Monologue: The Updated Pale Horses of the AIpocalypse

This is an iHeart podcast.

Be honest, how many tabs do you have open right now?

Too many?

Sounds like you need Close All Tabs from KQED, where I, Morgan Sung, Doom Scroll so you don't have to.

Every week, we scour the internet to bring you deep dives that explain how the digital world connects and divides us all.

Everyone's cooped up in their house.

I will talk to this robot.

If you're a truly engaged activist, the government already has data on you.

Driverless cars are going to mess up in ways that humans wouldn't.

Listen to Close All Tods, wherever you get your podcasts.

Cool zone media.

Hello, and welcome to this week's Better Offline monologue.

I'm your host, Ed Zitron.

In August of last year, I put out a podcast called How the AI Bubble Bursts, where I ran down what I thought would be the signs that things were collapsing.

My pale horses of the AI apocalypse, if you will, which is something that reads really well, but when you say it out loud, not so good.

The reason I put these together is that the bubble is unlikely to have one specific moment where things explode.

Things like Bear Stearns' collapse during the great financial crisis were such significant moments in part because that was a public company, mortgages were a massive part and are a massive part of the economy, and indeed there was billions and indeed I think trillions of dollars resting on them.

And I don't really think there's a situation like that, nor is there one where a public company collapses as a result of the AI bubble bursting.

This bubble is different because it's so thoroughly based on vibes.

I'll end up at some point doing a longer episode about this, but the long and short of it is that the actual core of the system, the real weak point, is actually NVIDIA.

NVIDIA's continued success comes from the reliable quarter-over-quarter or year-over-year growth from selling GPUs, the graphics processing units that power generative AI and that people take and put inside of servers so that they can immediately start losing money.

More than 40% of their GPU revenue comes from the other six companies in the magnificent seven, Microsoft, Google, Meta, Tesla, Apple, and Amazon.

And Nvidia deeply depends on their continued and growing hunger for GPUs.

It isn't just enough that they buy the same amounts, they must buy more.

The rest of the Mag7's valuations are dependent on continued growth too, which is what AI is meant to provide them, but it's not really providing growth at all.

Indeed, their growth stories are not really based on numbers, but vibes.

They're based on feelings and the general warm, fuzzy feelings that dipshits get when they claim that AI is the future.

Let me give you an example.

Meta's stock jumped because they've announced a theoretical data center the size of Manhattan.

Now, important detail here as well.

Meta is making exactly minus dollars on generative AI.

They're just putting generative AI everywhere.

And yeah, giving people a little bit of psychosis, I imagine, filling feeds full of slop.

It's very bad.

And you know what?

It also isn't making them money.

And right now, this is really good for the Magnificent 7 because they're not actually having to prove that AI is a big deal.

People are basically believing it is because the AI trade, which means the value of these stocks, remains positive.

And honestly, it is currently a positive trade, even though it's based on effectively nothing.

Like seriously, Amazon may only make $5 billion in revenue this year on AI.

And I mean revenue, not profit.

Google, maybe $3 to $7 billion, Microsoft $13 billion, and $10 billion of that is OpenAI's compute.

These are pathetic numbers, but these companies are let off because the AI trade still works.

The possibilities are still there.

Profits that don't exist today and don't appear to have any path to them either, they're coming.

We swear.

If we say agents enough time, maybe it'll work.

The long and short of it is that everything about the AI industry is based on some level on lies, on lies both subtle and overt, and the general sense that generative AI is both the future and the future of the economy.

There are a few real tangible things to point at other than ChatGPT's 500 million weekly users and the tens of billions of dollars invested in AI startups.

Or maybe the weird acquisition deals like the $2.4 billion that Google just gave kind of AI code editor to WinSurf, except Google took the staff, some of the staff, the C-suite, the R ⁇ D team for the $2.4 billion, and then Cognition, who makes the Devin coding bot, took the rest for an undisclosed sum.

It's all really, really fucking weird.

It's all very strange.

And yet, when you look beneath beneath the surface, there really are few returns.

Winsurf's monthly revenue was $6.83 million a month, a ridiculously small amount, working out to about $82 million of annualized revenue.

Yet, based on the Information's Generative AI database, Windsurf was actually one of the highest-earning AI startups out there.

No, really, based on annualized revenue, meaning month multiplied by 12, most AI companies make about $100 million annualized or less.

No, really, that's about $8.3 million

a month.

With Sinsteers, Weka, XAI, Together, Abridge, Glean, Perplexity, and they're all making about $100 million annualized.

And I must be clear about how weird this stat is.

Annualized just means the month of revenue times 12.

Customers churn.

Customers churn.

So your ARR may go down when you have a few people drop out.

It may spike arbitrarily because you increased prices.

And it's just, it's a very bad statistic to use, yet people are not really looking at the other thing, which is that's not a lot of money.

$8.3 $8.3 million a month.

You have ad agencies that make several times that.

Ad agencies buying ads on Facebook and Google.

These are terrible fucking companies.

I mean,

outside of the 100 million ARR side, you've got Midjourney, who are inexplicably profitable.

We'll look into that.

And Neo4j, and they make $200 million a month.

Sorry, $200 million annualized.

So about $16.6 million,

which is, again, not amazing.

None of these are amazing.

These aren't great these would be okay startups if they were profitable and they're absolutely not but right on the tippy top behind anthropic and open ai is any sphere who makes cursor and they make 500 million dollars annualized or about 41.6 million dollars a month which is bad because in june all of their prices got increased by anthropic and on top of that they had to change the pricing and now cursor kind of sucks go on their subreddit check it out anyway this is why you don't like annualized as a stat by the way because i reckon their july revenue might be a little bit lower than 41.6 million dollars but we'll never know because well unless someone reports there if you've got any numbers on any of these companies my signal is ezitron ezitron.76 on signal please send me anything you have i love numbers Anyway, past that point, it's only Anthropic and OpenAI.

And I've left out a few companies like Turing and Scale.

They don't make AI products.

They're consultancies.

But the long and short of it is this industry doesn't make much fucking money.

The actual returns from it aren't that very good.

It's not very good at all.

But the reason I'm telling you all these numbers is that I need you to know that this industry is not held up by the actual businesses or their revenues.

None of these companies are profitable.

And it certainly isn't held up by the actual products themselves.

And I'm sure you've got an AI friend who's come to you and said, oh, I use it for this.

See how far you get them in the conversation before they just say they're using it like a search engine.

We had this whole thing with Brian on the on the podcast this week.

It's like, yeah, really emphatically talking about the fact you've got a slightly sexier search engine that may or may not get things right that can fool you into believing you're correct about something, which isn't a fucking product.

No, if it was, people would be making money.

It's ridiculous.

But even in areas like coding, coding, the one place where people can point and go, look, look, users like this.

Look, Edward, look at them.

A study from non-profit group Model Evaluation and Threat Research found that despite developers believing AI coding tools sped them up by 24%,

coding tools actually increased the completion time of their tasks by 19%.

It actually slowed them down.

Oh my god, every week with this industry, I feel like I'm going insane.

It's like every week there's this goddamn story like this.

I still have people who tell me it's the future.

I feel like I'm going insane.

But as a result of AI being a nearly entirely vibes-based economy, the pale horses of its apocalypse will be vibe shifters rather than defined events that would logically change things.

Remember DeepSeek?

That was a clunky story.

So you had a Chinese model developer that may be cheaper, it was cheaper to train, much cheaper to train, but still cheaper to train.

And their model might be cheaper and it is comparable.

I mean, I've certainly made my hay with the episodes and interviews I did around it, but I'm surprised that shocked the market.

There are bigger, nastier stories.

But if that can do it, well, basically anything could.

One thing that really spreads like information poison into the heads of business idiots everywhere who are saying, AI now, please, without really understanding what it does.

It's unlikely to be one big pop, but several little duts from the bubble deflating.

So let's get to it.

Here are the signs to look out for.

Freshly updated.

And if you have any ideas, please do reach out.

I'd love to hear from you.

We'll start with an obvious one.

Any price increases or decreases are open AI and Anthropic.

Now, I called these increases in August, and I had decreases because, as I've mentioned in the subprime AI crisis, there is now a race to the bottom for model developers.

OpenAI cut the price of their O3 reasoning model by 80% in June in an attempt to undercut Anthropic's Claude for Opus, which launched May 22nd, I believe.

And we've already begun to see price increases too.

Anthropic added service tiers and OpenAI priority processing for enterprise customers to give them uninterrupted service in June.

And I must be clear, when I say enterprise, I mean any AI company that has customers at scale.

So any sphere, cursor, glean, and the like, they are enterprise customers and they are very likely having to pay these prices.

Now, here's another one.

Anthropic and OpenAI may move away from monthly subscriptions and indeed any AI company may do this.

This is a huge pail horse.

When companies start moving away from simple business models, so 20 bucks, 100 bucks a month, or they start changing those so the returns aren't on so the returns the actual usership isn't obvious isn't obvious how much you can or can't use that's a sign that the company is trying to contain costs and indeed fuck with their customers a bit and we've already seen a little bit of this with how cursor changed their pricing or claude max the hundred dollar or two hundred dollar a month subscription from anthropic which has now changed things to say you get more usage at the higher tiers than the twenty dollar a month pro subscription how much more it isn't clear go to claude subreddit you'll see a lot of people who are constantly saying i don't know what the rate limits are and that's working as intended by the way Anthropic is tweaking things very clearly on the back end.

We do not know what they are.

There are rumors that they are quantizing models, so making them smaller, dumber, cheaper to run, but still doing comparable things, or just not having as much access within peak hours.

These are unsubstantiated rumors, but I've seen enough claims of them that they're worth mentioning.

These are signs that things are getting bad or that they need to make things more profitable.

And really, any and all cost increases on consumers by any AI service service are a payoff horse, as they suggest the company in question is trying to deal with these ruinous costs.

Remember, there are no profitable AI companies.

It's fucking crazy.

We are three years into this dog shit, and there's none of them.

Okay, Surge AI, they are an AI data training company.

That does not matter.

It does not matter.

If you've got a car that has a horrible fuel economy, you don't point at how well oil is selling to say that the car's good.

They're different things.

Anyway, any and all rate limits at any AI startup are a huge pail horse because these companies are spending a shit ton of money and they're having to run these massive infrastructural operations to keep things going.

Those things are expensive, power in particular is expensive, so any mediating those costs is necessary.

If you see it with other companies, it's even worse because it means that OpenAI and Anthropic are turning the screws on them.

Now, another obvious one, of course, is any funding or revenue problems from the remaining AI software companies, Glean, Harvey, AlphaSense any sphere who makes cursor as I mentioned or Replit for example it suggests that the funding required to keep these companies going is running out and their business models aren't working and I want to make one real clear point here I'm kind of repeating myself but it's necessary These companies are symbolic of whether generative AI will work.

I realize ChatGPT is the big sexy one, but if you can't build software on generative AI and sell it for a profit, there is no industry here.

Every single major cloud computing revolution has started because there is profit.

And fucking hell, it's nothing like Amazon Web Services when it grew.

I'm tired of hearing this point.

I think I have to do a thing about it because I'm fucking sick of it.

It's nothing to do with it.

Nothing.

It's not remotely similar.

The only similar thing is their goddamn data centers.

God fuck.

Calming down.

But on the subject of data centers.

Any delays to Stargate or Amazon and Anthropic's new Carlisle, Indiana build out?

Both Anthropic and OpenAI are running into capacity issues.

This is documented well, both by the companies and reporting from outlets like the information.

And they need these projects to happen.

Because right now, it's very obvious Microsoft is not going to expand their infrastructure for OpenAI.

And Anthropic, well, they need Amazon to do it, but data centers don't grow like weeds.

They take forever to build.

I don't even know when the Carlisle, Indiana data center is going to get built.

Fucking hell.

What a mess.

Every time I think about this, it's such a mess.

All right, one final pale horse, and this is my favorite one.

Anything that happens with SoftBank and their ability to raise money is a big pale horse, the palest of them all, actually, because SoftBank is critical to everything.

Softbank is the money behind OpenAI.

SoftBank is the money behind Stargate.

SoftBank is financially responsible for Stargate, by the way.

And if OpenAI doesn't move in to

Stargate, and indeed if Stargate never happens, Oracle gets left with $40 billion of chips unused and Crusoe, who they have to pay a billion dollars, I think.

They have a 15-year lease with them.

What a mess.

But as I've said, the actual bubble bursting will be slow, annoying, and unsatisfying at times.

But these power horses are the signs that things are falling apart.

I will, as ever, keep you up to date on when these signs are happening, what they mean, why you should care.

I'll probably say fucking shit a few times when I do so.

Really appreciate you.

Love you all.

Thanks for listening.

Hi, I'm Morgan Sung, host of Close All Tabs from KQED, where every week we reveal how the online world collides with everyday life.

There was the six-foot cartoon otter who came out from behind a curtain.

It actually really matters that driverless cars are going to mess up in ways that humans wouldn't.

Should I be telling this thing all about my love life?

I think we will see a Twitch stream or president, maybe within our lifetimes.

You can find Close All Tabs wherever you listen to podcasts.

Along with Colorado State, Vanderbilt, Virginia Tech, Western Kentucky, South Florida, VCU, and Wichita State, playing 12 games over three days.

It's basketball at its best, plus everything Atlantis has to offer.

Aqua Venture Water Park, White Sand Beaches, World Class Dining, and more.

Get your tickets and accommodations at battle for Atlantis.com.

Hey, it's Brian Christopher.

Ready to chill the summer?

You're in luck.

I'm hanging out at Chumpa Casino, and you're in for a treat.

Chillax with hundreds of games, daily bonuses, exciting spins, and epic prizes.

It's all here, always free to play.

Kick back, have fun, and head to chumpa casino.com.

Let's make this summer legendary.

Sponsored by Chumpa Casino, no purchase necessary, VGW Group, void where prohibited by law, VTNC's 21 Plus.

Cozy up with fragrance that feels like fall and smells unforgettable.

Pura's smart, app-controlled diffusers pair with premium scents from brands like Nest New York, Capri Blue, Anthropology, and more.

Whether you're craving spiced pumpkin, warm amber, or nostalgic woody notes, there's a scent to match every mood in every space.

Discover why Pura is the go-to for premium home fragrance.

Start your fall refresh now at Pura.com.

This is an iHeart podcast.