Home
Connexion
S'inscrire
Pratique d'écoute
Pratique d'écoute
/
Video
/
The Infographics Show
/
What China Understands About AI That the US Doesn’t
What China Understands About AI That the US Doesn’t
Choisir le mode d'apprentissage:
Voir les sous-titres
Choisir le mot
Réécrire le mot
Highlight:
3000 Oxford Words
4000 IELTS Words
5000 Oxford Words
3000 Common Words
1000 TOEIC Words
5000 TOEFL Words
Sous-titres (253)
0:00
The AI revolution is supposed to be happening in the cloud, a battle fought with code and software.
0:05
It's not. Right now, Microsoft
0:07
and other tech giants are quietly locking up America’s nuclear energy capacity to train the
0:12
next generation of AI models. They’re pulling massive amounts of power off the public grid.
0:16
And as demand explodes, regular consumers are left competing for what’s left.
0:20
The result? A private AI energy race is emerging behind
0:24
the scenes…. and you're picking up the bill. Chapter 1: The $500 AI Tax
0:30
PJM runs the power grid for 67 million Americans across 13 states. And once a year,
0:36
it holds a massive auction to lock in electricity… 3 years before anyone uses it.
0:41
In June 2025, that auction cleared almost $270 per megawatt-day. The year before? Just
0:48
$29. In only 12 months, prices exploded 833%. That single auction sent total household costs surging
0:57
from $2.2 billion to $14.7 billion. And then it got worse.
1:03
The next auction, covering June 2026 onward, smashed straight into the market price cap
1:09
of almost $329 per megawatt-day. And without that cap, the real price would’ve hit $389.
1:17
The cap has already saved consumers $13 billion. In December 2025, internal filings showed PJM’s
1:24
next auction prices hit an even higher cap, but that wasn’t the problem. PJM was short
1:29
of its own safety target. Even after prices exploded, the grid operator still couldn’t
1:34
buy enough power to guarantee its reserve margin. All across the country, prices are beginning to
1:39
rise. The average family inside the PJM footprint faces an estimated $70 more per month by 2028.
1:46
The total bill across the next 8 years lands somewhere between $100 billion and $163 billion.
1:53
The reason for all this is simple. Demand.
1:56
Data centers alone drove roughly 63% of the entire price surge. The explosion in
2:01
demand from AI server farms is happening so fast that the grid is struggling to keep up.
2:06
The cheapest, most reliable power in the country is being pulled out of the public pool and sent
2:11
into corporate data halls. Ordinary families are left covering the rising cost of keeping
2:16
the rest of the system stable. And it’s already on your bill.
2:20
Chapter 2. The Great Cannibalization Most Americans still believe their
2:24
country is leading a clean energy boom. But the real story is more complicated.
2:28
In September 2024, Constellation Energy signed a 20-year power purchase agreement with Microsoft to
2:34
restart Three Mile Island Unit 1 in Pennsylvania. The plant sits just a few hundred feet from
2:39
the cooling towers of Unit 2, the reactor that suffered America’s worst nuclear accident in 1979.
2:46
Unit 1 ran for decades until it shut down in 2019. Not for safety concerns but because it stopped
2:52
making enough money to justify staying online. The plant is set to come online in 2027,
2:57
a full year ahead of the original 2028 target. The Department of Energy closed
3:01
a $1 billion loan to lower the borrowing costs, while Constellation itself put $1.6
3:07
billion into the restart. The plant's operating license was extended to at least 2054.
3:12
Microsoft gets the reactor’s full output for the next two decades, enough electricity to power
3:18
roughly 800,000 average American homes for a year. The power itself still flows into the shared
3:24
grid. But the clean-energy credit is claimed by Microsoft. So while nothing
3:28
physically changes in how electrons move, the output is effectively reserved on paper for
3:33
AI workloads all the way through 2047. The deal also delivers $3.6 billion in
3:39
state and federal taxes. 3,400 jobs will be created across central Pennsylvania and $16
3:45
billion added to the state's economy. Washington has even called it the “poster child” of its
3:50
energy and AI strategy. Politicians point to the investment and the jobs as proof..
3:54
What they ignore is the community deprived of power.
3:58
Shared national power is becoming private reserves, and the households left behind end up
4:03
paying more just to keep the grid running. Chapter 3: The Five Nines Reality
4:08
People assume the answer to AI's power hunger is simply more renewables. Build enough solar in
4:13
Nevada, add enough wind in Texas, throw in some big batteries, and the problem solves itself.
4:19
Except it doesn’t. AI data centers run on a reliability
4:22
standard called Five Nines. 99.999% uptime, or just about five minutes of downtime per year.
4:30
That kind of precision exists for a reason. A single frontier model training run can
4:34
cost tens of millions of dollars and run for weeks. One brief power flicker can
4:39
corrupt the entire run, forcing engineers to restart from the last saved checkpoint.
4:44
And that’s where the grid problem becomes unavoidable.
4:47
Solar produces nothing at night, and wind can disappear for days at a time.
4:51
Bridging those gaps to meet Five Nines reliability would require battery storage
4:55
at a scale no country has ever built. The cooling problem is just as brutal.
5:00
A single rack of NVIDIA servers pulls about 8 to 10 times more power compared to a standard server
5:05
rack just a few years ago. All of this creates immense heat and that’s got to go somewhere.
5:11
New AI data centers now run on direct liquid cooling, pumping water across the chips at flow
5:16
rates that would supply small American towns. That’s why hyperscalers are
5:20
picking nuclear reactors. Nothing else delivers carbon-free,
5:24
around-the-clock power at the size they need. The patchy stuff gets left for the
5:29
public grid to absorb, along with every headache that comes with balancing it.
5:33
A Reuters review of letters filed with PJM tells the rest of the story. While the cleanest power
5:38
gets locked into corporate contracts, the dirtiest plants are getting a second life. About 23 coal,
5:43
gas, and oil plants in the PJM region were scheduled to retire in 2025 or just after.
5:49
Since January 2025, 13 have had their shutdowns delayed or canceled outright. Brandon Shores,
5:54
a coal plant in Maryland that was the state's top climate polluter in 2022, will now run
6:00
until May 2029. It's owned by the same Talen Energy that sells nuclear power to Amazon.
6:06
The public is being told that green energy is the future. But the physical system is
6:10
already splitting into two tiers: firm nuclear capacity secured by corporations, and intermittent
6:16
renewables carrying the rest of the load. And even with firm power being locked up,
6:20
the sheer scale of new demand is now straining something more basic…
6:24
The grid itself. Chapter 4: Data Center Alley
6:28
Loudoun County, Virginia sits about 40 miles (64 km) west of Washington, D.C., a stretch
6:32
of rolling farmland has given way to suburbs and endless warehouse-scale buildings. The region now
6:38
handles roughly 25% of all North American internet traffic. Locals call it “Data Center Alley,” and
6:44
at its center is zip code 20147, the closest thing the internet has to a physical brain.
6:50
The county now holds around 200 data center buildings, with another 117 in development.
6:56
Each new project adds more pressure on the grid. Most new campus proposals request
7:00
enough electricity to run an entire county, roughly equal to 150,000 American households.
7:06
One campus. One company. Sometimes even one customer.
7:10
To carry the load, Dominion Energy is building massive new transmission
7:13
lines meant to move power into Loudoun County’s data center belt. One of the biggest proposals,
7:18
the Golden-to-Mars line, would cut a 22 mile (35 km) path through neighborhoods, schools,
7:23
and farmland. All carried on steel towers rising hundreds of feet into the skyline.
7:27
And the spending is scaling with it. Dominion’s investment plan for 2025
7:32
through 2029 has climbed to $50.1 billion. But the cost doesn’t stay local. It gets
7:38
spread across the entire regional grid, 13 states in total. So households far from Virginia will
7:44
help pay for infrastructure built to serve data centers they may never interact with.
7:49
Virginia's own Joint Legislative Audit and Review Commission ran the numbers. They said
7:53
the state would need a brand new 1.5-gigawatt natural gas plant every 2 years for 15 straight
8:00
years just to feed data center demand. These costs are buried in utility bills
8:05
and wholesale market fees that most customers never directly see. And
8:08
the pressure isn’t just coming from the wires. It’s also coming from what’s plugged into them.
8:14
Chapter 5: The Efficiency Trap NVIDIA's Blackwell platform is
8:18
the current top-of-the-line AI chip, rolling out across data centers through 2025 and 2026. At the
8:24
system level, it uses up to 25 times less energy per inference token than the previous generation.
8:30
That should solve the whole problem. It doesn't.
8:33
Global data center electricity demand is still set to more than double by 2030.
8:38
The International Energy Agency's base case puts global use at roughly 945 terawatt-hours by then,
8:45
up from 415 terawatt-hours in 2024. Slightly more than the entire annual
8:52
electricity use of Japan. By 2035, the same forecast climbs to 1,200 terawatt-hours.
8:58
In the United States alone, data centers could consume more electricity than all
9:02
heavy industry combined - steel, cement, chemicals, and aluminum - put together.
9:08
A 2024 Lawrence Berkeley National Laboratory report says US data centers made up 4.4% of
9:14
all electricity use in 2023. Within just a few years, that could nearly double - or even
9:20
triple - depending on how fast demand accelerates. In raw terms, electricity use could jump from
9:26
176 terawatt-hours to as much as 580 terawatt-hours in five years.
9:32
So how do both things happen at once? Efficiency improving, while total use still explodes?
9:38
That question isn’t new. A 19th-century English economist named
9:43
William Stanley Jevons figured it out long before anyone thought about AI. Watching steam engines
9:49
get better at burning coal, he noticed something strange. England's total coal use kept going up,
9:54
not down. Cheaper energy unlocked new uses, and the explosion of new applications buried whatever
10:01
savings the efficiency had produced. The same trap has now closed on AI.
10:06
When the cost of a single query drops from cents to fractions of a cent,
10:10
every product manager finds a new opportunity. Chatbots get embedded in customer service.
10:14
AI overlays appear on every search bar. Code helpers run quietly in the background of every
10:20
developer's day. Automated agents start handling work people used to do by hand.
10:25
Each one is cheap on its own. But multiplied across the global economy,
10:29
they eat every watt of savings and demand orders of magnitude more on top.
10:34
Cars on American roads tell the same story. Engines got much more fuel-efficient over the
10:39
past 40 years, yet total US gasoline use barely budged. Americans simply drove farther and bought
10:45
bigger trucks. Cheaper miles meant more miles. The IEA expects about half of all new data
10:51
center demand through 2035 to be met by renewables, with natural
10:55
gas covering another large chunk of the gap. That just shuffles how the energy is supplied.
11:00
It doesn’t solve the problem. And no chip breakthrough on the horizon changes
11:04
it. If anything, the crisis accelerates because technology keeps getting better, not worse.
11:10
America is trying to solve this by pushing more and more power through already stressed wires to
11:15
where the data centers are built. But China is taking a different
11:18
approach to the same problem. Chapter 6: China's East-to-West Masterplan
11:23
In February 2022, the Chinese state launched something it calls the East Data, West Computing
11:28
initiative. it’s a national restructuring of where computing actually happens. A coordinated
11:33
effort to move data centers away from dense coastal cities and into inland regions where
11:39
land is cheaper and power is more abundant. It’s being executed at a speed and scale no
11:43
Western government has attempted in our lifetime. It splits the country into 8 national computing
11:48
hubs and 10 data center clusters. The eastern clusters sit near Beijing,
11:53
Shanghai, and the Greater Bay Area, handling the fast-response work for the cities where
11:57
most of the economy lives. The western clusters are placed on purpose in the empty interior,
12:02
across Inner Mongolia, Ningxia, Gansu, and the edges of the Gobi Desert. As of late 2024,
12:08
China's National Data Bureau reported that roughly $6.1 billion had already flowed into the 8 hubs.
12:14
Why build out there? Because that's where China's huge hydro,
12:19
wind, and solar power sits unused. The western provinces produce far more electricity than
12:24
their thinly spread populations can use. The surplus had historically gone to waste because
12:29
sending it east was wasteful and expensive. The state's fix wasn't to build more wires.
12:34
It was to move the demand. Computing tasks that don’t need instant,
12:38
millisecond responses are now being pushed west along dedicated high-capacity fiber networks.
12:44
And for most workloads, it works. Latency between eastern cities and western data
12:48
hubs already meets China’s 20-millisecond target for routine computing. The result is a physical
12:54
reshaping of the internet itself. Data halls are built directly on top of the cheapest,
12:59
most abundant electricity in the country. In places like Inner Mongolia’s Ulanqab
13:04
cluster, that strategy has already attracted around 10 gigawatts of
13:07
planned data center projects from companies including 21Vianet, Huawei, and ByteDance.
13:13
New plants inside the national hubs must hit a power rating of 1.2 or lower, with some new
13:18
builds hitting 1.04, compared to a global average closer to 1.5. China's overall data center power
13:25
use is forecast to grow 19% per year between 2025 and 2030, faster than any other major economy.
13:33
If you can't efficiently move the power to the computers, you move the computers to the power.
13:38
While the United States lets corporations privatize nuclear plants one reactor at a time,
13:42
China treats energy and computing as a single state asset. The geography itself
13:46
wipes out the transmission chokepoints crippling the American system, and there
13:50
is no Loudoun County problem in China. America is breaking its power system into
13:55
privately controlled zones built around individual corporate demand centers,
13:59
rather than a shared network. China is doing the opposite. It’s reshaping geography
14:03
itself so supply and demand sit closer together. The American response to all of this is private,
14:10
expensive, and almost invisible. Chapter 7: The SMR Mirage
14:14
Through 2024 and 2025, the biggest names in American tech started pouring money
14:19
into a new kind of nuclear plant called Small Modular Reactors, or SMRs. They’re
14:25
smaller than traditional reactors, built in factories, shipped in pieces, and assembled
14:30
on site. Each unit typically produces between 50 and 300 megawatts. They can be scaled by
14:35
adding modules instead of betting everything on a single multi-billion-dollar megaproject.
14:40
Amazon led the push with a $500 million investment in X-energy, a company developing high-temperature
14:46
gas-cooled reactors. The plan is to unlock more than 5 gigawatts of new nuclear capacity by 2039,
14:53
enough to power roughly 3.8 million homes. The first project, the Cascade Advanced
14:58
Energy Facility in Washington state, is designed to stack 12 reactor modules into
15:03
three larger blocks, totaling 960 megawatts on a footprint the size of a few city blocks.
15:09
Google followed with a long-term deal with Kairos Power to bring a new generation of
15:14
nuclear plants online in the coming decade. The first project, Hermes 2 in Tennessee,
15:19
will feed electricity into the regional grid, but its purpose is effectively pre-committed
15:23
to Google’s data centers. It’s also the first time a US utility has struck a deal
15:28
tied directly to an advanced reactor design instead of a traditional plant.
15:32
Meta followed the same path, opening negotiations for multiple gigawatts of nuclear capacity.
15:37
Equinix, which runs a large share of the world’s data center infrastructure, signed
15:41
on with Oklo. Prometheus Hyperscale came next. And every single one of these reactors is being
15:47
designed to plug straight into or feed those campuses through behind-the-meter
15:52
setups. That’s when power moves from the generator straight to the customer through private lines,
15:57
and the shared grid never gets touched. Most of the tariffs and rules that apply
16:01
to grid-connected power don't apply either. These reactors are private utilities for corporate
16:07
use. The public is cut out at the blueprint stage. For most of the 20th century, American nuclear
16:12
power came through national labs, federally funded research, and rate-regulated utilities
16:18
that answered to state commissions. The benefits flowed, however imperfectly,
16:22
to ordinary families. The biggest nuclear push in 2 generations is now being paid for by three
16:28
corporations, built for their exclusive use, and engineered to stay outside the public system.
16:33
A tech company isn't just a tech company anymore. Chapter 8: The Sovereign Tech-Utility
16:39
Microsoft, Amazon, and Google have stopped being software companies that happen to run cloud
16:44
divisions on the side. Instead, they've become the largest unregulated energy buyers on the planet.
16:50
Amazon's deal at the Talen Energy Susquehanna nuclear plant in Pennsylvania shows how messy
16:55
this gets when private wires meet public rules. In March 2024, Amazon Web Services bought a data
17:00
center campus next door to the Susquehanna plant for $650 million. They plan to increase
17:06
the on-site load and it’ll be fed straight from the reactor through dedicated lines.
17:11
FERC had already approved a private power connection between Susquehanna and a nearby
17:17
data center in 2015, allowing electricity to flow directly into a single customer. When that
17:22
connection was later expanded, utilities pushed back hard. Their argument wasn’t technical. It was
17:28
about who ends up paying for the infrastructure when a single large user reshapes the system.
17:33
The federal regulator ultimately blocked the expansion, siding with concerns that
17:37
the setup would shift hundreds of millions of dollars in grid costs onto everyone else
17:42
in the region. Even after that decision, the fight didn’t end. It moved into the courts,
17:47
while the original private connection continues operating in the background.
17:51
The fight cracked the rulebook wide open. The same arrangement is being copied across the country
17:56
regardless. Each deal looks reasonable alone, but stacked together, none of them answer to voters.
18:02
The old utility system was built around everyone getting power. Utilities had
18:06
to answer to regulators and keep the grid running for entire communities,
18:10
not just their biggest customers. The new power players operate by a different rulebook. Their
18:16
job is to win the AI race, and the energy system is starting to reorganize around that mission.
18:21
Almost nobody voted for this. There was no congressional debate
18:25
or national plan, just a string of private deals announced through company press releases.
18:30
Pennsylvania Governor Josh Shapiro wrote letters to PJM helping fast-track the
18:34
interconnection. Every level of government has greased the wheels. None of them has
18:39
asked who the power is for, or who gets to use it, or who pays when the bill comes due.
18:44
The shared system is being carved into corporate territories, and most Americans don't yet realize
18:49
the shift has already happened. Chapter 9: The Reckoning
18:52
China has grasped something America still hasn't accepted. Two systems
18:56
are now competing for the future of AI, and they aren't playing the same game.
19:00
On one side is China’s system. A coordinated national buildout built around 8 hubs and 10
19:06
massive computing clusters, where data centers are placed directly next to the country’s
19:10
cheapest and most abundant power. The entire thing moves as one machine. Energy policy, transmission
19:16
planning, latency targets, and industrial strategy all line up under the same state-led plan.
19:22
There are no years lost fighting through overlapping regulators or court challenges.
19:26
Once Beijing decides to build, the system moves. On the other sits America's parallel grid.
19:32
A private nuclear network for the hyperscalers, sitting alongside a
19:36
decaying public network. Two grids inside one country, separated by who can afford which.
19:41
The AI century will go to whoever locks in scalable, always-on baseload power at
19:46
the scale that frontier model training and global queries actually require. Chip leads
19:50
are temporary, and software leads can be copied overnight. Energy infrastructure is
19:55
the one moat that takes decades to build, and the one thing no amount of code can replicate.
20:00
Americans will end up paying for both halves of the split.
20:03
They'll pay higher rates for a public system being stripped of its best power, while their
20:07
utility bills and tax dollars quietly fund the private system they'll never be allowed
20:12
to use. The June 2026 rate hikes are the first visible bill, and they won't be the last. Every
20:18
dollar of it comes from somewhere, and that somewhere is the same families now staring
20:23
at over $200 a year in new electricity costs. The new energy poverty isn't a forecast anymore.
20:29
It's already itemized on next year's bills, already approved by regulators,
20:33
already scheduled to land. The American story has always
20:36
been that markets sort themselves out, that competition drives prices down, that innovation
20:42
lifts everyone. That story works when the playing field is roughly level. It breaks when the same
20:47
handful of companies own the field, the stadium, the lights, and the power plant lighting them.
20:53
China built one national asset. America chose two grids, one for the future and one for the rest.
20:58
The physical war for AI supremacy is fought with atoms and cooling water, and the United
21:03
States has already outsourced the battlefield. But underneath all the talk about innovation
21:08
and free markets, something else is starting to emerge. The biggest companies aren’t just
21:12
competing in the market anymore. They’re starting to control the infrastructure
21:15
the rest of the market runs on. Find out in The Market Is Dead. AI is the Cartel. Or watch this.
