Hang tight while we fetch the video data and transcripts. This only takes a moment.
Connecting to YouTube player…
Fetching transcript data…
We’ll display the transcript, summary, and all view options as soon as everything loads.
Next steps
Loading transcript tools…
The AI Bubble Is About To Burst | Sasha Yanshin | YouTubeToText
YouTube Transcript: The AI Bubble Is About To Burst
Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.
Share:
Video Transcript
Hey guys, it's Sasha. The hype around AI
exploded this week as OpenAI released
Sora 2, its latest video generation
tool. Tech bros and AI fans began
dribbling all over their keyboards as
they sang praises on Twitter. I tried to
download and try the app out, but it is
apparently not available outside the
United States. So, I had to rely on the
experiences of people who have appeared
to be able to actually try it out. And
distilling through the social media, it
appears that although there are definite
improvements, the story seems to be much
the same as it was before. To get one
short usableish clip, you typically have
to go through at least dozens and dozens
of completely unusable clips with major
errors in them. Here you can see a
couple of typical examples that I saw on
Twitter. First, there's a skateboarder
who seems to grow arms while performing
about a dozen shoulder dislocations in
midair. And then a boxing video with the
typical signs of AI slop with breaks and
continuity, weird morphing just doesn't
look at all right. And as per usual,
there is no way to edit existing clips
that is not coming for some time. For
example, making a very small alteration
to an already generated video. Maybe
there's one little detail that you want
to change. No, instead you have to
change the prompt that you put in and
hope that the next video, which will be
completely different to the previous
attempt, is good enough and doesn't have
issues. It appears that the videos that
come out are a lot better if the type of
video that you want can rely on very
similar pre-existing videos being out
there already that were used by OpenAI
in the training data set for the model.
So, if you just want to blatantly
plagiarize or copy something that's
already been done and just with slight
tweaks, then your chances of the video
not looking completely broken increase.
And naturally enough, this opened up the
usual contest on Twitter between people
who want to defend the rights of
copyright holders, you know, people who
have produced the original content for
which OpenAI does not pay them money but
then uses, people whose work has been
outright stolen in order to train the AI
model so that it can then spit out near
identical output for commercial gain.
And by the way, I was just reading just
before recording that apparently in the
first few days, OpenAI is already
restricting your ability to generate
videos with some characters, especially
Nintendo based characters because
apparently Nintendo really likes suing
people. But obviously for smaller
creators, you can just go and get
[ __ ] right? Nobody cares about you.
They'll just steal your stuff and let
the AM model train on it because you
will never have enough dollars to sue
them. Tech bros on the other side are
busy shouting very loudly that copyright
law is for boomers. AI companies are
allowed to do whatever they want, of
course, and completely ignore the law
because it's the future, bro. Although a
number of lawsuits have already been
launched against Open AI and other AI
companies by the likes of New York
Times, a bunch of other ones. The
progress of these lawsuits seems
extremely unusually slow and there seems
to be a complete lack of interest in any
of this from the government including
the Congress and the Senate. They don't
want to do anything about this even
though it is so brazen which of course
is very important to remember. Very
important has nothing absolutely nothing
to do with the fact that all of these
politicians stand to benefit financially
personally from these AI companies
seeing the valuations go to the moon
because they were lobbyed by those
companies. The politicians were lobbyed
to look the other way. Anyway, a few
days ago, we found out that OpenAI made
$4.3 billion worth of revenue in the
first half of 2025. That's a roughly $130%
$130%
increase on the revenue run rate that
they had in 2024, which is very
impressive. The growth in revenues is
doing really well. New people are
signing up, but at the same time, they
are still continuing to lose eye
watering amounts of money. While they
made the $4.3 billion worth of revenue,
they then also spent $6.7 billion on
research and development, another $2.5
billion on inference apparently, and
another $2 billion on sales and
marketing, which means that overall they
lost almost $7 billion while taking in
$4.3 billion from its customers. And
that's before accounting for capex
spend, which is probably dwarfing all of
the other costs put together because the
billions upon billions of dollars that
are being spent on building out the data
centers and the Nvidia graphics cards
that go into those gra data centers
comes on top of these costs. But the
cost of buying graphics cards, building
all of these giant data centers just
gets amatized over long periods of time.
Even though the likelihood is that these
graphics cards will get replaced over a
much shorter period of time as new ones
come out and as they just break and stop
being as valuable. So in 2025 before
accounting for the cost of all of this
capex the graphics cards which probably
makes the cost of inference
significantly higher than the published
numbers. OpenAI spent $260 for every $1
in revenue. In 2024, using the same data
that was shared with shareholders,
OpenAI apparently only spent $2.34 per
$1 of revenue. So, the gap between what
OpenAI spends and how much money it
makes is actually currently growing
instead of getting smaller. And
according to inputs from other
companies, it is the same story with
others as well. Economies of scale are
not applying here because there is a
physical cost to generating the extra
amount of compute required to process
queries in newer models like GBT5. And a
big reason here is that the models are
getting progressively better because the
algorithm just uses more and more
compute. It chucks more and more compute
at every query instead of the efficiency
of that compute increasing. So each time
you submit a query to Open AI, it then
generates a whole bunch of internal
subquery. So the model has a chance to
structure its answer. You know, when it
says thinking, is doing all of those sub
queries. It has a chance to
self-correct, to doublech checkck
things, to make sure that the answer is
exactly what the user is using, asking
for, etc. before it spits the answer
back to the user. And this approach
costs a lot of money. And every time a
new model comes out, it does more and
more of this other stuff, which costs
even more money. And each time that the
model gets better, the cost therefore of
running that model jumps as well. But as
everyone was busy looking at Sora 2,
OpenAI quietly closed another funding
round 2 days ago. OpenAI sold 6.6
billion of shares to select investors
and employees valuing the company at 500
billion. They apparently authorized up
to $10 billion worth, but only sold
about 2/3 of that. If OpenAI was a
public company, this valuation would
place them just above Exon Mobile and
Netflix as the 19th largest company by
market capitalization in the world. Just
for comparison, in the first half of
this year, Netflix made 21.5 billion
worth of revenue, exactly five times as
much as OpenAI. And Netflix has a net
margin of 28%, which means that their
net profit is well positive and growing
instead of minus 130%. and going lower.
Open AAI says that they have 17.5
billion dollars worth of cash as of the
end of the first half of the year. And
at the rate of capex and general cash
burn, they will probably need to do more
funding rounds in the next 12 months, 24
months, and so on. And this is precisely
where this story about AI hype gets
seriously weird and interesting. But
just before I dive into this bit, trust
me, it's crazy. I wanted to briefly
mention that my investor playbook course
is on sale right now, today and tomorrow
only. If you're watching this later than
that, I'm sorry. You'll have to wait for
another opportunity. The course is all
about analytical investing. It teaches
everything from technical investing
knowledge, all the jargon, a full
breakdown of how to read financial
statements in a lot of detail. There's
literally many hours, multiple hours of
content just on that alone. And then the
course covers how to build valuation
models, good practices, investing
tactics, and a lot more. You only pay a
one-off fee to join. There's no monthly
subscription, but then you get access to
all the new content as it comes for
free. I am expanding the course. I have
some big plans for the next couple
months. And you get to join live
sessions with me where you can ask me
questions. You can listen to a deep dive
on a different topic every week. And if
you can't join the live session, you can
always watch those later as they get
recorded. If you want to sign up, make
sure you use the code flash10 at the
checkout to get your 10% discount.
That's only available today and tomorrow
only. And thank you so much in advance.
I really look forward to seeing you in
the live sessions, including next week.
So, here is the bit that I found really
interesting about OpenAI. Just a few
days ago, Nvidia announced that they
will invest $100 billion into OpenAI. On
September 22nd, 2025, Nvidia announced
this $100 billion investment in OpenAI
to build 10 gawatt of AI data centers, a
scale that would require 4 to 5 million
GPUs, nearly double of Nvidia's annual
output. So, here is the way this works.
Early investors go and give a bunch of
money to OpenAI in funding. Open AAI
then goes and hands over that money to
Nvidia to buy a bunch of graphics cards
because that's their business model.
They use those graphics cards. Those
graphics cards are currently selling at
four to five times the price that they
were before the AI boom started. But it
still costs Nvidia the same amount of
money to produce them. So Nvidia sits
there printing cash. They make a giant
boatload of money. Nvidia's balance
sheet explodes. Their current assets go
from $16 billion to $12 billion in just
four years. But wait, even after that
meteoric rise, Nvidia still doesn't have
a hundred billion to invest in OpenAI.
Their total amount of cash and cash
equivalents and short-term investments
is $57 billion in their most recent
filing. So, what's happened here? Well,
it's a circular game. Imagine you have
this imaginary, completely non-real
company called company O. Company O
wants to raise their valuation so that
the founders can make a ton of money by
cashing in and selling their shares.
Their valuation is based on hype and a
essentially a multiple of the amount of
capex investment that they make. The
bigger the data centers, the more
graphics cards are in there, the more
their shares are worth. And you can see
how huge these data centers are getting.
OpenAI is already using more energy than
entire countries like Slovenia. And the
energy usage is only going up in a
vertical line. But we're of course not
talking about Open AI. We're talking
about this completely imaginary company
called Company O. So Company O loses a
lot of money. The amount of money they
lose goes up over time per user, per
dollar that they make. and doesn't the
company doesn't have enough cash to keep
buying millions of graphics cards to
stick into all those data centers that
they're building. So, the boss of
company O goes and has a little chat
with the boss of company N that makes
the graphics cards and they say, "Look,
we don't have enough cash to keep buying
your cards, and we're not expecting to
have the cash in the next few years, if
ever. So, how about we do a bit of
collusion and we give you shares in our
company in exchange for an investment of
$100 billion. We know that you don't
have $100 billion, but check this. You
give us just $1 billion to get us
started. We take it and then we give you
the $1 billion right back to buy the
graphics cards. You then take the $1
billion that we just gave you and you
give it back to us again, the second
billion. And we just rinse and repeat
this. a hundred times over and over
until the total investment is a h 100red
billion. In the end, shares of company O
are going to go absolutely through the
roof because we're going to have the
biggest data centers stuffed with
millions of company N's graphics cards.
And then company N can sell those shares
of company O that would make way more
money for them than by just selling the
graphics cards in the first place.
That's basically roughly what's
happening here. Nvidia is now not only
selling the picss to the people going up
the mountain in search for gold, but
they figured out a way of also making a
load of money from the Klondike hype
itself as well. Yeah, good for them. And
if you thought that it's really
beginning to take the piss, just listen
to the most recent Jeff Bezos interview
from 2 days ago.
One of the things that's going to happen
in the ne it's hard to know exactly when
it's 10 plus years, but I bet it's not
more than 20 years. We're going to start
building these giant gigawatt data
centers in space. So these giant
training clusters,
those will be better built in space
because we have solar power there 24/7.
Uh and it's and and the solar power
there is there no clouds and no rain, no
weather. So you can build we will be it
will we will be able to beat the cost of
terrestrial data centers in space in the
next couple of decades.
So according to Jeff Bezos in the next
few years companies will be building out
entire data centers like huge vast
buildings with millions of graphics
cards in them in space because you get
solar power 247 there. This of course
makes complete sense and is in no way
completely [ __ ] delusional. The fact
that you can use energy storage, nuclear
power, our other forms of power such as
wind or tidal for a fraction of the
price that it would cost to lift all of
the stuff you need into orbit and then
keep it in orbit. That fact is just
irrelevant. If you ever wonder if there
is a bubble happening at any point in
time, if you hear leaders in the space
not only continuously generating hype
because that's generally what they do,
but also making these types of
statements that are just clearly
insanely stupid. And when you hear at
the same time large volumes of fanboys
amplifying this and clapping wildly when
they hear this [ __ ] that's a pretty
shorefire way of knowing that you're
entering some kind of peak hype
territory. The problem with stating the
obvious, as I am right here, is that you
inevitably attract the tech bro crowd.
Just go and read the comments who will
loudly complain that you don't know what
you're talking about. You know, you
don't understand the future. It's the
technology, man. In fact, I am a really
big fan of AI technology. I use it every
day, just like I was a big fan and an
active user of the internet during the
com crash. And at the same time as being
a fan, I could also appreciate that the
company valuations and the hype around
those valuations was complete nonsense.
The two can happen at the same time. As
more people are posting clips of short
videos generated by Sora, you know that
one good clip that they generated after
generating hundreds of unusable clips,
the cost of generating these clips is
only going up. And there is a weird take
from everyone that soon there is going
to be entire movies created by AI. I
think we're nowhere close, but soon is
happening. It's happening soon. And we
will all want to watch these movies
apparently because you'll just be able
to watch exactly what you want. And we
will want to go and look at art
generated by AI. And we will want to
listen to music generated by AI. But
here is an interesting thoughtprovoking
example for you. Chess AI is complex but
way simpler than things like generating
a video. So chess AI has basically been
solved. It's been here a lot longer. It
happened a lot earlier. It became better
than humans of playing chess somewhere
around 20 years ago and today it's not
even remotely close. On every online
chess platform, there are chess bots
that you can play against. There are
chess bots for different skill levels,
for different types of games, chess bots
that imitate the style of famous chess
players or even famous non-chess
personalities, just celebrities, and you
can play against them whenever you want.
You don't have to wait for them to move.
You don't wait to have to be matched to
another player. You can take as long as
you need. They won't drop out of games.
They won't act like a dick to you
because they are a bot. They will always
say nice things. And you know what?
Absolutely nobody plays bots. There are
millions of people who play chess
online. None of them want to play bots.
All of the humans on these platforms
prefer playing other imperfect humans
because humans like and enjoy and seek
out the raw imperfection and the random
nature of how the human mind works. They
enjoy interacting with the real human at
the other end. Even if there is almost
no visual difference in terms of how the
chess game online looks on your screen,
I think about this a lot. These new
generations of AI are going to be
absolutely revolutionary in increasing
the productivity of humans. A good
developer can produce much higher
volumes of code at a higher quality than
before using AI. A CGI artist will
probably be able to do a lot more good
work in a shorter space of time by
incorporating AI in their workflow. It's
like the personal computer and the
smartphone revolution all over again.
But it can also be true at the same
exact time as acknowledging this that
the hype around the technology far
surpasses even these very lofty benefits
that are definitely coming. And the
valuations of companies in this space
can also be in a massive bubble even
though long-term AI is here to stay. It
will be here for decades and it will
benefit all of us long term. The S&P 500
is once again at all-time highs thanks
to this ongoing growing AI hype wave.
The Shellipe ratio has just broken 40
for the only the second time in history.
The only other time, the only previous
time being just before the dot crash
happened. Investors, the majority of
whom have only ever seen bull markets
and have never seen an actual stock
market crash. The two months in 2020
doesn't count. The last one happened in
2008, are tripping over each other.
These investors are just wanting to
invest in lossmaking companies at
increasingly absurd valuations. It will
be very interesting to see how all of
this unravels. Stay tuned. I will be
here to share my thoughts unfiltered on
everything to do with the economy and
with the stock market. I know I've been
busy over the last few months. Sorry. I
will make more content. Remember, my
course is on sale until midnight
tomorrow, Sunday, midnight New York
time. If you want to get get live weekly
sessions where you can ask me any
questions that you want or suggest
topics for deeper dives. If you want to
understand more about how to do
analysis, how to build valuation models,
how to look at companies as an analyst,
investing strategy in the good times and
the bad, you can get 10% off my investor
playbook course. Use the code flash 10
to get 10% discount. Only available
today and tomorrow. And thank you to
those of you signing up. It means a lot.
I really do appreciate it.
I will see you in the next live session next
Click on any text or timestamp to jump to that moment in the video
Share:
Most transcripts ready in under 5 seconds
One-Click Copy125+ LanguagesSearch ContentJump to Timestamps
Paste YouTube URL
Enter any YouTube video link to get the full transcript
Transcript Extraction Form
Most transcripts ready in under 5 seconds
Get Our Chrome Extension
Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.
Chrome Extension