Beware Sam Altman over-exAIgerrations

0:01
[Music]
0:07
In November of 2025, Google launched its
0:09
latest AI chatbot called Gemini 3, which
0:12
was much better than its previous
0:13
chatbot, Gemini 2. According to many
0:15
users, it's also much more capable than
0:17
OpenAI's Chat GPT. They concurrently
0:20
launched a new image generation model
0:21
called Nano Banana Pro, which can
0:23
generate hyperrealistic images and
0:25
infographics with readable text. It is
0:27
far superior to chat GPT's image
0:29
generation capabilities. This has been a
0:32
disaster for OpenAI. Upon the release of
0:34
Gemini 3 and Nana Banana Pro, Gemini's
0:36
app downloads have surged. Google's gain
0:39
is OpenAI's loss. According to some
0:41
estimates, ChatGpt's usage numbers have
0:43
recently started declining. Open AAI
0:45
incurs billions of dollars of losses per
0:47
quarter. In the past, Sam Alman was able
0:50
to convince venture capitalists that
0:51
OpenAI is a technological leader and
0:53
they need billions of dollars of
0:54
investment to stay ahead. Despite its
0:57
massive investments, OpenAI is no longer
0:59
the leader. With Google's Gemini now
1:01
appearing to be better than Cad GBT, Sam
1:03
Alman declared a code red internally.
1:05
He's telling his employees they need to
1:07
make Chad GBT better, otherwise they'll
1:09
lose market share to Google. This
1:10
couldn't come at a worse time for
1:12
OpenAI. In addition to losing billions
1:14
of dollars per quarter in their core
1:16
business, they have $1.4 trillion of
1:18
data center spending commitments, which
1:19
they have no way to pay for. OpenAI is
1:22
reportedly in discussions with sovereign
1:23
wealth funds and other large investors
1:25
to raise $100 billion in early 2026.
1:28
This would be the largest funding round
1:30
for any company ever. But why would
1:32
anybody invest so much money into OpenAI
1:34
when they're no longer the technological
1:36
leader
Gemini Gains Market Share
1:38
since its release in late 2022. Chad GPT
1:40
has grown rapidly. Open AAI is not a
1:43
publicly traded company, so they're not
1:44
required to officially disclose their
1:46
financial results or key performance
1:48
indicators. But Sam Alman has
1:49
sporadically disclosed some information
1:51
about OpenAI in an attempt to garner
1:53
media attention. In October of 2025, he
1:56
claimed that Chat GBT has 800 million
1:58
weekly active users, a 60% increase from
2:01
the 500 million users they claim to have
2:02
in April. But there is reason to believe
2:05
that growth has started to slow down and
2:06
may in fact have turned negative. The
2:08
third party data provider Apptopia
2:10
provides estimates for Cad GPT's daily
2:12
active users. The number of daily active
2:15
users was increasing strongly throughout
2:16
2024 and most of 2025, but in September
2:19
it flatlined and even decreased
2:21
slightly. Apptopia also estimates the
2:24
average amount of time that each user
2:25
spent on the app each day. Back in July,
2:28
the average Chat GBT user spent about 27
2:30
minutes per day on the app. By October,
2:32
this had declined to 21 minutes. The
2:35
decrease in time spent doesn't
2:36
necessarily mean that consumers want to
2:38
use AI less. It's more likely the result
2:40
of increasing competition. The vast
2:43
majority of consumers only use the free
2:44
version of these AI apps. Maybe you used
2:47
to only use ChatGpt, but now you also
2:49
use Gemini sometimes, so you use ChatGpt
2:52
less.
2:54
OpenAI's biggest competitive threat by
2:56
far is Google. The data provider Similar
2:58
Web provides market share estimates for
3:00
generative AI companies. As of October
3:02
2024, OpenAI had about 85% market share.
3:06
By October of 2025, this had declined to
3:08
about 75%. Google has about 15% market
3:12
share. In July, Google disclosed that
3:14
Gemini had 450 million monthly active
3:16
users. By October, this had increased to
3:19
650 million, and that's only as of
3:21
October. In November, they launched
3:23
Gemini 3, which was a big improvement
3:25
over Gemini 2. Their monthly active
3:27
users and market share likely increased
3:29
significantly since then. When Google
3:31
launched Gemini 3, they published its
3:33
results on various benchmark tests.
3:35
according to the benchmarks is
3:37
significantly better than Claude Sonnet
3:38
4.5 and ChatGBT 5.1. These were the two
3:42
best models made by Anthropic and OpenAI
3:44
at the time. Most benchmarks rely on
3:47
narrow artificial tasks that can be
3:49
gained through fine-tuning and prompt
3:50
engineering. Companies choose which
3:52
benchmarks to highlight and when to
3:54
refresh them, making the results as much
3:55
a marketing exercise as a technical one.
3:58
So, the best way to compare various AI
4:00
models is to just tinker around with
4:01
them and see how useful the results are.
4:04
As an experiment, I gave both Gemini 3
4:06
and ChatGBT 5.2 the same prompt. Tell me
4:09
about OpenAI recently declaring code red
4:11
as I lose market share to Google's
4:12
Gemini 3. What implications could this
4:15
have for OpenAI and their ability to
4:16
raise venture capital or complete their
4:18
expected upcoming IPO? They both gave me
4:21
very similar results. Both of them
4:23
explained that OpenAI is worried about
4:24
losing market share to Gemini. This has
4:26
caused OpenAI to prioritize improving
4:28
Chat GBT at the expense of some other
4:30
initiatives. And while OpenAI still has
4:33
major venture capital backers, the loss
4:35
of their technological leadership could
4:36
impair their ability to raise capital in
4:38
the future. I've tinkered with both
4:40
Gemini 3 and ChatGBT 5.2 for a few hours
4:43
each, and to me, they seem approximately
4:45
equivalent.
4:46
But one area where Google is
4:48
indisputably ahead of Open AI is image
4:50
generation. Google's new Nano Banana
4:52
image generator can generate
4:54
hyperrealistic images that look much
4:56
better than anything Open AI can make.
4:58
OpenAI does have the ability to generate
5:00
images and even videos, most notably
5:02
with their video generation app, Sora.
5:05
Sora videos can look shockingly real at
5:07
first glance, but upon careful
5:08
inspection, it's pretty obvious that
5:10
they're fake. The image quality of Nano
5:12
Banana is unquestionably better than
5:14
anything we've seen from ChatGpt or
5:16
Sora.
Code Red
5:17
Shortly after the release of Gemini 3
5:19
and Nano Banana Pro, Sam Alman sent a
5:21
memo to OpenAI employees. He declared a
5:24
code red emergency. He told his
5:26
employees they need to make ChatGBT
5:28
better, otherwise they'll lose market
5:29
share to Google. OpenAI will postpone
5:32
work on other initiatives, including
5:33
advertising, AI agents, and a personal
5:35
assistant called Pulse. Instead, they
5:38
should focus all their resources to
5:39
improving Chad GBT, including its image
5:41
generation capabilities.
5:43
On December 11th, OpenAI launched its
5:46
latest model, Chad GBT 5.2. This was
5:49
just a few weeks after Google launched
5:50
Gemini 3, and less than 2 weeks after
5:52
Sam Alman declared code red. The timing
5:55
is probably not coincidental. Alman saw
5:57
that Chad GBT was losing market share,
5:59
so he rushed to push out whatever new
6:00
update he could. Sam Alman claims that
6:03
Chad GBT 5.2 is the most advanced AI
6:05
model in the world, far surpassing
6:07
Plaude and Gemini on various benchmarks.
6:09
We'll reiterate again that these
6:11
benchmarks aren't very useful because
6:13
there are ways to manipulate them or
6:14
specifically fine-tune the models to
6:16
perform well on narrow benchmarks. The
6:18
actual consumer reception has not been
6:20
great. The most common criticism is that
6:22
it feels significantly more restrictive
6:23
than prior versions. Many users describe
6:26
it as an overfiltered or nanny model
6:28
that injects safety disclaimers,
6:30
warnings, or moral framing into
6:31
otherwise normal requests, which they
6:33
feel breaks conversational flow and
6:35
makes the assistant tiring to use. On
6:38
December 16th, OpenAI launched a new
6:40
image generation model called Chat GPT
6:42
image 1.5. This allows you to make
6:45
hyperrealistic images. You can also
6:47
upload a picture of a real person and
6:49
have it incorporate the real image into
6:50
a fake background or make other changes
6:52
to it. While image 1.5 is a significant
6:55
upgrade over OpenAI's previous image
6:57
generation tools, it still lags far
6:59
behind Google's Nanobanana.
7:01
In a recent article on CNET, reporter
7:04
Caitlyn Chadawi compared ChatGpt image
7:06
1.5 to Nano Banana. The images on the
7:09
left were made with Nano Banana, while
7:10
the images on the right were made with
7:12
ChatGpt image 1.5. The first image is a
7:15
cat sitting on a Monopoly board. The
7:17
Chat GBT image is obviously fake. The
7:20
cat's fur is too smooth to be real. If
7:22
you look closely at the Monopoly board
7:24
and Monopoly money, they have incoherent
7:25
squiggles where words are supposed to
7:27
be. The Gemini image on the left looks
7:29
far more realistic. Had you told me it
7:31
was a real photo, I'd probably believe
7:33
you. In the second image, Gemini also
7:36
looks far more realistic than Chad GBT.
7:38
In the Chad GPT image, the falling
7:40
snowflakes are far too big and fluffy to
7:42
be real. Chat GPT continues to struggle
7:44
with the minute details of images,
7:46
especially when they contain text or
7:48
intricate shapes. It can make images
7:50
that look real at a distance, but gets
7:51
the small details wrong. On the same day
7:54
that GPT image 1.5 was launched, Sam
7:56
Alman tried to showcase his capabilities
7:58
on his Twitter account. He posted a
8:00
graphic of a sexy Christmas calendar
8:02
showcasing himself as a jacked
8:03
firefighter. Creepiness aside, notice
8:06
that it shows December 1st as a
8:07
Thursday. In 2025, December 1st was
8:10
actually a Monday. That's the problem
8:12
with ChatGpt. It just generates slop. It
8:15
can't fact check itself.
8:18
Open AAI, Google, and all the other AI
8:20
companies are constantly improving their
8:22
models. So, it's not particularly
8:23
important to compare the capabilities of
8:25
their products at any given point in
8:27
time. It's more important to look at the
8:28
bigger picture and we can see that
8:30
Google has a few key advantages over
8:32
Open AI. Open AI burns billions of
8:35
dollars per quarter and is completely
8:36
reliant on venture capital funding to
8:38
sustain its losses. Today, OpenAI has
8:41
raised approximately $60 billion of
8:43
venture capital. Google is also spending
8:45
tens of billions of dollars on AI
8:47
research and compute. Gemini is almost
8:49
certainly losing money, just like Chat
8:51
GBT. But Google can afford these losses.
8:54
In the first 9 months of 2025, Google's
8:57
parent company, Alphabet, generated $86
8:59
billion of operating cash flow. Just in
9:01
the first nine months of this year,
9:03
Google's generated more cash flow from
9:05
its core businesses than all the venture
9:06
capital OpenAI has raised since
9:08
inception. This gives Google a huge
9:11
advantage. They can afford to invest as
9:13
much money as they want into Gemini
9:14
indefinitely. Market sentiment and the
9:17
health of the venture capital industry
9:18
are irrelevant to Google because they
9:20
generate more than enough money
9:21
internally. Secondly, Google has massive
9:24
distributional advantages. They
9:26
incorporate Gemini into Google search.
9:28
People get used to Gemini as they
9:29
interact with it on Google search. Some
9:31
of them will go on to download the
9:33
Gemini app. Google also makes Gemini
9:35
integrate seamlessly with their other
9:37
products like Google Docs and Google
9:38
Sheets which are already used by
9:40
hundreds of millions of people. Finally,
9:42
Google pays huge amounts of money to
9:44
smartphone makers including Samsung to
9:46
have Gemini pre-installed.
9:48
Open AAI had a first mover advantage
9:50
allowing Jad GBT to gain dominant market
9:52
share. But in the long term, it should
9:54
be expected that Google with its massive
9:56
advantages will become the dominant
9:58
player in generative AI. Open AAI
10:00
currently loses money on chat GPT. Their
10:03
computational costs are so high that
10:05
they lose money even on the paid chat
10:06
GPT plus and pro subscriptions which
10:08
cost 20 and $200 per month respectively.
10:12
And they obviously lose money on free
10:13
users who generate zero revenue.
10:16
According to the information, only 5% of
10:18
ChatGBT users pay for plus or pro
10:20
subscriptions. So, OpenAI generates zero
10:22
revenue from 95% of its users. To
10:25
address this imbalance, OpenAI has long
10:27
planned to introduce advertising inside
10:29
Chat GBT as a way to monetize free users
10:32
and has reportedly had engineers working
10:33
on building an ad platform. However,
10:36
when Sam Alman declared a companywide
10:37
code red, those resources were
10:39
reportedly pulled away from advertising
10:41
and redirected towards improving Chad
10:42
GBT's core capabilities in response to
10:44
competitive pressure from Gemini. The
10:47
result is a strategic trade-off. Even if
10:49
OpenAI succeeds in closing the technical
10:51
gap with its competitors, it is pushed
10:53
further into the future any credible
10:54
plan to monetize the vast majority of
10:56
its users.
10:58
This begs questions about how OpenAI
11:00
plans to generate enough revenue to
11:01
cover its $ 1.4 trillion of data center
11:04
spending commitments. Alman claims the
11:06
company has an annual revenue run rate
11:07
of $20 billion. This isn't nearly enough
11:10
to cover their ongoing compute costs,
11:12
let alone their massive data center
11:14
expansion plans. On October 30th, Sam
11:17
Alman appeared on a podcast where he
11:19
attempted to explain OpenAI's plans to
11:20
increase its revenue.
11:22
We do plan for revenue to grow steeply.
11:24
Revenue is growing steeply. We are
11:26
taking a forward bet that it's going to
11:28
continue to grow grow and that not only
11:30
will chache keep growing but we will be
11:33
able to become one of the important AI
11:35
clouds that our consumer device business
11:38
will be a significant and important
11:40
thing that AI that can automate science
11:43
will create huge value. So man says he
11:47
expects open AAI to increase its revenue
11:49
with three new business lines. Firstly,
11:51
OpenAI will create an AI cloud. It
11:53
sounds like he wants OpenAI to build its
11:55
own data centers and sell computing
11:57
power to other AI companies. I have no
11:59
idea how he intends to do this. Open AAI
12:02
does not own or operate its own data
12:04
centers. They purchase computing power
12:05
from cloud service providers such as
12:07
Microsoft, Oracle, and Coreeave. Open
12:09
AAI's $1.4 trillion of current spending
12:12
commitments is not to build its own data
12:14
centers. The new data centers are being
12:16
built by the likes of Oracle and
12:17
Coreeave. OpenAI just promised to
12:19
purchase compute from them in the
12:21
future. I honestly have no idea what
12:23
Alman means when he says that OpenAI
12:25
will become one of the important AI
12:26
clouds. Maybe they're going to purchase
12:28
compute from hyperscalers and resell it
12:30
to other AI companies as a middleman.
12:32
Secondly, OpenAI plans to develop and
12:34
sell its own consumer hardware device.
Consumer Device
12:37
In May of 2025, OpenAI paid $6.4 billion
12:41
to acquire a company started by former
12:42
Apple chief designer Johnny IV. OpenAI
12:45
wants to use Iive's expertise to build a
12:47
new AI consumer device. It's supposed to
12:50
be some Amazon Echo type thing that can
12:51
fit in the palm of your hand. It will
12:53
have cameras, microphones, and speakers
12:55
on it. You can talk to it and ask it
12:57
questions. AI devices have been tried
13:00
before. For example, in 2023, a startup
13:03
called Humane AI unveiled an AI pin that
13:05
clips onto your shirt. The co-founders
13:07
of Humane previously worked as designers
13:09
at Apple just like Johnny IV. Alman was
13:12
a big believer in Humane. He personally
13:14
invested into one of Humane's private
13:16
funding rounds. Humane started selling
13:18
its AI pin in early 2024. It received
13:20
extremely poor customer reviews and
13:22
generated minuscule sales. The company
13:25
shut down in early 2025. The Humane AI
13:28
pin was basically just a more expensive
13:29
and less reliable version of the Amazon
13:31
Echo. We actually made a video about
13:33
Humane last year, which is linked in the
13:35
description below.
13:37
Given that you can already use Chat GPT
13:39
or any other AI model on your phone or
13:41
computer, I see no reason why anybody
13:43
would want a dedicated AI device. The
13:46
humane AI pain was an idiotic idea that
13:48
was doomed to fail from the beginning,
13:50
but Sam Alman thought it was a good idea
13:52
and invested his own money into it. The
13:54
development of OpenAI's consumer device
13:56
is reportedly not going very well.
13:58
They'll probably eventually make
13:59
something similar to the Humane AI pin
14:01
and it will be a commercial failure.
Automating Science
14:04
Finally, Alman says OpenAI will create
14:06
tremendous economic value by automating
14:08
science. Alman has long made grandiose
14:10
statements about the potential for AI to
14:12
make scientific discoveries. For
14:14
example, in a recent blog post, he said
14:16
that if OpenAI had 10 gawatts of
14:18
computing power, they could figure out
14:20
how to cure cancer. To date, neither
14:22
Chat GPT nor any other AI model has made
14:24
any significant scientific discoveries.
14:27
I'm sure that many scientists use AI
14:29
tools, including ChatGpt, to do some of
14:31
their grunt work, like crunching data.
14:33
Maybe this can speed up their scientific
14:35
discoveries, but I don't see how OpenAI
14:37
can generate huge revenue from this. A
14:39
scientist pays the same subscription
14:41
cost for Chad GBT as anyone else. As an
14:44
analogy, scientists use laptop
14:45
computers. They are far more productive
14:48
with a computer than they would be
14:49
without one, but they still pay the same
14:51
normal retail price to buy the computer.
14:53
If a scientist makes a great scientific
14:55
discovery, this doesn't benefit the
14:57
laptop manufacturer. Even if Chat GBT
15:00
becomes useful to scientists, I fail to
15:02
see how this will bring in a large
15:03
amount of revenue to OpenAI. Open AAI
15:05
claims that Chad GBT 5.2 2 is marginally
15:08
better at doing some science tests as
15:10
compared to Gemini 3. But just because
15:12
you can memorize a bunch of answers to
15:14
science tests doesn't mean you can make
15:15
new discoveries. The intelligence of
15:17
large language models is constrained by
15:19
their training data. It may not even be
15:21
possible for them to make novel
15:23
discoveries. When Sam Alman talks about
15:25
automating science and curing cancer,
15:27
this is an artifist to fool the
15:28
gullible. He makes grandiose predictions
15:30
to justify his massive spending
15:32
commitments, but there's no proof that
15:34
any of this is even possible.
Desperate for Revenue
15:37
Alman claims he can eventually cure
15:38
cancer with AI, but in the meantime,
15:40
he's resorting to some far less
15:42
glamorous initiatives to bring in
15:43
revenue. In October, Alman said they
15:46
will soon roll out an age verification
15:47
feature for ChatGpt. Adult users will
15:50
then be allowed to use ChatGpt to make
15:52
erotica. Maybe we'll see an erotic
15:54
cartoon series about Sam Alman, the
15:56
firefighter.
15:57
Altman also has a plan to start
15:59
monetizing the Sora video generation
16:01
app, which has to date been a cash
16:02
incinerator. In early December, they
16:04
announced a partnership with Disney.
16:06
Disney agreed to license 200 of its
16:08
characters, including Marvel and Star
16:10
Wars characters, to be used on Sora.
16:12
Open AI incurs astronomical
16:14
computational costs to operate Sora and
16:16
to date is believed to generate
16:18
minuscule revenue. Starting in early
16:20
2026, Sora users will be able to pay
16:23
real money for the privilege of
16:24
generating AI slop videos with Disney
16:26
characters. For example, you can make a
16:28
fake video of yourself as a Jedi. Disney
16:30
will get a cut of this revenue. I would
16:33
be shocked if a significant number of
16:34
people would be willing to pay real
16:36
money for this. I expect the Disney Open
16:38
AAI partnership to generate minuscule
16:40
revenue.
Contagion
16:42
Open AAI needs to increase its annual
16:43
revenue by more than 10fold if it is to
16:45
pay for its massive data center spending
16:47
commitments. Now that they're losing
16:49
market share to Google, this outcome
16:50
seems increasingly remote. We're already
16:53
starting to see fear spread across the
16:54
market. In September, OpenAI signed a
16:57
$300 billion cloud computing deal with
16:59
Oracle. Starting in 2027, OpenAI will
17:02
need to pay Oracle $60 billion per year
17:04
for cloud computing services. Oracle
17:06
needs to spend hundreds of billions of
17:08
dollars to build these data centers
17:09
before OpenAI starts paying them. Oracle
17:12
doesn't have enough cash to fund this
17:13
construction on its own, so they're
17:15
borrowing money from various banks and
17:16
other financial institutions. One of the
17:19
lenders is a private equity firm called
17:20
Blue Owl Capital. Blue Owl already
17:23
agreed to fund one of Oracle's data
17:24
centers in Texas. They were in talks to
17:26
provide an additional $10 billion to
17:28
fund another data center in Michigan,
17:30
which will also be leased to OpenAI. On
17:32
December 17th, it was reported that Blue
17:34
Owl pulled out of the deal. They were
17:36
probably concerned about OpenAI's
17:38
ability to make good on it spending
17:39
commitments. If OpenAI can't make good
17:42
on its payments to Oracle, it's unclear
17:44
how Oracle will be able to pay its
17:45
massive data center loans. Credit
17:47
default swaps on Oracle's bonds have
17:49
recently surged to over 100 basis
17:51
points. Investors are now betting that
17:53
there's a 1% chance that Oracle will
17:54
default on its debt within the next 5
17:56
years. While 1% doesn't sound like a
17:59
lot, that's pretty high for a company as
18:00
massive as Oracle.
18:02
Open AAI itself continues to incur
18:04
operating losses. They've raised $60
18:07
billion of venture capital funding
18:08
today. It seems like they've already
18:10
burned through most of this. The company
18:12
is currently trying to raise $100
18:14
billion at an eyepopping $830 billion
18:17
valuation. They want to do this funding
18:19
round in the first quarter of 2026. $100
18:22
billion is an absurd amount of money to
18:24
raise in a private funding round. It
18:26
would be the largest funding round in
18:27
history. I don't know what investors
18:29
even have this kind of money. And why
18:31
would any investor pay such an absurd
18:33
valuation for a company that has $20
18:35
billion of revenue, incurs massive
18:37
operating losses, and is losing market
18:39
share to Google? Over the coming months,
18:41
we'll be watching very carefully what
18:43
happens with this $100 billion funding
18:45
round. If this funding round fails, it
18:47
will mark the beginning of the end for
18:48
Open AAI. They will default in their
18:50
spending commitments and the entire
18:52
House of Cards will come crumbling down.
18:55
All right, guys, that wraps it up for
18:56
this video. What do you think about
18:58
OpenAI's code red? Let us know in the
19:00
comments section below. As always, thank
19:03
you so much for watching and we'll see
19:04
you in the next one.