Latest / In the Loop AI / OpenAI's Existential Crisis: Are GPT 5.2 and a $1B Disney Deal Enough?

Transcript

OpenAI's Existential Crisis: Are GPT 5.2 and a $1B Disney Deal Enough? · In the Loop AI

0:00
Ejaaz: OpenAI just shipped GPT 5.2, the new best model in the world.
0:05
Ejaaz: And yet, why does it feel like they're still fighting for survival?
0:09
Ejaaz: 80% of Vibe coding is done using Claude Anthropics model, and more people are
0:15
Ejaaz: using Google's Gemini models versus ChatGPT.
0:18
Ejaaz: So it begs the question, what's OpenAI even useful for?
0:22
Ejaaz: In fact, this year alone, they're projected to lose $20 billion and only projecting
0:27
Ejaaz: to be profitable by 2030. That's five years from now. A lot can happen.
0:31
Ejaaz: And on top of all of this, they have a $1.4 trillion bill to pay on compute and data center costs.
0:38
Ejaaz: So whichever way you look at it, the walls seem to be closing in on OpenAI.
0:43
Ejaaz: And Sam is feeling the pressure. He called in a code red two weeks ago,
0:47
Ejaaz: which companies only really do when they're in crisis mode.
0:50
Ejaaz: But all might not be lost.
0:53
Ejaaz: ChatGPT is still the number one AI app in the entire world. And just last week,
0:58
Ejaaz: they signed a $1 billion deal with Disney to license their IP to create AI-generated video.
1:04
Ejaaz: And rumors are amongst the grapevines right now saying that OpenAI may be IPO-ing
1:10
Ejaaz: next year at a $1.2 trillion valuation, which will give them the much-needed
1:14
Ejaaz: capital that they need to win.
1:17
Josh: Yeah, in 2023 and 2024, the world was basically like, ChatGPT equals AI.
1:22
Josh: If you're using AI, you're using ChatGPT.
1:24
Josh: In fact, most people off the street If you ask them what AI was,
1:28
Josh: they would tell you it's ChatGPT.
1:29
Josh: There is a complete and total domination of the marketplace by this company.
1:35
Josh: Up until this year this year the question has become
1:37
Josh: not chat gpt does equal ai but like which ai
1:40
Josh: do i use for which task there's not one ai that
1:43
Josh: does everything anymore because they've kind of siloed themselves into these
1:46
Josh: individual buckets where if you want to write code you don't go to chat gpt you
1:49
Josh: go to anthropic if you want to create images you go to google and chat
1:53
Josh: gpt is no longer the all-in-one app it's
1:56
Josh: a place where you go to solve a specific set of problems and like sticking with
2:00
Josh: the war analogy because this is the game of thrones it's like open ai used to
2:04
Josh: have the only aircraft carrier in town everyone had to come through them to
2:07
Josh: use them but now everyone has an aircraft carrier and the differentiator really
2:11
Josh: becomes like supply chains alliances
2:14
Josh: distribution and your ability to build applications that create actual value
2:18
Josh: for users like us who want to use them so that takes us just to i guess 5.2
2:22
Josh: which is what we're going to talk about today that open ai just recently launched
2:26
Josh: and it should give us some hints as to where the company is going as they try
2:30
Josh: to fight this resisting force that's closing in around them from all ends.
2:34
Ejaaz: Yeah. So the resisting force comes in the form of Google's Gemini 3 Pro,
2:39
Ejaaz: which released two weeks ago, and then Anthropics called Opus 4.5,
2:43
Ejaaz: which is the frontier model when it comes to coding.
2:46
Ejaaz: So really, when OpenAI released this model, people were looking at these two
2:49
Ejaaz: competitors and seeing how it weighed up against it.
2:51
Ejaaz: And the answer, Josh, is really, really well. It technically beats Claude Opus
2:57
Ejaaz: 4.5 encoding and it beats Gemini 3 Pro against a range of different generalized AI activities.
3:02
Ejaaz: So technically, I think it is the best or number one model in the world.
3:06
Ejaaz: And the reason why I'm kind of like a little uncertain about this is I just
3:10
Ejaaz: feel like, to your point, they're gamifying towards benchmarks that don't really
3:14
Ejaaz: matter or don't really apply to the real world.
3:16
Ejaaz: But there is one particular benchmark that I thought was really cool that it
3:20
Ejaaz: excelled in, which was GDPVAL, which basically measures how GPT 5.2 performs
3:26
Ejaaz: in real-world expert-level tasks.
3:29
Ejaaz: So I'm talking about things that people do in work that professionalize in their work.
3:33
Ejaaz: And it was measured that in 70% of cases, Josh,
3:38
Ejaaz: human experts would choose GPT 5.2's output and answer and the way that it deals
3:43
Ejaaz: with that particular task versus an expert professional human that that's paid
3:48
Ejaaz: hundreds of thousands of dollars a year which which kind of blew my mind and
3:51
Ejaaz: that was the most valuable kind of takeaway um with gpt 5.2.
3:55
Josh: So chat gpt feels like it's stuck between a rock
3:58
Josh: and a hard place where they're optimizing for these vanity metrics but the
4:01
Josh: actual day-to-day experience for a lot of users like
4:04
Josh: myself and you are just not really changing much
4:07
Josh: the product has mostly stayed the same the models are getting better
4:10
Josh: but the only difference that i notice is that the
4:13
Josh: tick in the little top left corner goes from 5.0 to 5.1 and
4:16
Josh: now it says 5.2 but in terms of the actual outputs in terms
4:19
Josh: of my experience using the platform not a whole lot has changed in
4:22
Josh: addition to this they have tweaked the cost a little bit which is pretty pricey
4:25
Josh: it's um 21 per million tokens in 168 per million tokens out so you're paying
4:32
Josh: for a frontier model for sure and this is probably a result of them just using
4:36
Josh: a tremendous amount of compute to optimize for these benchmarks,
4:39
Josh: which is bizarre because Sam Allman came out and said, this is a 390x cost reduction
4:44
Josh: in a year, which is pretty remarkable and
4:49
Josh: You really have to take a step back and appreciate what they've been doing,
4:51
Josh: because a lot of the sentiment around OpenAI right now is fairly bearish in
4:56
Josh: the sense that they have a lot of hard challenges they need to overcome in order
4:59
Josh: to make this whole thing work.
5:01
Josh: But a 390 times cost reduction in anything in a year is pretty remarkable.
5:06
Josh: And that's relative to, I believe, 03, which was, if you could believe this,
5:10
Josh: it was released a year ago, which feels like an eternity ago.
5:14
Josh: But it was literally not even 12 months ago. And since then,
5:17
Josh: we've gotten, I mean, you could look at this Arc AGI chart, which is the benchmark
5:20
Josh: that kind of everyone uses to gauge where these companies are on the path to AGI.
5:25
Josh: They've made such an incredible improvement. And that token that used to cost
5:29
Josh: a dollar now costs, what is one four hundredth of a dollar, whatever that is,
5:33
Josh: just a fraction of the cost.
5:35
Josh: And these, I mean, these changes are remarkable. So even though we're being
5:39
Josh: particularly bearish on OpenAI, you do have to admire the fact that this industry
5:43
Josh: is advancing so quickly.
5:44
Josh: And what cost $400 yesterday now cost $1 today for something that is far higher quality.
5:51
Ejaaz: All of that being said, Josh, I am not entirely sure that models are kind of
5:56
Ejaaz: like the thing to optimize for, at least from a benchmarks perspective.
5:59
Ejaaz: I think that OpenAI should be focused on one thing and one thing only that they
6:03
Ejaaz: do really well, which is the consumer product mode.
6:07
Josh: And we have a deal based on this that has come into play over the last week,
6:11
Josh: which is the Disney deal, the $1 billion deal that they inked with Disney in
6:15
Josh: order to hopefully, I guess, make a little bit more of a mainstream splash.
6:18
Josh: I guess the way you could describe it is if you can imagine Disney as a sneaker
6:22
Josh: brand and Sora is the machine that could generate these unlimited custom shoes.
6:26
Josh: Until now, people were basically making these fake Nikes. And this is a big problem.
6:30
Josh: There's a lot of these fake products that are reflective of the real thing,
6:33
Josh: but lack in the quality. They're messy.
6:35
Josh: They're legally very sketchy. and then disney is finally coming
6:38
Josh: and they're kind of capitulating they're like okay we're going
6:41
Josh: to officially license the designs but we're going to set the rules and
6:43
Josh: we're going to make sure that we get paid so our brand doesn't get wrecked
6:46
Josh: so this very much feels to me like a hedge against the
6:49
Josh: destruction that was already starting so now fans
6:52
Josh: can go and they could generate short videos using disney characters or marvel
6:55
Josh: star wars whatever it is and then disney gets the opportunity to skin that the
7:00
Josh: way they like put the guardrails in place and then disney also even offers plans
7:04
Josh: to stream curated clips on disney plus so they're actually integrating this
7:08
Josh: into their ecosystem so what you're seeing here
7:11
Josh: And it is really funny. It's like they're letting you play with the characters,
7:14
Josh: but not steal the actor's identity.
7:16
Josh: And this feels like a 180 from where they've stood previously,
7:18
Josh: because you just from what I understand, they're simultaneously suing Google
7:22
Josh: over like copyright infringement stuff. So they're saying, no, no, no, no.
7:26
Josh: If we're going to play this game, we're going to play this on our terms.
7:29
Josh: And they committed at least a billion
7:30
Josh: dollars to open AI to form this relationship and to start doing it.
7:34
Josh: And to me, it feels interesting. And on the screen, we have a bear case,
7:36
Josh: too, where it seems like they have just totally capitulated.
7:40
Ejaaz: Yeah, effectively, what's being described here is, you know,
7:44
Ejaaz: Disney, in his own words,
7:46
Ejaaz: Disney is a $200 billion company sitting on only $5 billion worth of cash and
7:50
Ejaaz: a $1 billion investment at a $500 billion valuation of OpenAI doesn't matter
7:54
Ejaaz: to them at all on the upside case.
7:57
Ejaaz: And so if OpenAI is going to IPO at a $1 trillion valuation, that's just a 2x.
8:01
Ejaaz: So it doesn't make any kind of meaningful dent or impact on Disney's profit line.
8:06
Ejaaz: Where I think the real value is coming in here, Josh, is long-term.
8:11
Ejaaz: If you can have endless amounts of content generation for Disney's IP without
8:17
Ejaaz: Disney needing to spend money on really expensive animators and teams to produce the content.
8:23
Ejaaz: And, you know, animation takes like years to produce. Open Air can generate
8:26
Ejaaz: it in a couple of minutes.
8:28
Ejaaz: That is, it's kind of like free marketing. It's kind of like free PR and the
8:32
Ejaaz: fans can make of it what they will without kind of Disney having to get their
8:35
Ejaaz: hands dirty. So it kind of makes their IP endless.
8:39
Ejaaz: And if Disney ends up owning all of it, then great.
8:41
Ejaaz: And in fact, as part of this deal, if I'm not mistaken, OpenAI gets exclusive
8:46
Ejaaz: rights on the IP for a full year before Disney kind of reclaims IP access or
8:53
Ejaaz: IP licensing on the AI generated content.
8:56
Ejaaz: So they really see that OpenAI has some kind of advantage here and they were
8:59
Ejaaz: able to strike a pretty sick deal.
9:01
Josh: Yeah. And the big boundary line is, is you can use the characters,
9:03
Josh: but not the real performers, forces, faces or voices.
9:07
Josh: So for me, this is exciting because I mean, imagine you were a kid.
9:10
Josh: I always watched Disney Channel. I loved Disney Channel.
9:12
Josh: And there's a world in which you start to get a custom curated Disney Channel
9:15
Josh: just for you, where you have the characters, you have a universe,
9:18
Josh: but it's all hyper tailored to the stuff that you are most interested in.
9:21
Josh: And that seems exciting. It seems directionally
9:25
Josh: correct in the sense that we're going to start leaning more into hyper
9:28
Josh: custom content per person and disney's kind of leaning into that so props to
9:33
Josh: them but that doesn't end our ai our open ai concerns for this week there there
9:38
Josh: is more so as we were going through the financials we so we started to look
9:42
Josh: in at them and they're losing a lot of money
9:46
Ejaaz: How much to the tune of 20 billion dollars 20 billion dollars yeah yeah.
9:50
Josh: Okay so they got 1 billion from disney so yeah they're exactly 120th of the way there
9:54
Ejaaz: Right exactly and okay to give
9:57
Ejaaz: them a few flowers 13 billion dollar annual recurring revenue is great right
10:02
Ejaaz: but they're projected to be spending twice if not three times that amount of
10:07
Ejaaz: money next year on data center costs alone josh so then you might be like well
10:12
Ejaaz: they have paid subscribers right like so they should be fine.
10:15
Ejaaz: Well, 5% of the 800 million weekly active users that they have actually pay. So that's 40 million.
10:24
Ejaaz: And of that percentage, I guess quite a large percentage of them are only the
10:27
Ejaaz: 20 bucks a month subscribers.
10:29
Ejaaz: So the point I'm trying to make here is the subscription model for OpenAI does not make sense.
10:35
Ejaaz: In a world where it needs to make sense, I would guess they need to turn on advertisements.
10:40
Ejaaz: As soon as they do that, in my opinion, it's going to make the product experience
10:43
Ejaaz: even worse, which will push people towards or back towards Google and Claude
10:49
Ejaaz: to kind of use their models for free.
10:51
Ejaaz: And Google has an endless amount of cash flow so that they don't really care about this, right?
10:55
Ejaaz: Another crazy number is how much of their revenue they're losing to Anthropic.
10:59
Ejaaz: This year, Anthropic unseated them as the top enterprise AI model that is favored
11:04
Ejaaz: by enterprises globally, a crown which OpenAI held for like a number of years prior to this.
11:10
Ejaaz: So they're losing on that front as well. And then when it comes to just like
11:14
Ejaaz: generally projecting their revenue, Josh, something that a number that they've
11:18
Ejaaz: floated is, you know, by 2030, we'll have $200 billion worth of revenue.
11:22
Ejaaz: But what people aren't contemplating
11:24
Ejaaz: is how much they're still measured to be losing in that same year,
11:28
Ejaaz: which is around the same amount, if not a little bit more.
11:31
Ejaaz: And of course, like these bar charts and stuff are all just estimates.
11:34
Ejaaz: No one knows how this will actually end up transpiring, but it's worthy to think about.
11:38
Ejaaz: But the final one is the $1.4 trillion worth of compute that they're projected
11:45
Ejaaz: to spend over the next five years.
11:47
Ejaaz: And I actually want to come in here with a positive note, Josh,
11:50
Ejaaz: because this is the biggest kind of debt on their balance sheet that people
11:54
Ejaaz: kind of like bear about, right? They're saying like, hey, this is a really bad thing.
11:58
Ejaaz: OpenAI is not going to be worth a lot of money because of this.
12:01
Ejaaz: It's just a number. And it's just a clickbait number because it's not real.
12:06
Ejaaz: If you look into the contracts and the terms that they've signed with all these
12:09
Ejaaz: cloud providers, they don't actually need to spend this amount of money.
12:14
Ejaaz: It's just the most amount of money that they could potentially spend with these
12:18
Ejaaz: cloud providers, with these compute providers.
12:20
Ejaaz: So realistically, if you look at the contractual deals for next year,
12:23
Ejaaz: OpenAI just needs to spend up to $10 billion, which is a lot more feasible than
12:29
Ejaaz: the $100 billion, which is contractually apparently all headlined to be for next year.
12:35
Ejaaz: So there's a bit of fear mongering on social media about all of this stuff.
12:39
Ejaaz: And I actually don't think the bear case is all that big for OpenAI.
12:42
Ejaaz: In fact, I think it's quite bullish.
12:44
Josh: Yeah, here's the thing. It's like with episodes like this, it's fun to talk about the drama.
12:48
Josh: It's fun to talk about the downswings. These are subtle nuanced changes that
12:52
Josh: are happening on a much longer, smoother trajectory, which is up and to the right.
12:56
Josh: And the reality is that OpenAI is one of the most remarkable companies on earth.
13:00
Josh: They are shipping the best AI in the world. And it's fun to nitpick.
13:05
Josh: It's fun to pick at the headlines.
13:06
Josh: But the reality is that these companies are printing tons of cash.
13:11
Josh: They're spending more. So there are real concerns. But directionally,
13:15
Josh: what we're seeing is this unbelievable progress,
13:18
Josh: like 930 times cost reduction for a leading edge token over 12 months is so
13:25
Josh: huge, especially considering that is one of many compounding factors.
13:29
Josh: There's the cost per token, but there's also...
13:31
Josh: Quality per token. And there is the cost per GPU, which has gone much lower.
13:35
Josh: And there's the product now that is significantly better than it was a year ago.
13:39
Josh: And we're getting all these compounding growth things that...
13:42
Josh: I'm watching this funny video on the screen.
13:44
Ejaaz: What is this? Is this from Sora? It just shows you what they're good at, Josh.
13:49
Ejaaz: It puts a smile on your face. People actually want to use and consume these products.
13:52
Ejaaz: And I feel like people are focused too much on the benchmarks.
13:55
Ejaaz: They're focused too much on the numbers.
13:56
Ejaaz: But the fact is, if people want to use your product, you'll end up being a pretty successful company.
14:02
Josh: You're watching this really funny video of a snowman snowboarding.
14:05
Josh: And this was not possible even a year ago because video generation has gotten much better.
14:09
Josh: And just last week, three days ago, actually, it was OpenAI's 10-year anniversary.
14:14
Josh: They've been around for a decade, which is pretty amazing because up until when? 2021?
14:19
Josh: AI didn't even really exist in the consumer marketplace. So for six of those 10 years,
14:23
Josh: nobody cared about open ai and then
14:26
Josh: they released chat gpt 3.0 and that
14:29
Josh: was the firing gun and the race started and now
14:33
Josh: in the next four years the world has changed forever and open ai is
14:36
Josh: very much responsible for that google had designed the
14:39
Josh: transformer they had the architecture they did nothing with it for a really long time until
14:42
Josh: open ai came around and started the race and is
14:45
Josh: now leading the race and will continue to be a prominent player where they stand
14:49
Josh: on the scale of things i don't know their 90 of the pie they had is certainly
14:53
Josh: going to go down but even if it drops down to 30 30 of this pie is an enormous
14:58
Josh: number and it leaves me just feeling excited and optimistic i mean the other wild card too
15:03
Josh: is that they have the hardware device incoming and no one else is really building
15:07
Josh: a dedicated hardware device with the world-class hardware designer.
15:10
Josh: So that leaves something really excited to look forward to next year.
15:13
Josh: But I think in terms of progress for this year, OpenAI had an unbelievable year
15:19
Josh: and it's easy to talk down on it because relative to others,
15:22
Josh: yeah, sure, they're losing some market share, but man, what a great company. What a great model.
15:25
Ejaaz: And that's our episode for today, folks. A lot of you have been commenting about
15:31
Ejaaz: giving our opinion on OpenAI, the bull case, the bear case, and the fact that
15:35
Ejaaz: a lot of these AI companies are going public next year.
15:37
Ejaaz: You know, a lot of you from the investing angle are super curious.
15:39
Ejaaz: We will continue to be putting out our thoughts and opinions and putting out
15:43
Ejaaz: the facts as much as we can. We hope you enjoyed today's episode.
15:47
Ejaaz: If you enjoyed it, and if you enjoyed all of our other episodes,
15:49
Ejaaz: we put out three or four, maybe even last week, Josh.
15:52
Ejaaz: Go check them out, subscribe, put on notifications, give us a thumbs up,
15:57
Ejaaz: give us a rating if you enjoy our content, and we will see you on the next one.
16:02
Josh: See you guys.