In the Loop AI

AI stocks just had a brutal week. Google lost more than $500 billion in market value, Amazon fell 12% after hours on its earnings, and Intel had its worst day since 1974, all while the biggest tech companies reported record cloud revenue and promised to spend even more on AI next year. So is this the dot-com bubble all over again? Ejaaz walks through how the selloff started, why search and the open web are cracking, and who is really paying for the AI buildout.

We also lay out the bear case (concentration, borrowing, VC-subsidized spending and the walls going up around Reddit and Amazon) and the bull case (real demand, cash-rich hyperscalers, labs with money in the bank, and what happened to the internet after 2000), and why this looks more like a correction than a collapse.

==
Timestamps
00:00 Is the AI Bubble Popping?
00:55 Big Tech Doubles Down
02:14 How the Selloff Started
03:11 Three Signals From Earnings
04:07 Search Is Breaking
04:59 Reddit Builds a Wall
07:20 The Overbuild Question
09:00 Who's Paying for It
10:13 Walls Around the Web
10:54 Amazon Blocks Meta's Muse
12:34 The Bear Case
13:46 The Bull Case
17:23 The Internet Still Won
18:51 Final Thoughts

#ai #stocks #bubble #nvidia #google

==
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

There's no easy way to say this, but we might be witnessing the collapse of

the AI bubble in real time. Over the last two weeks, we've seen over $2 trillion

wiped out in the stock market.

Amazon's down 12%, wiping off half a trillion dollars of its market cap.

Google is the same, wiping off another half a trillion dollars.

And Intel has seen a drawdown of 30% last Friday in a single day in the opening

hours, its biggest drawdown in history.

And it's all because people are losing faith with the hundreds of billions of

dollars that are being invested in AI CapEx to buy GPUs and to build these new models.

It's getting very costly and people don't see the return on that investment.

They don't see how AI is going to make money.

And this isn't just something happening in the West. China's stock market is

down 40% over the last couple of months, the biggest crash they've ever seen.

And the big question on my mind is, are we in another dot com era bubble that

is just a hundred times bigger and is going to result in a much

worse crash well the thing is someone's already answered that question for us

the hyperscalers amazon microsoft meta google and all the ai labs have made

their decision and it's clear

they're doubling down. Next year, we're planning to spend over $1 trillion on

compute, buying GPUs, building data centers, investing in power, etc.

To stand up these next gen AI models that we still haven't quite seen the ROI from.

The craziest part is all of this is being basically paid for by two companies, Anthropic and OpenAI.

So the big question mark is, are these companies, one, going to be able to make

enough money and two, continue growing at the rate that they are?

Well, the answer is there's a lot of nuance. It's partly like the dot-com era

bubble, and there's certainly things that we need to watch out for and manage

risk, but it's not entirely true.

If you look at any of the earnings reports from any of these companies that

are tanking over the last week, you'll see that they're making money and they're

having record quarters making more money than they have in the prior 10 years.

So the question we have to ask is, why is the market worried?

What do they see that companies don't?

And why they're still questioning that the money that's being made isn't really

real. So we'll get into all of this and I'm going to break it down for you on

today's episode brought to you by Qualcomm.

So first, some quick context. This drawdown really started around two weeks ago.

And it was from one single tweet from Elon Musk, where he responded,

yikes, to the over leveraging of debt seen in neoclouds, neoclouds being the

ones that buy up this compute GPUs and serve it to some of the AI labs.

And the company in focus at that time was CoreWeave. CoreWeave subsequently

then dropped down around 30 to 35% in its stock price,

as well as other competitors that were close to it, like nebius and iron and

so people really started to double down and ask the question are we spending

way too much money on gpus and all of these gpus actually being used do we have

the power and the electricity to power them now that then tied into last week's.

Record earnings week where basically the majority of the companies that being

hyperscalers like microsoft amazon came out with their quarterly earnings and

it gives you an insight into the numbers as to what's actually going on behind the scenes.

Three things were very clear. One, they're spending more money than they can afford.

Their free cash flow has officially gone negative for the first time for all

three of these companies, Google, Amazon, and Microsoft, for the first time in a decade.

So they're technically in the red, and they're borrowing a lot of money and

using profits from other business lines of their business to double down on

this bet that we still haven't seen play out.

Then number two, the ROI on that investment is starting to be seen from them selling compute.

So they would sell compute to say like Anthropic or OpenAI, and they would make a profit on that.

And all of their cloud service providers, so AWS, Google Cloud,

Microsoft Azure, posted record earnings.

Then that leads me to point three, which is, I guess, why the market it is tanking,

the fundamental business model of the internet, which is basically search and ads.

Is crumbling and Google is kind of like the leader here.

The point is that Google has driven ads and sales and revenue for the entire

Internet economy since the start of it.

So the idea here is any company that sits on top of Google Search,

basically relies on Google Search and search is starting to break.

The main reason why it's starting to break is people aren't going to google.com

anymore. They're not searching things up. They're not scrolling down a list

of links and trying to select which one.

They're just using AI to do all of that for them.

They're outsourcing it to AI. And AI doesn't have a monetizable search revenue.

And most importantly, it doesn't belong to that system that Google set up.

It belongs to Anthropic. It belongs to OpenAI and any other competitor that has an AI chatbot.

And so that's why you're starting to see this affect companies that aren't necessarily

affiliated with the AI bubble. Let's take Reddit, for example.

They announced their earnings. Again, you guessed it, record earnings,

all positive, no reason why the stock would crash.

And yet it crashed around 26% in the opening hours of its trading. Why?

Because they noted two things.

One, search ad revenue wasn't driving enough traffic to their site as usual.

And that's because, as Google mentioned a day earlier in their earnings,

we're just losing kind of our grip on this because AI search doesn't monetize

in the same way. But there's also a second thing happening, which is arguably

more concerning. And I'll stay on Reddit as an example.

Reddit is starting to build up walls around its service to stop people accessing

it for free, which is crazy because the internet economy, the way that it was

able to prosper in the first place, was because we ended up becoming more open.

It started off very closed with the intranet, and then we moved over for it

becoming much more open. And society and the world benefited a lot more.

We had all free access to information.

And Reddit has been kind of like the bedrock and core of the cultural pulse

on society, if that makes sense. And now if you want to access Reddit information,

you need to create an account and log on to it.

And the main reason why they're doing this is they're concerned that their entire

company in Moat will get replaced by AI.

Reddit basically has an enormous trove of data from people's comments,

subreddits, threads, et cetera. And so this information is very valuable if

it's working to inform you to.

Make a purchasing decision or research something or whatever that might be.

News often breaks on Reddit.

And to date, they've been given that away for free. But what AI is doing is

they're taking this data, they're training models to then become smart on this

data, to effectively become your own personal Reddit.

And then Reddit's not getting paid for any of this. Now, traditionally,

Google would drive all of the traffic to Reddit.

Reddit can display ads, et cetera, et cetera.

But if you're just communicating to the internet via your AI agent or your AI

chatbot, why would you go to any other website bar just your chatbot LLM interface,

such as Claude ChatGPT or your personal agent interface?

And so that's one of the main reasons why I think we're seeing a negative correction

towards the entire AI bubble.

But the second thing is equally as important, and that is a potential overbuild

in AI capex. So right now, we have invested on order of, I believe...

Just over a trillion dollars buying GPUs, standing up data centers,

connecting it with power lines, all to train AI models that we don't quite know

how it's going to end up benefiting society in the long term, right?

There's a lot of question marks around, okay, what's the necessary ROI from this?

In fact, we had a report from the chief economist of Apollo,

I think a few weeks ago, basically saying, we haven't seen where the money's

being made outside of the seven companies, the seven top companies in the world.

Seven top companies in the world are these hyperscalers who are buying these

GPUs to then sell to two companies, Anthropic and OpenAI.

So you can see how the circular economy is being criticized right now,

basically saying, I don't know where the money is coming from.

But it's not just where the money is coming from.

It's also the ability to deliver on this. Two points here.

Microsoft's intended spend for next year is around $170 billion.

That's huge. but a chunk of that comes from this year where neoclouds or third

parties which they engaged with to deliver and stand up these data centers weren't

able to deliver on it and the main reason is,

they couldn't construct it fast enough there was too much regulatory red tape

and it was just hard to get their hands on certain credit financing and gpus

and the second point is they're realizing that it's so much harder

to build these data centers than they initially projected and it's not a matter

of, oh, it'll get delayed by a few weeks or maybe a couple months in the worst case scenario.

We're looking at potentially years. So there's two forces at play here.

One, who's paying for all of this? Right now, it's two companies,

Anthropic and OpenAI, and a heck ton of financing and loans.

Just to put this into perspective, I think I saw a JP Morgan report,

which basically said that around $2.9 trillion is planned on being spent by

some of the top, I think, 10 companies in the world over the next couple of years, up until 2028.

Half of that is being financed by the hyperscalers and these companies themselves.

So they will pay for it off their own bank balance sheet as well as any money

that they make in the future.

But the other half is being financed. And that financing is getting quite expensive right now.

We saw last week where CoreWeave's borrowing debt basically rose to around,

I think, 16% when it was 10% literally two months ago.

And we're starting to see companies,

even the top labs like Anthropic are struggling to get financing.

They recently tried to close a, I think, $63 billion deal, which Hocktan from

Broadcom had to meet with the bankers and really push them to give them better rates.

Otherwise, the deal wasn't going to go ahead. So there's a lot of pressure in

this bubble when it comes to money. The money is effectively thinning.

And the second force is the business model for AI relies on these models being

able to train on the open data of the web.

So we're talking about all the data in the web that can allow it to get smarter

and smarter with every model generation. But if we have companies like Reddit,

which owns a large portion of the data on the web.

Putting up walls and banning access for AI labs or models to kind of get smarter,

you can see how all of this money being invested in CapEx and GPUs might actually

not be put to good use because we're going to see a slowdown in models getting smarter.

And on top of that, it also makes it harder to build a really cool product.

I use the example of Metamuse, which is like one of my favorite personal agents

and honestly, favorite AI products of the year. And I have no vested interest

in meta at all. I just enjoy the product. It's so good because I could just

talk to it and it goes off and does things for me, right?

But one of the main enablers allowing it to go do things for me is having access

to these different types of information. And one of the things that it does

for me is it can do some of my Amazon shopping or Instacart shopping or DoorDash ordering, right?

But Amazon recently shut off access to Metamuse for the exact same reason that

Reddit has, which is I don't want you.

Meta, being the interface that kind of drives traffic towards our website that

drives more purchasing utility.

I want it to be the old model where I'm in control and I'm able to figure out

what customers want and what they don't want.

In the traditional mode, you and I would go to the Amazon website and we would

scroll and Amazon would see that. They would see what we click on.

They would see what we review, etc, etc.

If it's instead an AI agent doing that, if it's instead an interface that I

interact with that has nothing to do

with Amazon's website, then they lose a lot of important customer behavior and

analytics that will inform how they are to build billions and millions of dollars

more value for their company.

So the conundrum here is if they build up walls, if a company as big as Amazon

builds up walls, then every other company is going to do the same thing.

And what that world looks like, AI agents interacting with AI agents is still

unclear. Okay, so are we in a dot-com era bubble? Let's answer that question

on today's episode brought to you by Qualcomm.

Firstly, the bear case is, let's see, there's a concentration risk.

So seven of the top companies in the world are the ones that are spending all

of this money to fund this CapEx build out. And it's important to note that

if it wasn't for these seven companies, the entire Fortune 500 index would be down this year.

A lot of companies on average performed badly. It was only these seven companies

that were keeping the index profits and chart go up mentality alive in the first place.

A lot of the money is being borrowed, around 50%. Again, of $3 trillion that

are being committed over the next couple of years, half of that is being borrowed

at pretty high interest rates.

The question is, can they be able to pay all of that back? Well,

that depends on AI making money in the first place.

And that leads me to number three, which is, if you believe what Chamath is

saying, around 75% of every dollar spent in AI is through VC subsidized dollars.

So it's this whole circular economic loop. And the fourth final one is companies

are now starting to build up walls.

So we're doing the opposite of what we hoped or intended, which is AI just goes

out and does everything for you to AI can go out, but then they're just going

to get blocked by a bunch of walls because these companies don't want to give access to them.

So that's the bear case. But I want to focus on the bull case because this is

ultimately how I view it and how I feel things are going to play out over the

next six months to a couple of years.

Firstly, if you want to compare this to the dotcom era bubble,

there's one very obvious distinction.

In the dotcom era bubble there was a lot of supply being built up.

Zero demand, like literally not enough demand to satisfy even 20% of what was

being built. It was a massive overbuild.

We are seeing the opposite here. Back then, it was promises.

It was like, yeah, we're going to pay you this amount of money.

Over here, we're seeing quarterly earnings result of money, cash in hand,

being paid to these companies that are investing a lot of money into this capex.

So they have demand of people actually buying these products using the models.

If you look at Anthropik's recent S1, you'll see that they are making not only

a lot of revenue, which is the fastest growing revenue of any company that we've

ever seen in the past, but the revenue is also lean, meaning the amount that

they invest to get that revenue is now overwhelmingly positive.

Number two, the people or companies that are spending in this case versus the

dot-com era bubble are companies that can actually afford it.

I'm talking about Amazon, Microsoft, Meta, Google, and the likes,

which have such a huge amount of money on their cash balance sheet.

And yes, they may now be going into cash flow negative. But the idea is they

have other business lines that are generating billions and billions of dollars

every single month, every single quarter, every single year.

So they feasibly have a way to be able to fund this. If you compare that to

the dotcom era bubble, that wasn't the case. Okay, but then you might add,

how is Anthropic and OpenAI, their startups, that's where a lot of this purchase demand comes from.

That is right. And Anthropic has committed to around half a trillion to $750

billion worth of spending over the next five years. A heck ton of money.

I looked at the S1 earlier this week and I saw, well, the main thing being that

they have around $132 billion on their cash balance sheet. But most importantly,

they only spent $7 to $9 billion on compute.

The rumor had it was that they were overspending, that they would be in the

loss. But actually, if you remove that and some of the share equity positions,

which was recorded as a loss on their balance sheet, they would actually be

turning a profit. Now, it doesn't matter if you agree or disagree with that.

That's just the direction that they're heading. They're eventually going to

be turning quite a big profit.

So the idea here is if OpenAI is also working at a similar rate,

which my inclination leads me to believe, then we end up with these companies

being able to pay for a lot of the GPU compute that they're paying the likes

of Amazon, Microsoft, etc.

To pay for. And remember, Amazon, Google, and Microsoft aren't just selling

to these two companies. They have a heck ton of enterprise contracts of which

around 50% of Anthropics revenue actually feeds through.

So Google Cloud, Microsoft Azure, AWS all have enterprise customers that are

paying those companies in order to get access to Claude and ChatGPT.

That's not direct to open air. So the idea is these companies are generating

quite a heck ton of money. And so when I look at these enterprise companies,

I'm like, okay, well, a lot of these enterprise companies are in the Fortune 100.

They're not VC backed. So this is coming from money that they're making,

and they're not going to be spending this money if it doesn't end up with them making more money.

And when I look at the net revenue retention, which is a NRR stat,

basically tells you how much money.

Someone spends at the start of the year using a product versus how much they're

estimated to spend at the end of the year.

For Anthropic Claude, at least, it's 900%.

So you end up spending nine times more because the product is that good.

I don't think that's made up. And the final point I'll make,

which I think is arguably the most important and biggest point to make,

the internet error still ended up panning out pretty damn well.

Yes, we saw a massive crash in correction, but that also led to the birth of

some of the biggest, most important, relevant AI companies that back the entire

economy. Forget about the internet economy. The internet economy is the economy today, right?

And so whilst we may see, yes, a huge inflation in this AI bubble,

and I definitely think we need to have cause for concern, call out the risks,

monitor the risk, and make sure that we don't overstep,

we should also expect some kind of deflation or correction of this entire bubble

and i would actually argue that that is what we're seeing right now

and a main reason why we probably won't result in an overbuild or dot-com error

type bubble is because it's really hard to overbuild right now

in the dot-com error bubble you could build out the telecom lines and like over

invest in that and that's what ultimately led to the collapse here.

You're restricted by a lot of electricity production and the ability to stand

up power to fund these GPUs and data centers.

So I think inherently, a lot of these big hyperscalers and AI companies are

seeing this and thinking, yeah, as much as I want to double down,

we aren't going to be able to spend money as fast as we would like because we just can't.

And that works as a indirect risk manager, I guess, in a way,

where it prevents us from inflating the bubble too much.

But I do believe that the market is in two buckets right now.

Some people that are thinking long-term about this and are betting that this

AI thing is gonna pan out pretty impressively over time. It's gonna consume

anything and everything and generate a huge amount of wealth for the world.

I sit in that bucket. And if you're a long-term investor, I'm just buying some

of these dips and just waiting to see how the market kind of pans out and making

sure that I don't recklessly throw money into this market.

But if you're a short-term investor and you're kind of like looking at these.

Inflated valuations, you know, Anthropic potentially coming out at a $2 trillion

valuation, SpaceX had a $3 trillion valuation or whatever that might be,

you're kind of looking at this and thinking, well, damn, like this might be

overvalued and you might be right over the next six months.

So take both of these things into account and hopefully this was a good enough

deduction, at least on my internal monologue as to how things are actually panning out.

Like it's easy to look at the markets and panic, but it's also important to

understand the nuance between historical context and what's actually going on today.

So yeah, I guess, you know, let me know your thoughts. I'm sitting here on my

own kind of like working through these things trying to figure out what makes sense and what doesn't

and maybe there are folks out there with different perspectives that might make

sense and change my own perspective so if you're on Spotify or Apple give us

a rating leave us a comment helps us out massively and also DM or comment.

Any kinds of thoughts or feedback that you have any kinds of topics that you

want me to cover I'm all ears and I see all of these things and I really enjoy

making these episodes so yeah I guess that's it thank you so much for listening

to today's episode brought to you by Qualcomm.