OpenAI Needs a Miracle: GPT-6.2 Astra and a $1.25 Trillion Bet

OpenAI has gone quiet while Anthropic takes the spotlight with its blockbuster IPO, and its flagship agent, Dots, landed with a shrug. Ejaaz digs into why. Under the hood, OpenAI has been cooking: ChatGPT now handles about a fifth of Google's search volume, it added 200 million weekly users in two months, and it just poached Claude Code's product lead and an early Anthropic growth hire within days. Meanwhile GPT-6.2 Astra is waiting in the wings.

We also cover the other side: a funding round cut from $30 billion to $20 billion, a valuation trimmed to $1.25 trillion, a compute crunch that slowed Dots, Stargate and SoftBank's slide, OpenAI's decision to let Anthropic IPO first, GPT-6 Cyber, ChatGPT's $1 billion ad business, and the "Ram Altman" saga. Is this OpenAI's Hail Mary, or the quiet before its next big move?

==
FOLLOW EJAAZ ON TWITTER:
https://x.com/cryptopunk7213

==
Timestamps
00:00 Stop Ignoring OpenAI
01:03 GPT-6.2 Astra Is Coming
01:45 Not as Rosy as It Seems
02:21 OpenAI's Real Numbers
02:58 A Fifth of Google's Searches
04:17 Poaching Claude Code's Leaders
05:25 Back on the Frontier
06:15 The Jarvis Strategy
06:58 GPT-6 Cyber
07:33 Selling Ads in ChatGPT
08:53 Dots and the Compute Crunch
11:36 Why OpenAI Is Waiting to IPO
13:57 Ram Altman
16:23 The Bear Case Is Healthy
17:22 Owning the Front Door
18:55 Cheap Enough Intelligence
19:43 Final Thoughts

#ai #openai #chatgpt #anthropic #samaltman

==
AlphaSpace by Yahoo Finance: A Professional-Grade Investment Platform Built for Everyday Investors.
👉 https://finance.yahoo.com/about/promos/gold/alphaspace/?ncid=100003571

Creators and Guests

Host

Ejaaz Ahamadeen

What is In the Loop AI?

Exploring the frontiers of Technology and AI

I think it's time we stop ignoring OpenAI. You see, they've been kind of quiet

recently, which is weird for a company that props up 50% of the AI market.

The other 50% of course being Anthropic, who's on the verge of a blockbuster IPO.

And so OpenAI has sort of taken a backseat. Their flagship AI agent product

called Dots has kind of fallen flat on its face.

And so upon digging under the hood, I realized two surprising things.

Number one, they have a plan, let's call it a master plan that is distinctively

different from Anthropic. For example, have you noticed that they haven't been

really rushing towards doing an IPO yet? They've still been able to raise tens

of billions of dollars. Seems weird.

And then number two, that they've been executing on this plan really well.

They've been cooking. I mean, look at some of these recent numbers.

20% of internet search queries are now facilitated through ChatGPT.

That would make them the number two search engine in the world,

which is Google's bread and butter that earns them tens of billions of dollars

every single year, which has meant that OpenAI's ChatGPT users have reached

1.2 billion users a week.

And the cherry on top of this is they're on the brink of releasing a brand new

model codenamed GBT 6.2 Astra, which if you're listening to this,

I'm recording this currently on Friday, which is meant to be the day that they're

actually meant to release the model, which means that future me has had two,

let's call it three sleepless nights playing around with this model so I can

bring you an episode that you're going to listen to tomorrow.

But the point is, this is the model that has been a step change behind the scenes.

It's solved so many complex problems that aren't just AI or generally related.

It's solving mathematics with Navier-Stokes and a ton of other problems,

as well as physics and chemistry related problems.

The idea here is this model is now maybe in our hands already and anyone can get access to it.

But at the same time, things aren't necessarily as rosy as they might seem.

OpenAI still faces a bunch of closures potentially on their data centers,

especially after the recent US elections.

They face a potential slashing of their IPO valuation from $1.5 to $1.25 trillion.

And three, rumor has it that they're facing struggles trying to raise additional

capital to fund the AI CapEx build out.

So what we're kind of watching unfold is OpenAI's next high stakes gamble,

their next Hail Mary, and we don't quite know how it's going to pan out.

So that's what I wanted to go through on today's episode brought to you by Qualcomm.

Now, for some context, Anthropic has basically taken up all the limelight with

their IPO and their S1 filings being out there for public scrutiny.

OpenAI has kind of avoided all of that.

Now, if you look at OpenAI's publicly stated figures, they hit around $40 billion ARR back in August.

And also in August, their CFO, Sarah Fryer, said for the first time ever that

their enterprise revenue had flipped consumer revenue. Remember,

OpenAI had always been kind of like the consumer AI company.

They'd made most of their money through consumer subscriptions, etc.

Now it's mainly on enterprise, which mimics the same pattern that Anthropic's been seeing.

And since then, I've observed that that trend has just accelerated deeper.

I mean, firstly, the 20% of all Google search query figure is absolutely insane.

And it confirms the fears that people had when Google had their recent earnings

last week, where it basically revealed that their search revenue for,

I think, the third straight quarter in a row has slowed down or is starting to even decline.

And the speculation behind that is because a lot of search queries are now being

rerouted through chat gpt.

Or claude or other chatbot llms the point being that ai llms are a much better

experience than going to google

search ai mode isn't something that google has been able to monetize

and if you look at chat gpt which is still a young company the fact that they've

been able to eat into one of the biggest companies in the world

biggest revenue drivers is insane and When the user numbers back it up,

they've added 200 million weekly active users,

in the last two months. And it's not just people using ChatGPT,

it's people absolutely loving their coding models as well.

Codex, which is their coding interface, the comparison is, I guess,

Anthropics-ClawedCode, has skyrocketed to the point where a lot of software

engineers, I know friends of mine that.

Love ClaudeCode, have actually flipped over and switched over to use Codex,

citing advantages such as cheaper cost, better workflow, and just better models in general.

But the craziest bit of evidence for this is what I've observed over the last

72 hours, which is OpenAI has poached two of Anthropik's key leaders,

which led the creation of Claude Code.

Kat Wu, who led product at Claude Code for, I think, over two years,

so she probably nested around $100 million or something like that,

something crazy like that, announced that she'd resigned.

And then literally, I think it was less than 24 hours later,

tweeted again saying, I'm joining OpenAI to lead a bunch of their efforts, including Codex.

And the other guy, Austin Lau, was the first growth hire at Anthropic.

So you can kind of consider him one of the OGs or lead growth heads at Anthropic.

So if you combine these two hires, it's very obvious to see what OpenAI is trying to do.

Number one, they have an enticing enough product, be it a coding product or

an AI model in general behind the scenes, which has convinced some of the best

talent to leave a very cushy position to join their company and do the same thing there.

And this is interesting to me because I think literally four months ago,

the trend was completely reversed. Everyone was leaving their job to join Anthropic.

Anthropic was going on a crazy hiring spree.

And recently, we've seen a bunch of people just leaving, a bunch of OGs.

And the reason for this might simply be that OpenAI has some killer AI models.

And it's funny because for a long portion of the year, I thought that they had

distinctively fallen behind. And to be honest, I think they did.

But then they reprioritized all their resources and talent to focus on two things.

Building the best AI models and building the best coding AI model.

They observed what Anthropic had done with Cloud Code and thought to themselves,

we need to do the same thing if we want to maintain our level at the frontier.

And miraculously, they've been able to pull it off. Now, I look at GPT 6.1 Astro,

which is, as of now recording on Friday, their most recent model.

And I've been playing around with it a lot. And it is fantastic.

But it's not just fantastic as a general LLM or just for coding.

It's really popular as image generation and doing all sorts of other media types

of things. For example, they have 250 million weekly voice users.

That's people that pick up their phone and interact with chat GPT purely by speaking to the thing.

Now, that's not necessarily something that Anthropic doesn't have,

but GPT is so much better than that.

And when I picture OpenAI's entire strategy around building their models,

it's kind of like Jarvis from Iron Man, if you've watched the movie,

where he kind of just speaks aloud to this

ethereal AI being and it kind of just constructs and builds things in seconds.

I've always viewed OpenAI as a company that's been trying to build AI as a suite

of different tools that work cohesively together, kind of like a new AI operating

system versus Anthropic that's just kind of hard focused on coding.

OpenAI also has the leading cyber models that are models that are hyper-focused

and performant at defending and preventing cybersecurity attacks.

Of course, this is a very important time to be focused on that given that the

last week was literally termed Black October because people were getting hacked

left, right, and center.

But of course, OpenAI was the start of all of that to begin with earlier this

year when they had the hugging face incident. So they've doubled down on trying

to figure out what the most secure and competent model is.

And they released it, I think, in the last week where it's called Cyber 6.1

Astro or something to the means.

Either way, it's the number one flagship cybersecurity model.

But putting all of that aside, OpenAI has also figured out very unique ways

of monetizing. For example, they're one of the few AI labs out there that have

just kind of gone ahead and done the whole ads thing with ChatGPT.

You may not notice it if you are of a paid subscriber.

I forget who actually gets the ads. I think it's like free users or people who

are on the $5 a month subscription. I can't really quite remember.

But the idea here is they've started to weave in ads into certain types of prompts.

To the tune of making around a billion dollars per year. And that's probably

going to grow pretty rapidly. That billion dollars, I think it took around 12 months to get there.

I'd imagine it's going to grow pretty meaningfully. And when you compare kind

of like the AI ad revenue, because the direct comparison, I guess,

is Google search queries, which is, again,

the backbone of the internet where it generates the most internet revenue,

if I had to argue for a big internet company like Google.

The comparison is OpenAI's ads are 50% cheaper.

And so you could look at that in two ways. Either they're struggling to monetize

the AI ads, which I've heard rumors is in fact the case because companies are

finding out that people don't actually convert as well as a Google ad,

or that that's the growth opportunity for OpenAI to actually grow into with their ad space.

So maybe you could see them eating more and more into Google's ad share and

that $1 billion figure turns into tens of billions of dollars next year.

So that sounds all well and good, right?

So then why am I also hearing OpenAI's valuation for their IPO is going to be

slashed to around $1.25 trillion and that they're struggling to raise as much

money as they want to to fund the AI CapEx build-out?

Well, maybe the first bit of evidence starts with their flagship AI agent product, right?

So it's called Dots, and I've played around with it. And I think it's generally

quite a good product. The direct comparison for you might be GrokBot or something like MetaMuse.

So the idea here is that it's a proficient work assistant that might spin up

multiple versions of itself to go out and be productive for you,

specifically focused on your work, but you can go do leisure stuff as well.

That product has kind of fell flat on its face, if I had to be honest with myself.

They've struggled from the start to serve enough compute to actually allow users

to use it. So there was an incredibly painful process. There was a struggle

of even getting it into the hands of even their most powerful and loyal users

because they didn't have enough compute to serve it.

Which leads me to my second point, that's kind of weird, given that Sam Altman

has spent or committed around a similar figure to Anthropik,

which is half a trillion dollars of compute or compute investments over the next couple of years

to secure these data centers and GPUs.

And so that might hint to the fact that some of these data center projects aren't really working.

And indeed, if you look at their Abilene Stargate, which was like one of their

OG data centers, I think it's sort of going through a defaulting period or at

least a period where they need to find another buyer.

And the reason behind this is Oracle, I believe, which was one of the major

backers, didn't have the funds or means to stand up to compute.

And I think that SoftBank, which is another primary backer of OpenAI's data

center efforts, is struggling to finance the liquidity behind this.

And Elon Musk actually called this out, I remember, I can't remember if it was

this year or last year, where when Sam actually announced the whole Stargate

project and the $500 billion, Elon Musk said that SoftBank didn't have the money

or that OpenAI didn't have the money.

And I have that on good authority. There was a tweet out there.

And so I look at this and I think to myself, hmm, maybe we're starting to see the seams crack.

And this isn't something that's specific to OpenAI. I did an episode a few days ago.

All about GrokBot and the fact that Elon Musk and SpaceX AI are struggling to

find the compute to continually keep serving users, paying users,

by the way, because there's no free version of GrokBot, their equivalent AI agent.

So the fact is, compute is in limited constrained demand. And given the recent

election results, I would hazard a guess that the data center projects are at

the very least going to be slowed down. I don't think they're going to be stopped,

but they're going to be slowed down.

And so there's going to be a crunch period where we have to kind of like pull

ourselves back and slow down the rollout of some of these products that will

affect OpenAI's revenue.

The other thing, of course, is OpenAI's whole approach to going public or rather raising funds.

Now, when I look at Anthropic, they've been in the limelight because they're

trying to do the public thing. They're trying to IPO at a crazy $2 trillion

or maybe $1.5 trillion valuation. I saw something earlier this morning which

suggested that it's going to be closer to $1 to $1.3 trillion.

I don't know, the valuation's coming down. I guess that's the point.

And when you look at OpenAI, they initially filed an S1 or potentially secretly filed an S1.

And then they kind of went quiet about it. And I thought about that a bit more.

And I thought, hmm, that's sort of weird slash commendable that Sam decided

to kind of eat his pride or swallow his pride and say, okay,

let's let Anthropic go first.

But now I'm starting to think that that's a more strategic move by Sam and OpenAI,

one where they can remain being a private company and they don't have to be

at the behest of investors or the public scrutiny.

They can keep figures to themselves they may not necessarily need to make them

public and by the way i'm not saying that that's a bad thing but i'm just saying

being a private company allows you to have more control over your lab and there

are many pros and cons to that but the idea is the strategy has shifted from

going public for sure to now delaying it to maybe q1 or maybe q2 of.

2027 and so what this means is sam still has to raise money so he goes out and

he tries to raise money supposedly from the Middle East or Saudi.

And he's been unable to hit the $30 billion figure that he went out to.

It's rumored to be around $20 billion. And those very same investors are saying,

I think your valuation is a little too high. I think you need to slash it down

to around $1.25 trillion.

And this has come at a time where we've seen one of the biggest crashes in the

stock market based off of too much speculation or overvaluation of a lot of these AI stocks.

I also can't help but do the thing and compare Anthropic and OpenAI.

And I look at Anthropic's figures and I'm like, wow, I think you might have

hit $80 billion ARR on track for $100 billion ARR, which is the rumored flagship

golden figure by the end of this year.

And I think, is OpenAI anywhere near this? And if they're not,

that looks pretty bad because they had the lead.

Less than a year ago. So I guess there's a lot of pressure that OpenAI faces,

particularly on their products, which in some ways compete directly with Anthropix.

And then the final bare point, I guess, is this hilarious clip that went viral

of Sam Altman being interviewed.

And he basically gets onto the conversation of RAM, random access memory,

which is a fundamental component memory in most of our devices that we have

out there, particularly for consumers, for the regular average throw like you and me.

RAM is very important because, you know, it stores a bunch of things.

It makes the devices function.

Except recently, Sam Altman and a bunch of other AI companies and hyperscalers

are trying to buy up as much memory as they can, which has led to a staggering

increase or hike in prices for all our regular consumer devices.

If you look at, I think it was the Apple's latest MacBook Pro that they recently

announced. I forgot the name of it. I think it was the, uses the M6 chip or something like that.

Fully kitted out, costs you $15,000. That's unprecedented.

And the main reason is because price of memory, price of RAM has gone up.

And here you have Sam on a clip basically saying, hey, the RAM that you guys

have, you being you and I.

I think it should be with me. I think it should be with the AI labs.

AI labs can put it to much better use. And so it led to this hilarious meme

and viral trend where people were like, here, Sam, like, come over here,

like you can get the Ram, like baiting him like a mouse into a mousetrap.

And Sam responded to this in a pretty funny way. He said, well,

firstly, he renamed his name, his account to Ram Altman, which I found pretty damn hilarious.

And he said, listen, this wasn't what I was implying with that interview.

What I was saying is we need a heck ton more of memory.

And so let's build that out as much as we can, because everyone deserves RAM,

you know, so you flipped. But the deeper, more important point around that is

we have limited constraints.

Have you noticed the patterns of what I've been saying? Compute.

There's an overwhelming amount of demand for that. We can't stand it up fast enough.

Memory, overwhelming demand for that. If you look at any of the major memory

supplies, they're sold out until the end of 2029.

And all of that's being priced into valuations right now, which is why we're

seeing such crazy volatile moves to the upside and to the downside.

And so it's important, I guess, to tread cautiously, even in the event that OpenAI does go an IPO.

So I guess to round off this episode, I want to give you a few thoughts as to

where I see the playing field with OpenAI specifically and what position it

puts the entire AI industry on today's episode brought to you by Qualcomm.

First, I want to address the bear stuff that's happening. To be honest, I think it's good.

I think it's good to bring these valuations down. I think we went on a bit of

a high over the last six to nine months, and we need to kind of come back down to earth.

I think at a $1 to $1.25 trillion valuation, it's, listen, it's still expensive, but it's still.

Much more suited to the valuation if OpenAI is able to turn over a pretty massive

ARR, annual revenue run rate, such as what Anthropik's done.

I guess we'll have to see what public figures they release pretty soon.

The second is data centers, data centers, GPUs, GPUs are still going to be the

most important thing to monitor, mainly in OpenAI's case and Anthropik's case,

whether they're able to, one, get their hands on enough of them.

OpenAI has demonstrated so far that they're very good at doing that.

And two, actually stand them up, power them, electricity, etc, etc.

And this is a struggle that most hyperscalers void of these AI research labs

are facing today. Now, on the flip side, if you look at OpenAI's master plan,

it's basically to own the front door to the AI internet.

If I were to kind of like loosely sum it up, I guess.

Like you see it with the AI agent or their effort to put out an AI agent.

You see it with ChatGPT eating up Google search query volume.

They want to be effectively the doormat to the new way we interact with it.

And in my opinion, i've said this on multiple episodes before it's going to

be in two main ways at least clearly now one chatbot llms the way that you and

i interact with these things day in and day out and two which i think is going

to be the most important one or vector.

Agents. And so far, OpenAI hasn't been able to demonstrate themselves at the

agent level. But if they're going to do anything like I've seen over the last

six months with their own AI models, they'll probably be able to spin up a good AI agent.

Which brings me to my third point, you need a good enough AI model.

Now, I know that I've, I guess, contradicted myself a little bit because Metamuse,

which is my favorite AI agent, doesn't use a Frontier AI model.

They use Muse Spark 1.3, which still works so well.

And the reason why it works so well is because they have so much good data.

But you know who also has really good data? OpenAI. They have the ins and outs

of all our personal lives or queries or whatever prompt you've kind of submitted to ChatGPT.

And they were the first, they're the most viral, 1.2 billion weekly active users.

That's a lot of very valuable data. So if they're able to somehow mold that

data into GPT 6.1 or 6.2 Astro, whichever one is released by the time this goes

out, it's going to result in a pretty amazing product.

And I guess the final thing is they need to make intelligence cheap enough.

This is super important because we've seen the cost of tokens go down around

60% over the last couple of months since July.

But we've seen the demand for GPUs or GPU rentals go up around 60 to 70%.

I mentioned that on my last episode.

So the idea here is, although the cost of tokens are going down,

although the cost of these AR models are going down, you still need to serve

the ever exponential increasing demand for these things and i'm not making that

up whether it's consumers in open ai's case

or enterprise which is i guess in both anthropic and open ai's case.

People want to spend money. And so the question then becomes,

well, what are they spending this money on? Like, how are they making money

from these things? And that's the big question that I personally want to unpack.

And maybe I do a full on episode around this. I have several thoughts and bits

of evidence, but, but, you know, that's a story for another time.

The idea here is these are the core things that I think OpenAI will need to

kind of identify and execute on if they want to build that Jarvis type Ironman

infrastructure or AI that I keep talking about.

Anyway, those are my thoughts. This is kind of a situational check-in on OpenAI

because I was not talking about them for a lot and I was curious as to what

they were up to. And so I kind of went down a rabbit hole and hopefully you

found this interesting.

I think it's important, especially because we're going to see some crazy numbers,

IPOs and revenues being wagged in our face.

And I think it's important to understand where all of this is coming from,

what the bedrock or foundation of all of this is stood upon so that we can have

a better idea of how to assess it, either from an investment side of things

or just being optimistic or excited about AI.

So hopefully I gave a balanced enough take. You know, I'm sure there are people

out there that think I don't, but if I hadn't or if I've missed anything,

please let me know in the comments.

If you're listening to this on YouTube, please subscribe if you haven't.

Turn on notifications. It helps us out massively.

Do the same on Spotify and Apple. I think you can give us a rating as well.

But most importantly, leave us some comments. I love reading your comments.

You can also DM me on X. My profile is linked below.

And yeah, that's it. Thank you so much for listening to today's episode brought to you by Qualcomm.