits and construction.

Gartner is already projecting that power shortages could constrain 40% of AI data centers by 2027.

So you have companies committing hundreds of billions of dollars to build-outs that physically cannot happen yet,

and quietly signing deals for their own nuclear reactors to get around this limitation.

That is not the behavior of an industry that has everything figured out.

In addition to the massive cost of AI, it might seem as though projects keep on failing even with the help of AI.

An MIT report found that roughly 95% of enterprise-generative AI pilots delivered no measurable return on investment.

However, there's a bit more to that number.

It does measure whether a pilot moved the bottom line within about six months, and a lot of them were in marketing, which is a hard place to show fast returns.

Also, internally built AI tools succeeded about a third of the time, while vendor-based ones succeeded about two thirds.

So most companies seem as though they're using it to solve problems that don't return much.

This is the phenomenon even in large companies.

For example, Amazon set a target of getting 80% of its developers using AI tools every week

and built an internal leaderboard called KiroRank that scored engineers on how much AI they used.

So then naturally, as you can imagine, engineers started pointing AI agents at pointless, made-up tasks purely to run up their token counts and climb the board,

a move the industry has coined token taxing.

It drove up Amazon's own compute bill for basically nothing.

So in May of 2026, they officially killed the leaderboard, and a senior VP had to send around a note saying, please don't use AI just for the sake of AI.

Also, a recent survey of more than 2,500 enterprises found

that around 74% of companies that put a live AI customer service agent into production had to roll it back.

So while the demo looks incredible, productionization is harder, especially in legacy systems full of edge cases the demo never saw or trained on.

The reason why these AI agents are getting switched off isn't only because they're wrong.

It's because that when they're wrong, the company can get into liability and big legal trouble.

Late last year, the Australian government paid Deloitte around $440,000 for a report on a welfare system.

A researcher started reading it and noticed something off.

The report cited academic papers that just didn't exist and quoted a federal court judge saying something that the judge never said.

It turns out that Deloitte had actually been using Gen.ai to help write it, and the AI had just hallucinated its sources.

Deloitte ended up issuing a corrected version and refunded part of the fee.

The lesson here is pretty stark.

If your AI puts something in front of a client, you are liable for what it outputs.

You don't get to blame it on the software.

Now, stack that on top of hallucinations would turn out to be really stubborn.

One 2026 benchmark that tested 37 different models clocked made-up answer rates anywhere from 15 to 52 percent.

And oddly, several of the newer reasoning models actually did it more often than the older ones, not less.

On legal questions in particular, Stanford put the range of hallucination at somewhere between 58 percent and 88 percent.

So put yourself in a Fortune 500 boardroom.

You've got a tool that that will confidently make things up some meaningful fraction of the time, a core precedent saying you own every word it says,

and insurers who've started writing outright AI exclusions into their policies because they can't price the risk.

What do you do?

You quietly kill a lot of those customer-facing agents.

So we have now a grim picture of the AI ecosystem.

The spending is unsustainable, the pilots mostly fail, and the agents are getting retracted.

So what should you do with this information if you're building a career?

There are three things.

Number one, don't confuse the bubble with the technology.

For example, the dot-com crash didn't mean that the internet was fake, but it meant the money had run way ahead of the value

100

and the correction wiped out everyone with no real business underneath.

101

The same sort of principle can be applied here.

102

You can think about it like this.

103

Would a tool still be worth using if AI got 10 times more expensive to run?

104

Real models with real products underneath will survive that.

105

Number two, become the person who owns the output, not the person who produces it.

106

Every failure we talked about in this video, for example, the dead pilots, the agents getting rolled back, Deloitte handing the government a faulty report, is caused

107

because of the lack of type of this person someone who can catch what the AI got wrong

108

and take responsibility for it.

109

That role is growing while the just generate code role shrinks.

110

So you should use AI every day, ship fast with it, and stay accountable for what it produces.

111

And build robust eval and verification systems.

112

And number three, last but not least, go where the deflation sends the money.

113

So when a bubble sorts itself out, the money that was pumped into it doesn't just disappear, but instead it gets deflected toward products that can actually prove their worth.

114

AI security is a good example.

115

Every one of these systems getting shipped is an attack surface almost nobody can defend yet.

116

Same with the people who can integrate AI into real systems, evaluate it, and make it reliable instead of just impressive on stage in a controlled environment.

117

So you should specialize into the part that needs human judgment.

118

So in conclusion, here's the real story of whether AI is all hype or not in 2006.

119

Yes, AI stocks are getting hammered.

120

Yes, 95% of those enterprise pilots went nowhere.

121

And yes, the AI replaces everyone mantra is getting walked back in boardrooms across the industry.

122

But none of that means AI itself is finished.

123

The internet survived the crash and came back bigger.

124

So AI will be fine.

125

The only question here is whether you'll come out of this as someone who chased the bubble

126

or if you'll become someone who understood what was actually valuable underneath it.

127

And that's all I have for you in this video.

128

If this perspective was helpful, hit that like button, hype the video, and subscribe for more videos on software engineering, AI, and tech.

앱 다운로드

유창하게 말하는 데 필요한 모든 것

AI 발음 평가문장마다 점수 확인

IPA 발음 연습모든 소리 마스터

단어 학습나만의 단어장 만들기

단어 게임놀면서 외우기

QR 스캔

Download on theApp Store Get it onGoogle Play

같은 채널

How I use AI as a senior software engineer and 5 principles of AI explained in 10 minutes

Daily ConversationWorking Adults

FOR YOU

이 레슨을 통해 배울 것

이 비디오는 AI 산업의 현황과 과열된 기대에 대한 이야기를 담고 있습니다. 빠르지만 명료한 발음, 기술 용어와 일상 표현의 혼합, 논리적인 흐름이 특징입니다. 영어 쉐도잉 연습에 좋은 자료로, 발음 교정과 동시에 복잡한 주제를 다루는 표현을 익힐 수 있습니다. 비디오에서 사용된 단어와 문장 구조를 분석하며, 실제 대화에서의 활용법을 연습해 보세요.

핵심 어휘 및 표현

backfire: (계획 등이) 역효과를 낸다. "The AI take over has completely backfired"에서 사용되었습니다.
capital expenditure: 자본 지출. 기업이 장비나 시설에 투자하는 비용을 말합니다.
free cash flow: 자유 현금 흐름. 기업이 운영 활동에서 발생한 현금 중 투자와 재무 활동을 제외한 잔액입니다.
hype: 과장된 광고나 기대. "the AI hype is cracking"에서 AI에 대한 과열된 기대가 풀리고 있다는 의미입니다.
payoff: 보상, 수익. "a payoff that hasn't arrived yet"은 아직 오지 않은 수익을 뜻합니다.

쉐도잉 연습 팁

이 비디오는 말하는 속도가 빠르고 정보량이 많아 shadow speech 연습에 적합합니다. 먼저 전체 비디오를 여러 번 듣고 내용을 이해한 후, 5~10초 단위로 나누어 쉐도잉을 시작하세요. 발음 뿐만 아니라 강세와 억양도 주의깊게 따라하세요. 특히 "the numbers don't seem to add up"나 "the cracks are becoming harder to ignore"와 같은 구문은 자연스러운 흐름을 연습하는 데 좋습니다. 영상 영어 공부를 할 때는 화자의 표정과 제스처도 관찰하며, 의미 전달에 중요한 역할을 하는 비언어적 요소도 학습하세요. 매일 10~15분씩 연습하면 발음 교정과 동시에 듣기 능력도 크게 향상될 것입니다.

이 영상의 문법

화자가 가장 많이 쓰는 문형을 영상 속 실제 표현과 함께 정리했습니다.

문형	영상 속 표현
수동태 be + 과거분사 — 누가 하는지보다 무슨 일이 일어나는지에 초점	is expected · being shut · I've been told
현재완료 have/has + 과거분사 — 과거의 일이 지금도 관련이 있을 때	I've been told · has told · hasn't arrived
관계절 who / which + 절 — 사람이나 사물에 대한 추가 정보	marketing, which is · person who owns · someone who can

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법

쉐도잉(Shadowing)은 원래 전문 통역사 훈련을 위해 개발된 언어 학습 기법으로, 다언어 학자인 Dr. Alexander Arguelles에 의해 대중화된 방법입니다. 핵심 원리는 간단하지만 매우 강력합니다: 원어민의 영어를 들으면서 1~2초의 짧은 지연으로 즉시 소리 내어 따라 말하는 것——마치 '그림자(shadow)'처럼 화자를 따라가는 것입니다. 문법 공부나 수동적인 청취와 달리, 쉐도잉은 뇌와 입 근육이 동시에 실시간으로 영어를 처리하고 재현하도록 훈련합니다. 연구에 따르면 이 방법은 발음 정확도, 억양, 리듬, 연음, 청취력, 말하기 유창성을 크게 향상시킵니다. IELTS 스피킹 준비와 자연스러운 영어 소통을 원하는 분들에게 특히 효과적입니다.

섀도잉 방법: 단계별 전체 가이드 읽기 →

쉐도잉 연습: The AI Take Over Has Completely Backfired (Here's the Proof) - 영상으로 영어 말하기 배우기

연습 모드 선택

앱 다운로드

같은 채널

How I use AI as a senior software engineer and 5 principles of AI explained in 10 minutes

추천 영상

The Limitless AI Lie. The Bubble Is Slowly BURSTING.

Something Feels Off About 2026...

Stanford CS Professor: AI Can Code. That’s Why You Should Learn Programming | Chris Piech

Elon Musk Explains How the AI Bubble Will Burst.

Meta’s Llama 4 is mindblowing… but did it cheat?

The AI Bubble… We Need to Talk

The Alignment Problem Explained: Crash Course Futures of AI #4

Leading AI CEOs Warn UN Security Council of Existential Risks | Sam Altman On AI Controls | N18G

How Smart Organizations Will Use AI: Jevons Paradox & Workforce Impact

What 50% Unemployment Will Actually Do to the Economy

Which Commercial property sectors will benefit from AI? | The Market in Minutes

How AI Became More Expensive Than The Workers It Replaced

이 레슨을 통해 배울 것

핵심 어휘 및 표현

쉐도잉 연습 팁

이 영상의 문법

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법