0:00
So the whole US stock market including
0:02
your 401k, your index funds and the
0:04
value of your retirement is based on a
0:07
story that might be coming to an end.
0:09
>> You mentioned earlier the dot bubble.
0:11
Are we doing bubble 2.0 right now?
0:14
>> Oh, this is much bigger. The AI the AI
0:18
buildout relative to the TMT buildout of
0:20
992000 is multiples even as a percent of
0:25
>> Okay. So, one of the big reasons why the
0:28
stock market is being held up right now
0:30
is because there's a story that American
0:32
companies are going to make trillions of
0:34
dollars in profits forever because the
0:37
world will be forced to use America's
0:40
>> You have a lot of data that you look at.
0:42
Do you think China is getting better at
0:46
>> two there are two relevant tech centers
0:47
on two and a half America, China,
0:49
Israel. Those are the tech centers of
0:51
the world. They will win or we will win.
0:54
Now, some people say that that story is
0:56
coming to an end because of all the
0:58
lies, the spending, and the competition.
1:01
Now, on June 12th, something interesting
1:03
happened. A letter was sent to a company
1:05
in San Francisco. That letter was from
1:08
Howard Lutnik, the commerce secretary of
1:10
the United States. And by the end of the
1:11
night, the most advanced AI in the
1:14
world, was shut down. Andropic, which is
1:17
the company behind Claude, was ordered
1:19
to cut off two of its most powerful AI
1:23
models from every foreign national in
1:26
the world, not just China, by the way.
1:28
That order included countries like
1:30
France, Germany, Japan, and even
1:33
anthropics employees. If they weren't
1:36
American citizens, they were also locked
1:38
out. Now, 4 days after that letter,
1:41
France fires Palunteer. The French prime
1:44
minister said that we can't depend on
1:46
partners who are capable of turning off
1:48
the tap. But Germany already walked
1:50
away. Spain told its companies to stop
1:53
signing deals. The same goes for
1:55
Britain. And it's because the world
1:58
found out that it has a choice. And the
2:01
choice is literally 7 and 12 times
2:04
cheaper. Cuz while America is spending
2:06
$1 trillion a year building AI, which is
2:09
3% of the whole US economy, China is
2:13
spending a fraction of that and giving
2:16
it away virtually for free. So in this
2:19
video, I want to explain how China is
2:21
competing with the US and how this AI
2:24
story might be coming to an end and some
2:26
of the things that you can use to
2:28
potentially see this bubble popping
2:30
before anybody else. So with that said,
2:32
let's get into it. Hi, my name is Henri
2:34
Jick. Hope you're doing well. Come for
2:36
the finance and stay for the AI bubble
2:38
everyone saw coming. So, let me just
2:40
start with a basic question. Do people
2:42
really want this AI technology? Because
2:45
there's a theory that says the reason
2:47
that this is such a prevalent story in
2:49
the market is so that these tech
2:51
companies could justify their insanely
2:54
high stock prices because in reality
2:56
they've run out of really good ideas. In
2:59
fact, there's a really good interview on
3:00
CNBC with Ed Zitron who brought up a lot
3:03
of really great points.
3:04
>> But fundamentally, large language models
3:06
are not the future. The only reason big
3:08
tech is investing in this is that
3:09
they've run out of hyperrowth ideas.
3:11
They don't have a next iPhone. They
3:13
don't have a new Google search. So,
3:14
they've put over a trillion dollars with
3:16
trillions more to come into a kind of a
3:18
deadend industry because when they when
3:20
that ends, they'll have to admit that
3:22
they don't have anything else. Now,
3:23
throughout the video, I'm going to show
3:25
you more clips from that interview, but
3:27
there was also an interview with Alex
3:29
Karp, who is the CEO of Palunteer, which
3:32
if you don't know is the company that
3:34
works closely with the government and
3:36
pretty much every three-letter agency in
3:38
the world. And Alex also brings up the
3:41
fact that nobody really trusts AI right
3:44
>> Who owns the data? Where is it cached?
3:47
Are the prompts secure? Is this being
3:49
transferred to you? Are you being comp?
3:52
Okay, if it was so valuable, let's say I
3:54
can make you a billion dollars right
3:55
tomorrow. Wouldn't I say I'll make you a
3:58
billion dollars and I want 30%. Why are
4:00
they charging for tokens if it's so
4:02
valuable? He is saying if the promise of
4:05
AI is as good as they are marketing it
4:08
to be in its current form, they would
4:11
not be charging us for tokens. Instead,
4:13
they'd be charging us for building a
4:15
billion dollar business idea where they
4:18
would take 30% of the revenue. Cuz think
4:20
about how you pay for anything in
4:22
business. You pay a lawyer to win a
4:24
court case. You pay a contractor to
4:26
remodel your house. The price you pay is
4:29
attached to a specific result. Now, AI
4:33
companies do not work that way. They
4:35
charge us per what's called token usage.
4:39
Now, a token is basically a word. Every
4:42
word the AI reads and every word it
4:44
writes for you, we pay for that. whether
4:47
the answer was good or was really bad.
4:49
And Alex Karp is basically saying why
4:52
would they price their business that
4:54
way? Why wouldn't Open AI chat GPT just
4:57
say only pay me when it works? If we
5:00
create a good idea for you, give us a
5:02
cut of your income. But I'm telling you
5:05
in this country at every single
5:07
enterprise I deal with they these people
5:09
are livid. They're like I am paying for
5:12
tokens that create no value. These
5:14
people are stealing the weights and
5:16
alpha of my business and they're
5:17
creating a wealth tax that does not help
5:19
the poor. It just punishes starts with
5:21
the billionaires. Every single person at
5:23
this table is going to be paying a
5:24
wealth tax only to punish us.
5:27
>> If they were confident that this thing
5:28
created this value, that would be the
5:31
easiest sales pitch in history. Pay us
5:33
nothing unless we make you money and
5:36
unless we build you a billion dollar
5:38
idea. But they can't offer that because
5:41
these models do what's called
5:43
hallucinate. Right? This is where they
5:45
confidently make things up. And nobody,
5:48
including the people who built them,
5:50
could tell you when or really why it
5:52
happens. Right? No one's been able to
5:54
figure out how to fix it completely.
5:56
>> You'll notice that both Anthropic CEO
5:58
Darama Day and Sam Wman have both said,
6:00
"We can't wait to see what you build
6:01
with this." Well, that's because they
6:04
don't know what you can build with this.
6:05
They want everyone else to do their
6:06
innovation for them. spend as much as
6:08
they can on tokens and then take
6:09
whatever's left except they lose too
6:11
much money for that strategy to actually
6:14
>> And that puts every corporation in
6:16
America in a very awkward situation
6:19
because let's say you're the CEO of a
6:21
corporation, right? You just spent $50
6:23
million on AI this year. So your board
6:26
of directors asks you a question.
6:28
They're like, "What did we get for this
6:30
$50 million we just spent? What's the
6:33
ROI?" And you're like, "I don't know,
6:35
right? We don't have a number. Nobody
6:37
has a number. Corporate America's paying
6:40
subscription fees on a technology whose
6:42
outcome we cannot measure. And it gets
6:46
worse though because not only can we not
6:48
measure it, we are also risking our
6:52
company secrets and potentially creating
6:54
a competitor. Here's what Alex has to
6:57
>> But something has gone completely wrong.
7:00
And the basic view among enterprises in
7:03
this country is I'm going to chill lax
7:06
uh and waste my time with tokens. I'm
7:08
going to get no value and they're going
7:11
>> The fear for all these CEOs is that when
7:14
your company uses these models, your
7:17
data flows through them, right? Your
7:19
process, your trade secrets, the special
7:21
sauce that makes you profitable, which
7:23
Alex Karp calls the alpha. So what
7:27
happens when the AI company that you use
7:31
learns from your business? It just
7:33
becomes your competitor. And this is not
7:34
a hypothetical thing by the way.
7:36
Anthropic launched a design product
7:38
called Claude Design while having a
7:41
relationship with a company called
7:43
Figma, which is a design company. Figma
7:46
CEO publicly said he was shocked. So,
7:50
picture being a business and then
7:53
watching that. You're paying your vendor
7:55
millions of dollars a year to use their
7:57
AI, but what you're actually doing is
8:00
paying them and training your own
8:03
replacement. So, what's the solution
8:06
then? This is where it gets really
8:08
>> But what is happening among the most
8:10
technical players is they're saying, "I
8:13
want something I own. This is my
8:15
business. I want to own the GPUs. I want
8:18
to own my data. I want to own the model.
8:20
I want to control the alpha. Why would
8:22
they get access to my data? If they're
8:24
going to build my alpha, why wouldn't I
8:27
>> Right? What that means is instead of
8:29
renting an AI from someone, instead you
8:32
just download one, you run it on your
8:35
own computers using your own data where
8:37
no one can see it and no one can learn
8:39
from it and also no one can shut it off.
8:42
He then goes on to say though that most
8:44
businesses don't even need the latest
8:46
and greatest cuttingedge AI because you
8:49
don't need the smartest one to process
8:52
insurance claims, right? You need one
8:53
that's specifically really good at
8:55
insurance claims. And what's interesting
8:57
is that Alex Karp profits from this
9:00
business model as well. So why would he
9:02
be saying all this? What's his motive?
9:04
Cuz remember France, Germany, Spain,
9:07
they're canceling their contracts.
9:09
Palanteer has been losing those
9:10
contracts across Europe all year. Now, 2
9:13
days before that interview, Palanteer
9:16
announced a partnership with Nvidia to
9:18
sell open models in sovereign
9:22
Basically, that interview was a product
9:25
launch for his new service, which is why
9:28
he's out there telling other companies
9:29
to download and own their own AI. So,
9:33
that is the first problem with AI.
9:36
Nobody trusts it. But there's a second
9:38
problem that's even bigger. Now, before
9:40
I explain the second problem, everything
9:42
in this video, like the AI spending and
9:44
whether this is a bubble at all, depends
9:46
completely on where you're reading it.
9:47
And that's where today's sponsor, Ground
9:49
News, comes in. Ground News is an app
9:51
that shows you the same story from
9:52
hundreds of different outlets at the
9:54
same time. And it tells you which ones
9:56
are left-leaning, right leaning, or
9:57
center, so you can see how the story
9:59
changes depending on the source. Perfect
10:01
example, South Korea just announced a
10:03
huge national AI and chip investment.
10:05
Over 159 news sources covered it. And
10:08
here's what ground news shows.
10:10
Left-leaning outlets frame this as a
10:11
historic industrial strategy, stressing
10:14
the huge scale, and they lead with the
10:16
market's outcome. Right leaning outlets
10:18
lean into the urgency and survival using
10:20
phrases like race against time, framing
10:23
it as existential for South Korea's chip
10:25
industry. Now, center outlets just skip
10:27
the drama and focus on policy execution.
10:30
Even the headline number changes
10:31
depending on the source. Some report a
10:34
thousand trillion one, others up to
10:35
2,000 trillion, some just say 1.2
10:38
trillion, but it's the same announcement
10:39
with three different narratives. That's
10:41
what ground news makes visible. I use it
10:43
when I'm researching for these videos so
10:45
I can separate what's actually happening
10:46
from how it's being told. If you want to
10:48
see the full picture of global events
10:50
and not just one side, go to
10:53
or click the link down below. You'll get
10:55
40% off the Vantage plan, which is the
10:57
one I use, and you'll be supporting
10:58
independent journalism and this channel.
11:00
Thank you to Ground News for sponsoring
11:02
this segment. And now, let's get back to
11:03
it. Now, the second problem with AI is
11:05
that even if every company in America
11:07
trusted these AI companies, the money
11:10
still does not make sense because the
11:12
business model is broken in a way we've
11:14
never seen from tech companies. Because
11:17
here's how software is supposed to make
11:19
money. Software is the greatest business
11:21
model ever invented because you spend a
11:23
lot of money building the thing at once,
11:26
right? And then every new customer is
11:28
basically free money. That's why tech
11:30
stocks have done so well over the past
11:32
few decades. When you buy Microsoft
11:35
Excel, right, Microsoft doesn't spend
11:37
anything extra to sell you that copy.
11:40
Their costs stay the same, but their
11:42
revenue goes up. And it's the gap
11:44
between those two lines that is their
11:46
profit. And that gap is why tech
11:48
companies, the most valuable companies
11:50
on Earth. Now, AI broke that model. And
11:54
how they broke it was every single time
11:57
you ask Chad GPT a question right now,
11:59
it costs Soap AI money cuz it uses
12:02
electricity. Chips are being worn down.
12:05
So more customers does not translate to
12:08
free money anymore. More customers means
12:11
more cost dollar for dollar. So AI is
12:15
not a software business. It's more like
12:17
a restaurant, right? Every time a meal
12:19
gets served, somebody has to buy the
12:21
ingredients every time. Except this is a
12:24
restaurant that loses money every time
12:26
it serves food. And its plan to fix it
12:29
is to serve more food. Now, let me give
12:31
you some context. In 2025, Open AAI
12:35
burned over $20 billion in just one
12:38
>> Well, they'd be the first to be this bad
12:40
other than we work. And even then, this
12:42
is so much worse than that. Open AAI
12:43
burned $20.9 billion in 2025. That's the
12:46
auditive financials that the FT and II
12:48
reported. And the problem with these
12:50
companies is their margins are getting
12:51
worse and they actually their costs
12:53
increase linearly with their revenues.
12:55
>> So he's basically saying that costs
12:57
increase linearly with revenues, right?
13:00
The two lines are going up together and
13:02
the gap never really opens up. Now for
13:05
25 years, every investor has been
13:08
trained to be patient with this cuz they
13:11
say, "Well, they're losing money now,
13:12
right? But at scale their margins will
13:14
get better, right? Amazon lost money for
13:17
years. Except with AI, we just keep on
13:20
waiting and the margins are getting
13:22
worse cuz every new model costs more to
13:26
run than the last one. And the market is
13:29
starting to notice it.
13:30
>> There is no proof that they can improve
13:31
their margins. No amount of specialist
13:33
silicon or supposed Vera reubans will
13:35
bring these costs down. And we're at a
13:37
point now where OpenAI is now
13:39
potentially pushing their IPO to 2027
13:41
because they couldn't get a trillion
13:42
dollar valuation. It's clear that people
13:44
are wising up to the problem of
13:46
generative AI, which is there's not
13:48
really a business there.
13:49
>> Now, all of this, by the way, is not
13:50
just open AI cuz look at who's paying to
13:53
build all of this. This is from Oracle's
13:57
>> Oracle is a particularly scary one
13:58
because they are building 7.1 gawatt of
14:01
capacity just for one customer. And they
14:03
even said in their annual report that
14:04
the risk was they might not get paid.
14:06
Open AI only loses money and I think I
14:09
estimate it's like $75 billion of
14:11
revenue annually that they will have to
14:13
pay for the full Stargate data center
14:14
project in annual compute revenue. Open
14:16
AAI can't afford that and if they can't
14:18
Larry Ellison can't afford to pay back
14:20
those bills and Oracle stock will be in
14:22
jeopardy along with the margin loans
14:23
that Mr. Ellison holds. It's genuinely
14:26
>> So Oracle is building the equivalent of
14:29
several nuclear power plants worth of
14:31
electricity for basically one customer.
14:34
a customer that just lost $20 billion.
14:37
Here's my favorite one, though. Nvidia
14:40
sells its chips to a group of smaller
14:43
cloud companies. They're called
14:44
NeoClouds. Now, those companies borrow
14:47
billions of dollars to buy Nvidia's
14:48
chips and Nvidia rents them back.
14:52
>> I think companies like Core and
14:54
especially Nebius and Iron and Cipher
14:56
Mining and all of them, Terowolf as
14:58
well, they are all very they're
15:00
basically outgrowths and they're
15:02
subsidiaries of Nvidia. Nvidia is now
15:05
according to the information going to be
15:07
paying them to rent back their GPUs when
15:09
they install them in the data center.
15:11
This is the this is something that only
15:12
happens in an industry without diverse
15:16
>> So what he's saying there is that
15:17
Nvidia's sales are partially funded by
15:20
Nvidia. That's like a car dealership
15:22
lending you money to buy a car and then
15:25
paying you to borrow the car back for
15:27
the weekend and then reporting all of it
15:29
as demand. Now look how much profit
15:31
we're making, right? Yeah, because you
15:33
are buying back your own equipment.
15:35
There's a name for when an industry
15:37
starts doing this. It's called not
15:39
enough real customers. So for companies
15:41
investing trillions in AI like
15:43
Microsoft, Google, Amazon, Meta, what is
15:47
the ROI from all their spending? They
15:50
won't tell you. These companies report
15:52
everything. Cloud revenue, ad revenue,
15:54
YouTube revenue. But AI revenue, they're
15:58
not telling us that. Microsoft, Google,
16:00
and Meta, and Amazon are all doing a
16:02
funny little I don't want to call it a
16:03
scam, but it's a a trick where because
16:05
their other businesses are still
16:07
growing, but they never disclose their
16:08
AI revenues, everyone conflates that
16:10
with AI driving their growth. In
16:11
reality, their other businesses are
16:13
growing and AI is losing them money
16:15
across the board. You'll notice that
16:17
neither Microsoft or Amazon, who both
16:19
share their run rate of AI, will share
16:21
the actual AI revenues. That tells you
16:24
that these companies are afraid. Public
16:26
companies love good news. If they had
16:27
good news, why wouldn't they share it?
16:29
That's because they've only got bad news
16:30
>> Now, as of right now, the stock market
16:32
is still patient and investors are
16:34
saying, "Okay, give it time still." But
16:37
all of it really depends on one big
16:40
assumption, which is that if and when
16:43
the profits do come, it's going to be
16:46
the American companies that will make
16:48
the profits because the world has no
16:51
other option. But the third problem with
16:53
AI is that the world has another option.
16:56
That option is called China. So, let me
16:58
show you what China is really doing.
16:59
Remember this chart from the beginning
17:00
of the video. The trillion that America
17:03
is spending. Well, here's something
17:04
interesting. This is China. On that same
17:07
scale, America, $764 billion this year,
17:12
and then 1 trillion next year, 3% of the
17:16
whole US economy. China, 102 billion
17:20
this year, 123 billion next year. 0.6%
17:24
6% of their economy, which means
17:27
America's outspending China almost 10
17:30
to1. Why? It's cuz China figured
17:33
something out. A developer took the
17:36
exact same coding task and gave it to
17:39
two AI models, Claude Opus, which is one
17:41
of the top American models made by
17:43
Anthropic, and GLM, which is a Chinese
17:46
open model. Both models finished the
17:49
same task, and both took about 5 1/2
17:52
minutes. The American model charged
17:57
and the Chinese model charged 31. That's
18:01
7 12 times cheaper. Now, before you
18:04
think that that's a cherrypicked test,
18:06
here is the industrywide data. This is
18:10
called the artificial analysis
18:12
intelligence index. And this is
18:14
basically the official rankings of every
18:17
AI model in the world. Now look at the
18:19
top. The best American model scores 60.
18:23
Now look right here. This is GLM. The
18:26
best Chinese open model 51. And look at
18:29
how much of this chart is from China.
18:32
Deepseek, Quen, Kimmy, Miniax. They're
18:35
not at the top. They are everywhere.
18:37
They are filling the whole middle of the
18:40
global rankings. Now to be fair, the US
18:43
still has the smartest AI in the world.
18:45
And that's true. But if you ask the
18:47
question that every business is asking,
18:50
do I need the fastest AI model to manage
18:53
my business? The answer is no. For most
18:56
businesses, that's things like answering
18:58
their customer service emails to
19:01
basically do the boring work that is 90%
19:04
of what companies actually use AI for.
19:08
Based on that logic, China's winning.
19:10
The US is winning the race for the most
19:12
dollars spent, and China is sort of
19:13
winning the race for the customer. Now
19:15
the question is how is China doing this
19:18
while spending 10 times less money? And
19:20
the answer is distillation. Okay, here's
19:23
how it works. When you train a frontier
19:26
AI model from scratch, that means
19:28
spending billions of dollars teaching it
19:30
everything the hard way. But there's a
19:32
shortcut. You can train your model by
19:35
studying the answers of a model that
19:37
already exists. This is basically like
19:39
copying someone else's homework and then
19:40
you know the answer for almost no money
19:42
spent, by the way. So the US spends
19:45
trillions of dollars doing the hardest
19:47
research in human history and then China
19:49
just sort of copies it by compressing
19:51
the results into smaller cheaper models
19:53
and then it gives them away for free. It
19:56
open sources them which means anyone can
19:58
download them. And if you think about it
20:00
every dollar of US AI spending it's kind
20:03
of like a donation to the Chinese AI
20:06
industry. And this has become common
20:09
practice for China. So much so that they
20:12
are doing it as a side hustle. Check
20:14
this out. This is a model, for example,
20:16
called LongCat. Look at the benchmarks.
20:18
It's going toe-to-toe with Google's
20:20
Gemini, and it's beating older versions
20:23
of Anthropic's flagship models on
20:25
realworld agentic tasks. But the thing
20:28
is, do you know who built Longat? It's a
20:31
company called Mtoan. Do you know what
20:33
Mtoan does? It's a food delivery
20:37
company, right? It's the Chinese Door
20:39
Dash equivalent. And they built an AI
20:41
that competes with the smartest labs in
20:44
the United States. Which means when a
20:47
company like that can do what a trillion
20:50
dollar US company is doing, is that
20:53
company still actually worth trillions
20:55
of dollars? Maybe not. Cuz remember the
20:58
assumption that's holding up the whole
21:00
AI stock market is that let's be patient
21:03
guys. the profits will come and when
21:05
they do all the US companies will
21:08
collect those profits because the world
21:09
has no other choice. But here's what the
21:12
actual cost is when there is a choice.
21:15
Right? This is the same type of work.
21:17
The American model shows 18 1.5 cents
21:20
per task and the Chinese model 4 cents.
21:24
Right? Within a few quality points of
21:28
discount. This is sort of the chart that
21:31
destroys the whole story because you
21:34
cannot make back a trillion dollars
21:37
selling something that your competitor
21:40
is giving away at 90% of the quality for
21:44
just 10% of the price. So, let me tie
21:46
all of this together. If all of this is
21:48
true, then when does this bubble pop, if
21:51
ever? And logic says it's when
21:54
corporations stop their capex, their
21:57
capital expenditures. It's when they
21:59
stop spending money building all these
22:01
data centers. But believe it or not,
22:04
that is not when the bubble pops.
22:07
According to the data, data shows that
22:09
the bubble could pop a lot sooner. And
22:12
here's why. During the dot bubble, the
22:16
NASDAQ index peaked in March of 2000.
22:19
Now, all the companies that were laying
22:21
the fiber optic cables at the time,
22:23
which are the data centers of that era,
22:25
they kept spending billions of dollars
22:28
well into 2001, even though the stock
22:32
market collapsed a full year before
22:35
their spending stopped. So, the market
22:38
did not wait for companies to stop
22:40
spending and to admit to anything. The
22:43
logic of the market changed when enough
22:45
investors stopped believing in that
22:48
story. So the trigger this time around I
22:51
think will be something a lot more
22:53
subtle. Something like a big tech
22:56
earnings call where a CEO says something
22:58
like we are moderating uh the pace of
23:01
our infrastructure investment or some
23:03
boring small thing like that. And that's
23:05
because the first company that gets
23:08
rewarded by Wall Street for cutting
23:10
their AI spending that will give every
23:14
other CEO permission to do the same
23:16
thing. In fact, according to Ed Zitron,
23:18
Goldman Sachs recently said that the
23:20
first hyperscaler to pull back on
23:23
spending will get rewarded by the
23:26
>> So, I heard Goldman analysts say
23:28
recently that the first hyperscaler to
23:30
pull capex will get rewarded by the
23:32
markets. I think the capex pullbacks are
23:34
they're the sign. I also think any
23:36
financing falling through Baro and AI or
23:38
anthropic would be a sign, but I think
23:40
we're going to start seeing AI companies
23:42
kind of start falling out of favor and
23:44
not being able to raise money. But the
23:46
big thing is debt. When data center debt
23:48
stops being issued, that will be when
23:50
it's bedtime for this industry because
23:52
even if they think AI is going to win,
23:54
we've got 100 gawatt or so of data
23:56
center capacity allegedly under
23:57
construction or in planning. That's
23:59
trillions of dollars of money that needs
24:01
to come from somewhere and we are
24:03
tapping out the debt markets. We saw
24:04
that with Google raising that $85
24:06
billion equity raise.
24:08
>> That's going to be one of the early
24:10
signs. Now, another sign that we could
24:13
be at the peak of the bubble is the bond
24:15
market. That's because unlike the stock
24:18
market, which runs on stories of hopes
24:20
and dreams, the bond market doesn't work
24:23
like that. All bond investors care about
24:27
is will I get paid my interest payment.
24:30
Right? The moment they get scared, they
24:33
start to demand a much higher interest
24:35
rate. Now, how we measure their fear is
24:38
something called a credit spread. Here's
24:41
how that works. In the world of
24:43
investing, there's a concept called the
24:45
riskfree interest rate. It's called that
24:49
because it is set by the US government
24:52
which is considered to be the safest
24:54
borrower on earth. That's government
24:56
bonds, right? Whatever they're at,
24:58
that's the risk-free rate. Okay? But
25:01
remember, companies can also issue
25:04
bonds. Except because companies are
25:06
risky, cuz they can go out of business,
25:09
their bonds pay that risk-free rate plus
25:15
something extra to compensate you for
25:18
the risk that their company could go
25:19
broke and never pay you back. Makes
25:21
sense, right? Well, that extra between
25:24
the risk-free rate and their rate, that
25:28
is called the spread. Think of it as an
25:30
insurance premium. When lenders feel
25:33
safe, the premium is small, meaning the
25:36
spreads are what's called tight, meaning
25:38
corporate bond rates are close to the
25:40
risk-free rate. But when investors feel
25:44
like there's some market risk, the
25:47
premium explodes, right? The spread
25:51
increases. That is one of the early
25:53
signs that we could start to see that
25:55
this is going to fall apart. Now, here's
25:57
an example. By the way, see this
25:59
increase in 2008. Spreads hit almost
26:04
Lenders started to charge very high
26:06
prices. Credit shut off completely and
26:08
companies that ran on borrowed money
26:10
just collapsed. Also see the jump in
26:13
2020. Now look at today. The spreads are
26:19
That is close to the lowest and the
26:21
calmst readings in recorded history.
26:24
What this means for now is that either
26:27
bond investors see no problem and
26:30
everything in this video is completely
26:31
wrong or bond investors are wrong and
26:35
they can be wrong. Look at early 2007.
26:38
The housing crisis was already underway.
26:41
Bear Sterns was months away from blowing
26:43
up and spreads were only 2 1/2%. They
26:47
were super calm right where they are
26:50
today. Right? The fear gauge didn't
26:52
predict 2008, though. That's because
26:54
spreads don't really measure what is
26:57
true. They measure what lenders believe.
27:00
In 2007, lenders believe the housing
27:04
market was safe. Which now we know
27:05
obviously that it wasn't. But the point
27:08
is is that when you see someone on the
27:09
news say, "Hey, don't worry. AI is
27:12
totally safe. It's doing great. Credit
27:14
markets, right? The spreads are not so
27:16
worried." Right? If you hear that,
27:18
remember that the credit market wasn't
27:20
worried in ' 07 either. Credit markets
27:22
can be wrong and they have been wrong
27:25
>> Although you you'd agree that spreads do
27:27
not do not imply that that moment is
27:30
>> Spreads have been wrong before. That's
27:32
the thing. And I think that perhaps the
27:34
timing isn't going to be immediate, but
27:35
at some point a hyperscaler is going to
27:37
pull back capex. And when that happens,
27:40
well, this is an industry of followers.
27:41
The tech industry doesn't have ideas.
27:43
They just copy each other. Everyone
27:44
copied Sachin Nadella when he put chat
27:47
GPT in Bing. And I think that whoever
27:49
breaks capex first, they'll follow them,
27:51
>> And finally, I just want to show you
27:53
what Michael Bur posted a few days ago.
27:55
And remember, he's the guy who predicted
27:57
the 2008 financial crisis. So, in chart
27:59
one, he shows chip stocks are trading at
28:02
the top of their 15-year valuation
28:06
range. Basically, the same peak that
28:08
they hit right before the 2024
28:09
correction, which is marked with those
28:11
red circles. The market is basically
28:13
pricing chips like the trillion dollars
28:15
has already been made. Now chart two is
28:18
even more interesting. This tracks the
28:20
three groups of AI stocks since last
28:23
year. Now the gray and white lines going
28:25
up to 200% are the AI winners, right?
28:28
The companies selling the chips and the
28:30
equipment. But the orange line way at
28:32
the bottom that's barely above zero are
28:36
the hyperscalers, right? Companies like
28:38
Microsoft, Google, Amazon, Meta. What
28:41
does that mean? It means the market is
28:43
telling us that the companies that are
28:46
doing the spending, the trillions of
28:48
dollars, right, they're getting almost
28:50
no credit for it. Their stock values
28:52
aren't really going up. Wall Street
28:54
instead is rewarding the companies that
28:57
are getting that money and it's ignoring
28:59
the companies spending to build it.
29:01
Right? That's basically the market
29:02
admitting it doesn't believe the
29:05
spenders will make it back. And in the
29:07
third chart he posted, it shows the
29:10
Silicon Data LLM token expenditure
29:12
index. It's a fancy name, but what it
29:15
shows is it shows us the price that
29:17
people pay for AI tokens. This index is
29:21
the price of AI itself. And look at it.
29:24
It's down almost 20% from its high in
29:25
May. Now, the question is, why would the
29:28
price of AI be going down during the
29:30
biggest AI buildout in history?
29:32
Bloomberg says either it's because
29:34
demand is going to cheaper models or
29:36
buyers are just not willing to pay more.
29:39
Look at the middle one. Demand is
29:41
shifting towards cheaper models. And
29:43
that is the China theory that's showing
29:46
up in this data. Now, to be fair though,
29:48
Michael Bur's been early before. And
29:50
when people say early in the market,
29:52
that's a polite way of saying he's been
29:54
wrong, right? He's made market crash
29:56
predictions over the years quite a lot.
29:58
That didn't really come true. And even
30:01
this index has dips that have recovered.
30:04
Basically, Bloomberg says that the
30:05
signal for all of this is ambiguous,
30:07
right? We can't really learn anything
30:09
from this data. It can mean anything.
30:11
So, basically, the real answer to how
30:14
long it will take for the AI bubble to
30:15
pop, if ever, is that no one knows. But
30:19
those are some of the early signs to
30:21
look for based on the data from the
30:24
past. Now, if you're interested in
30:26
seeing how I'm personally preparing and
30:27
more of my thoughts, those videos live
30:29
in the premium member section where
30:30
you'll also get access to my videos
30:32
earlier. And if that's valuable, the
30:33
link is down below. It allows me to make
30:35
more videos like this one and take on
30:36
fewer sponsors. Thank you for watching
30:38
and being a premium member. I hope you
30:40
have a wonderful rest of your day. Smash
30:41
the like button, subscribe if you
30:43
haven't already. Would love to see you
30:44
back here next time. See you soon.
