Limitless: An AI Podcast

Ejaaz:
A lot of people have written off OpenAI, but they might finally be turning things back around.

Ejaaz:
They just raised $122 billion, the largest ever private funding round.

Ejaaz:
They purchased TBPN for $200 million plus.

Ejaaz:
They cut out Sora. They're reconsolidating every single resource into one singular super app.

Ejaaz:
But there's tension amongst the rankings. Their CFO, Sarah Fryer,

Ejaaz:
is reportedly being frozen out ahead of a large IPO.

Ejaaz:
Their CMO just stepped down. And there's a lot of tension around whether OpenAI

Ejaaz:
will regain the throne from Anthropic.

Ejaaz:
And that's what we're going to be digging into in today's episode.

Ejaaz:
We think that OpenAI is undergoing a pivot that a lot of people are currently missing.

Josh:
This is the comeback story. And I think it's told in three parts, right?

Josh:
It's like we start with, we have this graveyard of things that have gone wrong

Josh:
that led people to believe that open AI is just kind of getting left behind

Josh:
by Anthropic and market shares reflected this.

Josh:
But then we have all the things that are actually going on behind the scenes,

Josh:
followed by this conclusion, their magnum opus, which is leading into the IPO

Josh:
and all of the things they're doing behind the scenes that maybe you're not

Josh:
privy to or recognizing.

Josh:
And starting with Stargate, which may be the biggest failure.

Josh:
I mean, the headline number was $500 billion for a data center.

Josh:
And if you remember, Sam was at the White House with Masayoshi from SoftBank

Josh:
and a few other people announcing half a trillion dollars for these data centers to exist.

Josh:
OpenAI has since scrapped the 600 megawatt Abilene extension because it actually

Josh:
would have installed GPUs at a site where the power grid wouldn't be ready until

Josh:
the next gen chips are available.

Josh:
So they've been having a lot of logistical issues. There was another problem

Josh:
that they had where the winter weather in Abilene, where they're building these

Josh:
data centers, literally knocked the liquid cooling offline for multiple days,

Josh:
straining the relationship.

Josh:
They've been having a bunch of issues and the competitors are kind of coming in to sweep up the mess.

Josh:
And it appears as if OpenAI is kind of shifting the strategy from building these

Josh:
gigantic data centers to actually just deferring a lot of the money that they

Josh:
raise into cloud infrastructure and just borrowing the compute from other people.

Josh:
So this is the first sign of things that weren't going so well.

Josh:
Grand ambitions signed at the White House didn't really convert into everything

Josh:
that they hoped and wanted for.

Ejaaz:
You could also argue that OpenAI were the singular prop for the entire GPU economy, right?

Ejaaz:
They started this whole circular economy by signing all these partnerships and deals.

Ejaaz:
But it wasn't just GPUs. It was also the memory supply, which is required to build the GPU.

Ejaaz:
So every single layer of the picks and shovels, OpenAI was involved.

Ejaaz:
They committed to purchasing, I think, around 40% of the global memory supply,

Ejaaz:
and they defaulted on that in the last two weeks.

Ejaaz:
So a lot of this stuff is starting to implode, and people are fearful of OpenAI

Ejaaz:
success based off of that. But it's not just the infrastructure side of things.

Ejaaz:
We have to talk about the competition between the other major competitive for

Ejaaz:
them, which is Anthropic.

Ejaaz:
And Anthropic, and they've been cooking in multiple ways, the primary one being enterprise.

Ejaaz:
Now, if you looked at this chart about a year and a half ago,

Ejaaz:
it would be the inverse. OpenAI had the dominance in the enterprise market share.

Ejaaz:
Everyone in Fortune 500 was using ChatGPT. And at some point in the last year

Ejaaz:
and a half, that flipped pretty aggressively.

Ejaaz:
People became really fond of coding AI. It started becoming productive within their own business.

Ejaaz:
And Anthropic took the lead. They now have a 73% market dominance on first-time

Ejaaz:
enterprise usage. And it's not just enterprise usage.

Ejaaz:
Anthropic has also flipped their own trend on retail.

Ejaaz:
Recently, they were adding around a million users or a million signups per day.

Ejaaz:
And this came after the Pentagon fallout where Anthropic stood their ground.

Ejaaz:
And a lot of retail users really admired that.

Ejaaz:
So there is this shuffle or mix that has happened over the last year and a half

Ejaaz:
where OpenAI kind of lost that gleam, that number one throne,

Ejaaz:
which Sam had held on to for the first couple of years of open AI.

Ejaaz:
And there's real competition in the play now.

Josh:
Well, for a while, ChatGPT was synonymous with AI for a lot of people.

Josh:
It's like people thought that ChatGPT was AI. It was the entirety of AI.

Josh:
And what Anthropocistan has kind of punctured a hole in that perception and

Josh:
saying that, hey, actually, AI is multiple companies.

Josh:
In fact, Anthropoc is another player. And a lot of people have switched over.

Josh:
Now, this very much feels like, what is it? Maybe the second inning,

Josh:
The first inning was dominated by Chad Shibuti. The second one now is showing some competition.

Josh:
There are some more elements in this graveyard that we need to get through before

Josh:
we can get to the good stuff.

Josh:
Sora being the next biggest one. I mean, Sora was costing allegedly $15 million

Josh:
to run per day with no revenue, and they signed a huge deal with Disney.

Josh:
Now, Disney didn't know that their deal was getting cut until the rest of the

Josh:
world did when Sora came out and announced that they were actually cutting the app.

Josh:
Now, this concerned some people. For me, it was this very exciting thing because Sora...

Josh:
I mean, it was the AI slop app, right? Like it's like TikTok, except for AI only.

Josh:
It was a totally different market than what it seems like they're moving towards

Josh:
now. Like we just mentioned, the enterprise market is a huge market that they're going for.

Josh:
And Sora has absolutely nothing to do with that. So although Sora getting closed

Josh:
down could be perceived as a positive or a negative, there is one last thing

Josh:
in the graveyard that is kind of just purely a negative and it has to do with

Josh:
the leadership shakeup that they're having right now.

Ejaaz:
Yeah, so if we look at OpenAI's products in general, they've been losing favor,

Ejaaz:
but the team is what binds the company together. So they're the people who are

Ejaaz:
building the products together.

Ejaaz:
Even they have been suffering a bunch of reshuffling and tension amongst the ranks.

Ejaaz:
So a few things happened over the last two weeks. Number one,

Ejaaz:
OpenAI's CEO of Applications, Fiji Simo. So not to be confused with Sam Altman himself.

Ejaaz:
She has taken a leaf of medical absence for the next couple of weeks.

Ejaaz:
There've been rumors for a while now that Fiji has been working remotely from California.

Ejaaz:
And so like the fact that she's not gonna be able to have focus on this next

Ejaaz:
stage of open AI, kind of suggests that maybe Sam needs to step in or they're

Ejaaz:
under pressure from constrained resources to be able to pull off and execute

Ejaaz:
their vision of securing AGI by the end of this year.

Ejaaz:
There's also Sarah Fryer, the CFO, who is reportedly being frozen out from IPO talks.

Ejaaz:
Apparently, she disagrees with Sam's valuation. Apparently, she thinks it's

Ejaaz:
too extreme and Sam doesn't really like that.

Ejaaz:
Then you have the CMO, Kate Rauch, also stepping down from her position.

Ejaaz:
So I understand that there's a period of consolidation of power under OpenAI.

Ejaaz:
I know that they called it Code Red about four months ago. So it's to be seen

Ejaaz:
whether this is playing off.

Ejaaz:
And I think the argument that we want to make here, Josh, on this episode is

Ejaaz:
that I think the pivot is working and a lot of people are missing that.

Josh:
The CFO news is really exciting. And if you stick around, we'll get into the

Josh:
CFO OpenAI IPO confrontation that's currently happening because I think that's

Josh:
probably the biggest part of the story.

Josh:
But this is where we go behind the scenes and kind of figure out what has been

Josh:
working. And I think that's the biggest part of the story. And I think that's the biggest part of the

Josh:
I think a lot of people don't really, they see the headlines that Sora got killed,

Josh:
but they're not exactly sure why and the reasoning of it being actually something positive.

Josh:
Now, Sora compute was a lot. And as we know, because their data centers haven't

Josh:
been going online as fast as they needed to, OpenAI is kind of constrained for compute right now.

Josh:
They have a very limited amount and a lot of that compute has to go to serving

Josh:
users like us, but a lot of it also has to go to researchers for doing research

Josh:
and also training in order to train these new models.

Josh:
Now by killing Sora it frees up quite a bit

Josh:
of GPUs because generating video consumes a lot of

Josh:
compute to possibly help them train

Josh:
for this new model now this new model is codenamed Spud which

Josh:
is possibly GPT 5.5 or GPT 6 and Sam Altman confirmed that pre-training actually

Josh:
completed on March 24th and Greg Brockman said this is the result of two years

Josh:
of research and it seems like this is their perhaps not AGI moment but this

Josh:
is the really big model they've been training for for a long time.

Josh:
And we've started to see the leaks coming out of what the outputs of this model can actually do.

Ejaaz:
Yeah, so a lot of things have been leaked out of OpenAI over the last week.

Ejaaz:
And we're not entirely sure if what you're seeing on the screen now,

Ejaaz:
their new ChatGPT Image 2 model is from the new Spark model that you just referenced

Ejaaz:
here, Josh, but it's pretty damn good. It's so good.

Ejaaz:
Yeah, like look at some of these examples, like what you're seeing on the screen

Ejaaz:
here is four different examples of four different types of medium that you can generate as an image.

Ejaaz:
So in this example, you've got the human anatomy in like incredible detail here.

Ejaaz:
And it's all accurate. Like you can zoom in. This is such a basic thing.

Ejaaz:
But a year ago, it couldn't do this. It couldn't spell English words correctly, let alone Latin words.

Ejaaz:
And it is absolutely nailing it in this example.

Ejaaz:
And then the one over here, it's nailed like the entire global map.

Ejaaz:
And then we have an example of like a real maybe photo of what is this Bath

Ejaaz:
and Body Works, where it says, Sorry, we're closed for the evening. See you tomorrow.

Ejaaz:
Now, a lot of these seem probably quite basic. We've got like a YouTube thumbnail graphic here as well.

Ejaaz:
In order to get the accuracy, spelling, like, look, this is not Mr.

Ejaaz:
Beast that you're seeing on your screen right here.

Ejaaz:
These aren't real videos that exist at all, has been incredibly hard.

Ejaaz:
But now if you were to scroll on X or any other social media platform and see

Ejaaz:
these images, you might be convinced immediately that that is real when really it's not.

Josh:
Yeah, whatever they're doing with this model, it seems like this is the moment

Josh:
in time in which we will no longer be able to tell what's air generated and what's not.

Josh:
Because if I saw any of these photos independently, I would have absolutely

Josh:
no clue that they were generated by AI.

Josh:
Down to the handwritten note that we're seeing on screen here,

Josh:
where it actually looks really handwritten, very much like a human wrote it.

Josh:
And that hasn't been the case with a lot of previous AI models.

Josh:
So whatever they have done to kind of cook this image gen model in the background

Josh:
has worked in a way that's very impressive and really powerful.

Josh:
And it's also important to note that Anthropic has no image gen capabilities

Josh:
at all. When it comes to anything outside of text.

Josh:
Anthropic just doesn't exist. They're not really a multimodal model at all.

Josh:
And it seems as if ChatGPT now is kind of making a move towards even the Google

Josh:
side of things when competing directly with Nano Banana Pro or Nano Banana Mini,

Josh:
both of which are unbelievable models.

Josh:
So this is really exciting to see. And I think it's an early step into their

Josh:
new strategy that they're leaning towards, which is the super app.

Josh:
And the super app, I think is probably the reason why a lot of people are switching

Josh:
over to Claude, because they have this unbelievable desktop and mobile app that

Josh:
has everything that you want under one roof.

Josh:
If you use Claude, you get access to like the anthropic models just to chat

Josh:
with you get access to co-work and you get access to coding along with all of

Josh:
the open claw functionality that they've recently built in like dispatch and

Josh:
we covered an episode last week.

Josh:
And I think that super app is a very powerful thing because the current chat

Josh:
GPT open AI ecosystem, it's very spread out. if you want to use Soar,

Josh:
you need to download an app, log into your account.

Josh:
If you want to use Codex and write code, you have to create a whole separate entity.

Josh:
You have to download a separate app. It doesn't all happen on the same application.

Josh:
And it appears as if OpenAI is officially doing the super app.

Josh:
And this is a really important pivot for the company.

Ejaaz:
Yeah. And the part I like most about this super app is their insane turnaround for coding AI.

Ejaaz:
Now I have to stress at the end

Ejaaz:
of last year, it was very clear what the number one coding AI model was.

Ejaaz:
It was Claude opus 4.6 from anthropic

Ejaaz:
and it was number one by a country mile

Ejaaz:
now open ai since then called a code red and

Ejaaz:
they consolidated all their compute resources all their research talent

Ejaaz:
to focus on two specific things building the best generalized llm and to upgrade

Ejaaz:
their coding model why because the argument was if you have the best coding

Ejaaz:
ai model you will basically win the ai race because you could use that coding

Ejaaz:
model to build every other single app that you want to including the next AI

Ejaaz:
models that you release.

Ejaaz:
And a lot of these companies, OpenAI and Anthropic, are currently using their

Ejaaz:
models to build the next model. So coding, very important.

Ejaaz:
Codex right now is amazing. Now, it's not seen that way amongst the average user who is vibe coding.

Ejaaz:
They still very much prefer Claude Code and Opus 4.6 because the personality

Ejaaz:
feels right. It sounds intuitive, it's easy to understand.

Ejaaz:
For those of you who are serious about software engineering,

Ejaaz:
a lot of the feedback has since been that Codex is a much better coding agent.

Ejaaz:
Now, that's not entirely true when it comes to spinning up multiple coding agents.

Ejaaz:
A lot of engineers have said Codex kind of loses its ability to coordinate amongst

Ejaaz:
themselves, and therefore they need to use Claude Opus 4.6.

Ejaaz:
But the turnaround that OpenAI has managed to achieve has been nothing short of amazing.

Ejaaz:
And I think Codex has around 2 million weekly active users, which was up from

Ejaaz:
like 100,000 at the start of Yeah, so very impressive turnaround.

Josh:
It's growing very quickly and it's a great product. I was building a little

Josh:
website just to test things out, comparing Codex directly with Opus 4.6 and

Josh:
Codex was faster, it was more effective, but Cloud was easier to use.

Josh:
It was much warmer, much more friendly and the design I found was just much

Josh:
better in general, the design sense.

Josh:
So they're both spiky in their own ways, but Codex is an amazing product for

Josh:
people who are hardcore coders and just want the absolute best tokens generated.

Josh:
Now, all of this to say that OpenAI is in a pretty good place and they there's

Josh:
been laying the foundation perhaps a little more quietly than they should have

Josh:
on the next step forward and that next step forward really is driven by this

Josh:
IPO and now the IPO has been a major point of contention because there is some

Josh:
conflict in the company but before we get into the conflict

Josh:
The new fundraising round is telling. EJ, this is probably the final fundraising

Josh:
round they're going to have that's private.

Josh:
I mean, the next one is going to be the full public offering.

Ejaaz:
And it's the largest ever private round. I want to show you this graphic,

Ejaaz:
Josh, just to put into context the scale of this race.

Josh:
Yeah, that's outrageous.

Ejaaz:
Prior to this, the company that had the single largest private round was also

Ejaaz:
OpenAI in 2025, where they raised around $40 billion. dollars.

Ejaaz:
This is almost 3x that at that point.

Ejaaz:
It is just a gargantuan round and puts things into perspective as to,

Ejaaz:
you know, people really believe that the scaling laws are intact.

Ejaaz:
What they mean by that is a lot of this round is going to be used to acquire

Ejaaz:
GPUs, compute, setting up energy infrastructure grids to be able to train the next model.

Ejaaz:
And Greg Rockland on a recent interview last week, I believe, said the same thing.

Ejaaz:
He said, if I could get my hands on all of the compute in the world,

Ejaaz:
I would, because it is a direct translation into revenue for OpenAI.

Ejaaz:
And they've proven that time and time again.

Ejaaz:
I think Anthropoc has done the same thing. But a part of me just feel a little

Ejaaz:
uncomfortable about this, Josh, because they've raised all this money, right?

Ejaaz:
They've given all their investors, I think I saw Ashton Kutcher in like a leaked

Ejaaz:
cap table, had made a 43X on his fund from his initial investment from OpenAI.

Ejaaz:
So they're already up massively.

Ejaaz:
And now they're going to do this huge IPO, which has rumored to be around $1.2

Ejaaz:
trillion and above on open market.

Ejaaz:
I feel like things might be getting a little bit frothy. I don't know if you

Ejaaz:
have the same take here, but I know that they need the funds for scaling their

Ejaaz:
AI, but I don't know if a lot of this, I don't know how much of this is just

Ejaaz:
kind of PR to get the biggest IPO ever.

Josh:
Well, there has been PR and it hasn't been very positive because it appears

Josh:
as if Sam Altman has been fighting with the CFO of the company,

Josh:
Sarah Fryer, as it relates to the IPO timing and how they're actually going to initiate it.

Josh:
So from my understanding, and this is kind of what intuitively makes sense,

Josh:
is Sam Altman wants to get out as soon as possible.

Josh:
We know SpaceX recently just filed for their IPO. It was privately,

Josh:
but it's going to happen sometime around June. They're going to come out of the gates first.

Josh:
They're going to raise $75 billion at a $1.75 trillion valuation,

Josh:
that's going to absorb a lot of money from the market.

Josh:
There's still a lot of enthusiasm for AI companies.

Josh:
OpenAI surely wants to get there first. I can't imagine a world in which you

Josh:
wouldn't want to be ahead of

Josh:
Anthropic when the resource that matters the most to scaling is capital.

Josh:
Now, Sarah Fryer is kind of, E.J.

Josh:
To your point that you were mentioning earlier, she's a little bit more cautiously

Josh:
optimistic where she's like, hey guys, we have a lot of money.

Josh:
We just raised a ton of money.

Josh:
Our problem isn't actually raising money, it's deploying it effectively.

Josh:
And we haven't proven that we can deploy this capital effectively.

Josh:
So why are we going to rush to raise so much more?

Josh:
And I think that's a totally valid argument. And I understand the case for both. One is very pragmatic.

Josh:
One is very CEO optimistic.

Josh:
We are going to raise a ton of money and we are going to deploy it effectively

Josh:
and bring 10 gigawatts of compute on by the end of

Josh:
Is right, but it appears as if Sam is going to win. And having more capital

Josh:
on the balance sheet, maybe it's not a bad thing.

Josh:
In the case that things are frothy, in the case that capital does start to disappear,

Josh:
they want to have at least some amount of runway to continue to scale these data centers.

Josh:
And that's what this IPO would afford them. So we'll see.

Josh:
But all of this is kind of guiding towards a gigantic IPO at some point this

Josh:
year for OpenAI, possibly ahead of Anthropic.

Ejaaz:
So what do we have as major IPOs this year? We have Anthropic that you mentioned.

Ejaaz:
I I think they're rumored to go IPO at around $800 billion.

Ejaaz:
So they don't quite hit the trillion dollar mark. And then you have SpaceX,

Ejaaz:
which last week Bloomberg reported that they're apparently valuing it at a $2 trillion IPO.

Ejaaz:
So OpenA, I think wants to- You think it's too high? 1.75, 1.75.

Ejaaz:
Okay, so you're fine with $250 billion discount. Got it, okay,

Ejaaz:
right. Okay, okay. But the point is, I think Sam wants to kind of.

Ejaaz:
Mark the ground, plant the flag that OpenAI is a serious company and that they

Ejaaz:
are capable of raising huge amounts of capital for a very necessary technology.

Ejaaz:
Now, there's the optimistic take where if things are going well,

Ejaaz:
you kind of want to add fuel to the fire.

Ejaaz:
But also, if things go horribly wrong, it's kind of nice to have $122 billion

Ejaaz:
in the back, I guess, to kind of bail yourself out. So I see the angle that they're going for.

Ejaaz:
Again, to my earlier point, apparently this is their leaked cap table that I'm

Ejaaz:
showing you on screen, which shows how much original investors in OpenAI are up,

Ejaaz:
including recent ones, by the way, such as SoftBank, who wrote a 64 or 65 billion

Ejaaz:
dollar check and is already up one and a half times.

Ejaaz:
The only one that's actually down, ironically, is NVIDIA, the richest company

Ejaaz:
in the world, who is underwater by 500 million dollars because they recently

Ejaaz:
invested 30 billion dollars into the 122 billion dollar round.

Ejaaz:
So a A lot of people are up. Ashton Kutcher, again, greatest investor of all time, is up around 43x.

Ejaaz:
The point is there's a lot of money here.

Ejaaz:
Evaluations are hugely inflated. OpenAI cannot afford not to deliver at this

Ejaaz:
point because if they do, I fear it would lead to the implosion of a lot of

Ejaaz:
capital markets, not just AI.

Josh:
Yeah, but if they do, if this new model codenamed SPUD does land sometime in

Josh:
April or May, and it performs as advertised, as this pseudo-artificial general

Josh:
intelligence, then every single open AI is dead narrative kind of reverses overnight.

Josh:
Yes. And then the GPT exodus becomes a thing of the past. And the bet that Sam

Josh:
is making is that he's going to

Josh:
be able to create the most consequential product probably since GPT 4.0.

Josh:
Yep. And if they could do that, if they could pull that off,

Josh:
and if they could do so leading up to their IPO, then that's a very strong headwind

Josh:
that they can have leading into this giant fundraising round,

Josh:
hopefully giving them a ton of money on the cap table, but also raising quite a large valuation.

Josh:
So there's a lot of things at play here. There's a lot of strategy being deployed

Josh:
and great products being built.

Josh:
And I think over the next coming weeks and months, we're going to see the fruits

Josh:
of their labor kind of come to life. And I'm hopeful that they are a grand slam,

Josh:
home run, incredible products.

Josh:
And I'm pretty stoked for the future of OpenAI.

Ejaaz:
Well, what if I told you Sam had written a industrial policy on post-AGI economics,

Ejaaz:
aka universal high income, where he dropped this this morning.

Ejaaz:
And he basically describes a world where AGI basically does and automates all human labor.

Ejaaz:
And so we need to figure out a way to give dividends to humanity.

Ejaaz:
So he talks about restructuring the entire tax system, giving people a salary

Ejaaz:
every couple of weeks or every month, and humans just get a four-day work week,

Ejaaz:
which sounds amazing probably to a lot of listeners.

Ejaaz:
So I'm just saying he wouldn't probably be writing this manifesto if part of

Ejaaz:
what he's claiming to deliver on is real.

Josh:
Yeah. So bullish on OpenAI as a company, bullish on the roadmap,

Josh:
bullish on the kind of foundational work they've been doing the last few months,

Josh:
and just very excited to see where they end up.

Josh:
I think a lot of people who are doubting them, who are canceling their premium

Josh:
subscriptions, are going to find themselves sadly mistaken over the next coming

Josh:
weeks and months, particularly starting with this new image model,

Josh:
which seems pretty incredible.

Josh:
And I imagine once it's fully released, and it's not just in its leaked form,

Josh:
we're going to be getting some really amazing demos.

Josh:
We'll be offering them up here on the show as always and as always thank you

Josh:
guys so much for joining us and for watching this episode we're kicking off

Josh:
another strong week in the world of ai if you enjoyed this please don't forget

Josh:
to share with your friends your family anyone who you think would be interested

Josh:
in this type of content and each has any final words before we take off for the day nope

Ejaaz:
Thank you guys so much for listening and we'll see you on the next one.