シャドーイング練習: it’s boring, but it’s what bursts the ai bubble - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

各モードは全画面の練習ルームで開きます。

252 文

文が短すぎたり長すぎる場合は、Editをタップして調整してください。

The single biggest threat to America isn't the national debt, it's not the Strait of Hormuz, and it's not even the AI bubble itself.

It's energy.

But not energy in the way you're probably thinking.

It's not blackouts or your monthly utility bill.

It's that America has pushed all its chips to the center of the table and bet the entire economy on one wager, that AI arrives on schedule.

And energy is the thing that decides the schedule.

Because right now, AI feels invincible.

The valuations keep growing, the spending isn't slowing down, and the stock market keeps pumping.

But if something's going to pop the AI bubble, it's energy.

And by the end of this video, I can prove it to you.

But first, the size of the wager.

Because betting the economy isn't an exaggeration.

In 2024, the biggest tech companies spent a little over $200 billion building AI infrastructure.

In 2025, that number hit almost $400 billion.

This year, their own estimates say more than $700 billion.

So the spending isn't just growing, it's nearly doubling every year, and it's that spending that's carrying the economy.

The level of concentration we're seeing in these tech stocks isn't subtle, it's practically screaming, it's a bubble.

Last quarter AI accounted for more than half of America's real GDP growth.

George Saravellos, head of FX research at Deutsche Bank, wrote in a message to clients, AI machines appear to be saving the US economy right now.

In the absence of tech-related spending, the US would be close to, or in, a recession this year.

But you probably already know all this.

You've likely heard AI is a bubble a million times by now.

And here's the thing about all those bubble takes.

Every single one of them is arguing about the wrong question.

They're arguing about whether AI is here to stay, whether it's truly going to revolutionize every industry, and whether a chatbot that can't count the letters in strawberry is maybe about to take your job.

But that was never the right question.

The real bet isn't whether AI works, it's actually about whether we can even physically deliver it on time.

Whether the AI future can arrive fast enough to justify the insane prices we're paying for it today.

Because right now, money is disappearing into AI faster than Thanos can snap his fingers.

It's the fastest and largest build-out in history, yet nobody's actually making any money on it.

Which is fine, for now.

Investors will happily fund losses as long as they think a payoff is on its way.

But eventually, the investors, the companies, and the lenders, funding all of this, need that payoff to actually show up.

Eventually the capital has to come home with a return.

That's how the AI bubble pops.

It's not when it fails, but when it's late.

When the capital starts to get sick of waiting around and decides to go somewhere else.

So what decides whether AI shows up on time?

And what determines whether these trillions of dollars from big tech will ever see a return on investment?

Well, before I get into that, let's take a quick pause.

As you've probably heard this a thousand times by now, learn AI or you'll get left behind, but nobody mentions the hard part.

ChatGPT is good at one thing, Claude at another, Gemini something else.

So you end up subscribed to all of them.

At roughly $20 a month each, that's $60 a month total, or $720 a year, all just to try and keep up.

Which brings us to the sponsor of today's video, ChatLLM by Abacus AI.

They've solved that problem.

Chat LLM by Abacus AI puts every top model in one place.

GPT 5.6, Claude Fable 5, Gemini 3.1, even Grok and DeepSeek, and when a new model drops,

it's instantly available, which means no waiting around and no new subscriptions.

You can pick which model to use yourself, or let the Route LLM feature read your prompt and automatically pick the best model for you.

And Chat LLM by Abacus AI goes past just chat.

It can generate images, video, and even build full apps and websites, with hosting included, no code needed.

The price for all this is just $10 a month, less than any single subscription it replaces.

If you're interested, you can check out Chat LLM by Abacus AI using the link below, or in the pinned comment.

Thank you to Abacus AI for sponsoring this video.

Now back to the single wager our entire economy's riding on.

This entire bet is trapped inside a race between two clocks.

Clock one decides how fast the AI future can actually arrive, how quickly we can build it, Plug it in and turn it on.

Clock 2 decides how long we can afford to wait for it.

We'll get to that later because it's the scarier of the two.

So, let's start with the first one.

Whenever a technological breakthrough like AI shows up, Wall Street loves to hand out the same Mark Twain quote as advice.

Don't bet on the miners.

Invest in the picks and shovels.

The phrase is more overused than the man in the arena

speech in a one-bedroom San Francisco apartment with four founders vibe coding an AI app.

What Wall Street's saying is, don't try to pick the winning chatbot.

Just buy the stuff underneath the boom, the data centers, the power, the infrastructure.

But here's what nobody mentions when they hand you that advice.

America's running out of picks and shovels, because you can't just build a data center.

A data center doesn't get electricity by plugging into the grid and asking nicely.

It has to join something called an interconnection queue.

That's the waitlist to get access to the power grid.

Think of it as the DMV, but for electricity.

Except instead of measuring the weight in bad attitudes and hours, the interconnection queue measures it in regulatory headaches and years.

In America's largest power grid, that weight has stretched from under two years in 2008 to over 8 years today.

And the waitlist is only where the problems start, because AI data centers don't use electricity the way an office tower does.

Busy during the day, quiet at night, asleep on weekends.

Data centers run full blast around the clock every day of the year.

They already eat more than 4% of America's electricity.

And by 2035, this number is expected to reach a fifth of the country's electricity consumption.

And the main workhorse feeding this entire build-out is this alien-looking device, the gas turbine, which is where the problems continue.

It's the machine behind roughly 40% of America's electricity, and they're among the most complex pieces of machinery on Earth.

Only a handful of companies can actually build one.

Three manufacturers control over three-quarters of the global market.

One of them is GE Vernova.

Its order backlog has ballooned from roughly 80 gigawatts to 116, and its own CEO says the company will be largely sold out through the end of 2028.

And in 2024 alone, developers ordered nearly three times more turbine capacity than every factory on Earth could physically produce in a single year.

Which means it doesn't matter how many zeros Big Tech puts on the check, the manufacturers are structurally constrained.

But it's not just the gas turbines either.

Transformers, which are the machines that step power down from the grid, now take over two and a half years to get with an estimated 30% supply shortage.

And for new transmission lines, they take 6.5 years on average to complete due to permitting delays, backlogs, and a mountain of other BS.

So it's not just a single waitlist that's backed up.

It's the entire American energy supply chain.

We're seeing tech move at Netflix speeds, and the infrastructure underneath it moving at the speed of blockbuster.

Which is why the companies trapped between these two speeds are starting to act desperate.

Two developers reportedly paid $25 million, for a turbine, just to reserve a place in line for one.

Others are rebuilding retired jet engines to makeshift the power they need.

We're seeing grown, profitable companies bolting old aircraft engines to the ground because they're desperate for power.

100

So the obvious fix is the one you're already thinking.

101

Just build more.

102

Big tech is great at throwing money at problems.

103

Why not throw money at the source until the clock speeds up?

104

Well, we already ran that exact experiment.

105

And money lost.

106

Remember Remember that 8 year wait just to plug into the grid?

107

It belongs to a power grid called PJM Interconnection.

108

PJM is the largest grid operator in America, serving 67 million people across 13 states from Chicago to Washington DC, roughly 1 in 5 Americans.

109

If you live anywhere in that footprint, your lights are on because of decisions they make,

110

and PJM's biggest decision of the year happens at an event nobody watches, the capacity auction.

111

Here's how it works.

112

PJM doesn't just buy electricity, it buys guarantees.

113

Once a year, PJM holds an auction to pay power plants to promise they'll be ready with electricity on the worst days of the year.

114

It's essentially a billion dollar pinky promise that there's going to be electricity when there's a January freeze, or an August heat wave.

115

You can think of it like insurance.

116

They pay a premium so the grid doesn't fail when it's needed most.

117

For most of this century, that insurance was dirt cheap, because American electricity demand flatlined for 20 years.

118

It made the capacity auction the single most boring event in American finance.

119

In the auction covering 2024, a promise of power cleared at about $29 per megawatt day.

120

Then, the data centers arrived.

121

Not one at a time, by the dozen.

122

Buildings

123

that each swallow as much electricity as a small city walking into a system designed for a world where demand barely grew.

124

And the data centers don't want power eventually.

125

They want it now, guaranteed, forever.

126

The next auction cleared at just under $270, a more than 830% spike.

127

The regional bill went from $2.2 billion to $14.7.

128

The same promise, seven times the price, the biggest leap since the capacity market was invented in 2007.

129

That got attention.

130

Governors across the region leaned on PJM, federal regulators signed off, and the auction got a ceiling, a legal maximum on what the grid was allowed to pay.

131

The very next auction hit that ceiling.

132

So did the one after it, and the one after that.

133

What looked like a spike turned out to be the new floor.

134

Then came December 2025, and something happened that had never happened in PJM's history.

135

The auction came up short.

136

Each auction is built around a target reserve margin.

137

It's the extra safety cushion of power supply required above expected

138

peak demand to guarantee the grid can handle those worst days of the year.

139

It's basically the backups in case of an emergency.

140

For the first time in history, PJM went shopping with more money than ever before, and paid the maximum price allowed under federal law,

141

but there still wasn't enough power supply to go around.

142

And that's the physical reality of the grid.

143

All that money can't summon a gas turbine, a transformer, or a transmission line out of thin air when the world's manufacturers are booked years out in advance.

144

Which means the grid failed to buy what it needed and then paid full price for what it got.

145

And the bill for all this doesn't just get sent to big tech, it gets sent to you.

146

It flows straight into your electric bills.

147

The average household in Maryland saw their electric bill rise by $21 a month, or nearly $250 per year.

148

Which may not sound like a lot at first, until you remember that's just the average, for one household, in one state.

149

When you remember that PJM covers 13 states and 67 million people, it's a lot.

150

PJM pays an independent watchdog called Monitoring Analytics.

151

They ran the numbers on the December auction and found that of the $16.4 billion the auction will cost,

152

roughly $6.5 billion can be traced directly to data centers.

153

Forty cents of every capacity dollar traced to data centers.

154

Across the last three auctions, the number is more than $23 billion.

155

So that's clock one, the power clock.

156

The infrastructure can't keep up with the demand, money can't speed it up, and there's no clear path to fixing it.

157

These companies just have to wait.

158

Which sounds survivable.

159

Annoying and expensive, but survivable.

160

Because late is better than never.

161

Unless the thing doing the waiting has an expiration date.

162

And spoiler, it does.

163

Which brings us to clock two.

164

The chip clock.

165

The one that determines how long we can afford to wait.

166

Because those data centers are really just massive warehouses for a single product,

167

semiconductor chips, specialized processors that train AI and the most expensive piece of the entire bet.

168

Of the more than $700 billion a year, the biggest chunk is spent on the chips and computing hardware.

169

And there are two problems with these chips.

170

The first is, they don't age well.

171

Think of that one grandma of yours who hasn't worn sunscreen since the 70s.

172

A cutting edge AI chip stays cutting edge for somewhere between two to six years, depending on who you ask.

173

Newer, faster chips arrive, and the old ones slide toward worthless.

174

But the infrastructure those chips are living on lasts for decades.

175

So the chips age like a smartphone, and the infrastructure ages like a bridge.

176

That mismatch has a price tag.

177

If a data center scales back, or its chips go obsolete after a few years, the multi-billion dollar grid equipment built to serve it stays on the utilities' books for decades,

178

and a chunk of that cost gets shifted to the utilities' customers, meaning you.

179

And then there's the second problem, the bigger one.

180

Right now, tech companies are reporting record earnings.

181

On the surface, you could maybe even say, what AI bubble?

182

But Michael Burry, the investor from the movie The Big Short, thinks those record earnings are an illusion.

183

He believes big tech's cooking the books, and like most things in finance, he believes it comes down to sneaky accounting.

184

When a company buys a machine, it spreads the cost across the years the machine stays useful, called the useful life.

185

That schedule is called depreciation.

186

Stretch the useful life out, and your annual depreciation drops, making profits look bigger today.

187

Shrink the years, and your annual depreciation expense rises, lowering your reported profits.

188

Which means one boring accounting guess, how long does a chip stay useful, quietly determines billions in big tech earnings.

189

And the problems arise when nobody agrees on the answer.

190

The tech giants started extending the official useful life of their servers from a 3 or four-year standard to six years.

191

Burry says the real number is much lower.

192

If he's right, the industry is understating depreciation by about $176 billion between 2026 and 2028.

193

By Burry's math, this would overstate Oracle's earnings by nearly 27% by 2028, and Meta's by almost 21%.

194

NVIDIA fired back immediately at this.

195

They said that their customers get four to six years out of these chips just fine.

196

What we're watching is a classic kindergarten game of he said, she said.

197

That's also what makes this interesting, because whether you side with Burry or NVIDIA doesn't actually matter, because the tech companies themselves can't agree on the answer.

198

In January 2024, Amazon decided its servers would last six years instead of five.

199

That single decision reduced its depreciation expense by $3.2 billion and added $2.5 billion to its profit for the year.

200

Twelve months later, Amazon changed its mind.

201

They walked the estimate from six years back down to five for subsets of their servers and networking equipment.

202

The reason Amazon gave in the filing was the increased pace of technology development, particularly in artificial intelligence.

203

And while Amazon was walking some things back, Meta went the opposite direction.

204

Meta extended the life of its servers and network assets to five and a half years.

205

That single change reduced its 2025 depreciation costs by $2.9 billion, and added roughly a dollar to Meta's earnings per share.

206

So, same technology, same time period, two accounting departments, looking at the same silicon, and reaching opposite conclusions.

207

And that's the race we have.

208

There are two clocks ticking.

209

One says new power infrastructure can take 5 to 10 years to build.

210

The other says today's AI chips may only be economically cutting edge for 3 to 6 years, depending on who you ask, of course.

211

If those clocks fall out of sync, billions of dollars of infrastructure could arrive for a very different AI market than the one companies are planning for today,

212

and the race is already being timed.

213

Goldman Sachs looked at how often data centers actually open on schedule.

214

Historically, only about 72% of projects due within the next year showed up on time.

215

When you adjust for today's supply chain and labor problems, Goldman expects about 60% of next year's promised capacity to arrive when promised look two years out,

216

and it drops to 50%.

217

Flip a coin, that's the current odds the AI future arrives on schedule.

218

And Bloomberg's energy research arm found something even uglier.

219

They ran two forecasts side by side.

220

The first asks a simple question.

221

Based on how many AI chips are actually being sold, how much data center capacity should America be building?

222

The second asks a harder one.

223

How much can America actually power?

224

They found that by 2033, the gap will be 63 gigawatts apart.

225

For context, every operational data center in America adds up to a little over 40 gigawatts.

226

So, the gap Bloomberg found isn't a small rounding error.

227

To deliver the future these companies have already paid for, America would have to find a way to power a gap half as large as every data center the country's already built.

228

That took decades to build, on a grid that now has an 8-year waitlist.

229

The math doesn't just look bad, the math looks impossible.

230

Which brings us to the age-old question.

231

How did we get here?

232

How does the entire energy supply chain get this backed up?

233

Well, here's the uncomfortable answer.

234

It's not because AI is failing, it's because it isn't.

235

The demand is painfully real.

236

Everyone wants in.

237

Data centers drove roughly half of America's electricity demand growth last year.

238

The line for power is measured in years because everyone showed up at once.

239

If you've seen my previous videos, it's the flood stage of the capital cycle.

240

Success is what broke the grid.

241

And the chips?

242

How does the most expensive hardware on Earth expire so fast?

243

Again, not because AI stalled, because it didn't.

244

Every new generation of chips buries the one before it.

245

The technology is moving so fast that its own products can't hold their value.

246

Success is what kills the chips.

247

So the better AI does, the more demand floods the grid and the longer clock one takes.

248

The faster AI improves, the quicker the hardware goes stale and the shorter clock two gets.

249

Success winds both clocks against each other.

250

Now if you'd rather get your financial news from people yelling about earnings beats, No hard feelings.

251

But if you want the deeper stuff, the stuff buried in government reports and depreciation schedules, hit subscribe.

252

Because it's those boring documents where the real stories actually live.

アプリをダウンロード

流暢に話すために必要なすべて

AI発音採点文ごとに採点

IPA練習すべての音をマスター

単語学習自分の単語帳を作る

単語ゲーム遊びながら覚える

QRをスキャン

Download on theApp Store Get it onGoogle Play

同じチャンネル

FOR YOU

このレッスンの語彙とスピーキングのポイント

このB2レベルのスピーキングレッスンは、動画「it’s boring, but it’s what bursts the ai bubble」を教材にしています。繰り返し出てくる語は次のとおりです：chip, center, grid, billion, clock。この動画には、シャドーイング用の文が252文、単語が3242語あります。音声の長さは17:51です。話す速さは速く、1分あたり約182語です。音のつながりや弱く発音される音が多くなります。英語の頻出3,000語に含まれる単語は80%だけなので、語彙は難しめです。

この動画の重要語彙

動画の中で特に難しい単語15語を、発音と意味つきで紹介します。

単語	発音	意味
grid 名詞	/ɡɹɪd/	格子, グリッド
depreciation 名詞	/dɪˌpɹiːʃɪˈeɪʃən/	減価償却
bubble 名詞	/ˈbʌbəl/	泡
turbine 名詞	/ˈtɜː.baɪn/	タービン
gigawatt 名詞	/ˈɡɪɡəˌwɒt/	ギガワット
bore 動詞	/boɹ/	いらいらさせる
investor 名詞	/ɪnˈvɛs.tə(ɹ)/	投資家, 出資者
mathematics 名詞	/mæθ(.ə)ˈmæt.ɪks/	数学
chatbot 名詞	/ˈtʃætbɑt/	チャットボット
shovel 名詞	/ˈʃʌv.əl/	シャベル, スコップ
chunk 名詞	/t͡ʃʌŋk/	塊
queue 名詞	/kju/	列, 待ち行列
spike 名詞	/spaɪk/	針
plug 名詞	/plʌɡ/	プラグ
ceiling 名詞	/ˈsilɪŋ/	天井

動画に出てくる句動詞

単語	発音	意味
show up 動詞		現れる
check out 動詞		試験する, 検査する
come up 動詞		上がる, 登る
hand out 動詞		配る, 配布する
leave behind 動詞		置き忘れる
slow down 動詞		遅らせる
sold out 形容詞		売り切れ
speed up 動詞	/spiːdˈʌp/	スピードを出す

この動画の文法

話し手がよく使っている文型を、動画の実際の表現とともに紹介します。

文型	動画での表現
受動態 be + 過去分詞 — 誰がするかより、何が起きるかに焦点を当てる	is trapped · is expected · was invented
現在完了形 have/has + 過去分詞 — 過去の出来事が今も関係している	has pushed · You've likely heard · you've probably heard
関係詞節 who / which + 節 — 人や物について情報を加える	available, which means · turbine, which is · Transformers, which are

注意したい発音

話し手はisn't, can't, doesn'tなど、短縮形や弱形を49回使っています。聞こえたとおりの短い形で発音しましょう。

「th」の音: mathematics /mæθ(.ə)ˈmæt.ɪks/, underneath /ˌʌndɚˈniθ/, breakthrough /ˈbɹeɪk.θɹuː/
「sh」と「zh」の音: depreciation /dɪˌpɹiːʃɪˈeɪʃən/, interconnection /ˌɪn.tɚ.kəˈnɛk.ʃən/, shovel /ˈʃʌv.əl/, specialize /ˈspɛʃəˌlaɪz/, recession /ɹɪˈsɛʃn̩/
長い単語（アクセントの位置に注意）: depreciation /dɪˌpɹiːʃɪˈeɪʃən/, interconnection /ˌɪn.tɚ.kəˈnɛk.ʃən/, mathematics /mæθ(.ə)ˈmæt.ɪks/, developer /dɪˈvɛləpɚ/, machinery /məˈʃiːnəɹi/

日本語話者が苦手な音：

/r/ と /l/ の区別 — /l/ は舌先を歯茎につけ、/r/ はどこにもつけない: profitable /ˈpɹɑ.fɪ.tə.bl̩/, historically /hɪˈstɔɹɪkli/, reportedly /ɹɪˈpɔːtɪdli/, practically /ˈpɹæk.tɪ.k(ə.)li/
/v/ — /b/ にならないように、上の歯を下唇に当てる: investor /ɪnˈvɛs.tə(ɹ)/, shovel /ˈʃʌv.əl/, developer /dɪˈvɛləpɚ/, invent /ɪnˈvɛnt/, vibe /vaɪb/
/f/ — 「フ」ではなく、上の歯と下唇で出す: profitable /ˈpɹɑ.fɪ.tə.bl̩/, smartphone /ˈsmɑɹtfoʊn/, fixing /ˈfɪksɪŋ/, handful /ˈhæn(d)fʊl/, freeze /ˈfɹiːz/

この動画での練習方法

まず声を出さずに動画を最後まで聞き、知らない単語をメモします。
まず0.75倍速で一文ずつシャドーイングし、慣れてきたら通常の速度に戻します。
自分の声を録音して元の音声と比べます。grid, depreciation, bubbleなどの単語に特に注意しましょう。

シャドーイングとは？英語上達に効果的な理由

シャドーイング（Shadowing）は、もともとプロの通訳者養成プログラムで開発された言語学習法で、多言語習得者として知られるDr. Alexander Arguelles によって広く普及されました。方法はシンプルですが非常に効果的：ネイティブスピーカーの英語を聞きながら、1〜2秒の遅延で声に出してすぐに繰り返す——まるで「影（shadow）」のように話者を追いかけます。文法ドリルや受動的なリスニングと異なり、シャドーイングは脳と口の筋肉が同時にリアルタイムで英語を処理・再現することを強制します。研究により、発音精度、抑揚、リズム、連音、リスニング力、そして会話の流暢さが大幅に向上することが確認されています。IELTSスピーキング対策や自然な英語コミュニケーションを目指す方に特におすすめです。

シャドーイングのやり方: ステップ別の完全ガイドを読む →

シャドーイング練習: it’s boring, but it’s what bursts the ai bubble - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

アプリをダウンロード

The Stock Market Is Being Used as a Dumping Ground

To Save the Economy... We’re About to Break It

The $2.5 Trillion Credit Wave Nobody Sees Coming

The World's Safest Market Is Breaking

The Recession's Already Here… Open Your Eyes

The New American Housing Crisis Has Landed

The Stock Market is BROKEN (just accept it)

Something Feels Off About 2026...

The SpaceX IPO... It's Worse Than You Think

The AI Bubble Survives on $1.65 Trillion in Hidden Debt

Bitcoin Is a Brilliant Scam and I Can Prove It (Again)

Private Credit: The Secret Industry Taking Over Wall Street

Every Type of Investor Explained in 6 Minutes

Why The Dumbest Investors Win (it's not why you think)

The Inflation Problem... It's Worse Than You Think

Crypto Is a Brilliant Scam and I Can Prove It

The 1973 Stock Market Crash that Nobody Remembers

The (Inevitable) United States Debt Crisis

The AI Bubble… We Need to Talk

Why the Oil Crisis Is (Secretly) Good...

No, Really — Is A.I. a Bubble? | The Ezra Klein Show

Why the Arab World never united

歐巴馬談 AI：喊放慢的公司「是真的在擔心自己的產品」【AI 大人說】

How America Got So Good At Buying Sh*t

The Limitless AI Lie. The Bubble Is Slowly BURSTING.

I MADE $15B WITH CHAT GPT

How AI is helping brands identify consumer trends faster

Is the AI Boom Just Another Bubble?

The Only SaaS Moats Left in 2026

Significant slowdown in AI payoff could tip economy into a recession, says Apollo Global's Slok

If You Don’t Have One of These Jobs by 2030… You’re Screwed