Betting on AI: The Delusion Driving Big Tech

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
These people are doing it because they're too stupid

(00:02):
to realize that what they're in is kind of a scam.
I think that they're scamming in the sense that they're selling something
they don't really understand, but they're not like, I think this will die.
I think that they think this will go forever.
Kind of like crypto and when I, this is the other thing, people
with my work, they're like, oh, you started like a year ago.
I was writing in 2020.
I was on crypto before.
Everyone.

(00:27):
Welcome to Screaming in the Cloud.
I'm Cory Quinn, and I've been looking forward to this episode for a while.
Ed Zitron is the host of Better Offline, the writer of Where's Your
Ed at Newsletter and has often been referred to as an AI skeptic.
I tend to view him as much more of an AI realist, but we'll get into that.

(00:48):
Ed, thank you for joining me.
Thank you for having me.
So you have been very vocal about AI and I, I wanna say your
skepticism of it, but that feels like it's an unfair characterization.
I think you're, you're approaching this from the perspective
of you are trying to sell me something, I'm not going to buy
it without examination, and the rest sort of flows from there.

(01:10):
Yes, and that's the thing right now, and I've said this
a few times in a few interviews, I am being and have been
characterized as a radical because I'm like, Hey, this company
loses billions of dollars and has no path to profitability.
Or, Hey, none of these companies appear to be making much revenue, let alone
profit, but they're spending so much money on this, is that sustainable?

(01:32):
And people like, ed, you are crazy.
You are crazy.
I irrational person believe that these companies will
simply work it out in some way that I cannot describe.
I have blind faith.
You are the radical for not
Right?
Strangely, when I try and pitch people on things with blind
faith in them, they, they tend to ask a whole bunch of
uncomfortable questions that I'm not prepared to answer.

(01:52):
But nowadays with ai, when you ask those
questions, you're treated like you are the problem.
Exactly.
You, it's like a gifted child that you have to, it's what is that, um, Twilight
zone episode with the child that can dream, dream, thinks of something.
It's just everyone wants to coddle AI in case it
hurts them or in case it breaks the narrative.
It's just, it's very frustrating because disagreeing with me is one thing.

(02:14):
I do not have trouble with that unless the person is just like, their
argument is no, because that is most of the response to my work is just nah.
No, it's not.
Well, I'm trying to sell something and if what you're
saying is true, that impedes my ability to do that.
So shut up, please.
Yeah, could you, but even then, it's like it goes beyond that.
'cause you've got tons of journalists as

(02:35):
well who are just like, no, it's not true.
They're gonna work it out.
This is all part of the plan.
What's the plan?
Can you tell me the rest of it?
Because it doesn't seem like a good plan unless the plan is raise as much
money and get as much awareness as possible, which they've succeeded at.
They've not made any money.
In fact, they're not making much money at all.
No one is making any money.
I think there's like two profitable companies within this space.

(02:59):
Maybe, maybe three.
If we go back in time and we look at the standard generative AI chatbot or
coding assistance and you offer it to someone in a vacuum, they would spend
thousands of dollars a month for this based upon the perceived utility.
Not everyone, but enough people would now though that has
been anchored at about $20 a month per user is the perception.

(03:21):
So I don't see a path to suddenly multiplying each price by
a hundred to wind up generating some sort of economic return.
They have anchored their pricing at a point where I don't see a
path to significant widespread industry defining profitability.
I'm gonna be honest, I don't think they would've paid thousands of dollars.
You had code auto complete things.
I talked to a client about this before COVID, like I was remember in a WeWork,

(03:46):
and they were telling me, yeah, it was kind like auto complete for coding.
I'm like, oh, seems useful.
I bet people would like that They, I think
they sold part of it to Microsoft or something.
I, I forget.
But also, if they would've paid thousands
for it, wouldn't they have paid thousands of.
Dollars for it already.
'cause GitHub copilot basically got out there before everyone.
Would people not have paid thousands for that?

(04:07):
That's the thing, Microsoft launching this product, they are the pricing gods.
If they thought they could get a thousand
dollars a month per head, they would do it.
They would do it today.
They would've done it yesterday or several years ago, but they didn't.
And there was a story that came out in like 2023, I think that said that, um,
Microsoft was losing like 20 bucks a customer with GitHub copilot as well.

(04:29):
But putting that aside, I think that people overestimate, or sorry,
underestimate how much software engineers love automating shit.
Like there, there was the whole platform as a service era.
Uh, the, the, I mean when that with containerization, when it came
in, I remember talking to a client at the time PR firm who was

(04:49):
just like, yeah, there are people who just do containerization.
'cause their boss said they'd heard it at a conference.
Nick Suresh Esh, uh, ity, he did a great piece about Monte Carlo.
I don't even know what Monte Carlo does, but he has a great
piece about how a bunch of people were trying to get Monte Carlo.
They were like, why am I talking about Monte Carlo?
And no one knew what it was.
This is the modern software era and we can't even make

(05:11):
real money off of AI in this kind of specious software era.
It's crazy to me.
Man.
It, the whole thing feels insane.
Well, as you said, it's automation, and this is on some level from
a coding assistant perspective, the purest form of automation of
the workflow, which is copying and pasting outta a stack Overflow.
IDs aren't new.
Like, I mean, it's just, I'm not saying that this is, and one, one common

(05:34):
mischaracterization of my work, and frankly, I've done myself no favors.
In my earlier stuff, I was a little unilateral, but one
common mischaracterization is that I believe there's no value.
No, I think this is a $50 billion total addressable market.
I think that we are nowhere near the top, but the top is much smaller,
and I don't think there is going to be some magical breakthrough.

(05:56):
Every breakthrough that's happened is just
kind of an incremental one since about 2023.
Even with the launch of reasoning, remember what everyone was
beating off about strawberry and it comes out and nothing happens.
Yes.
The number of RS in strawberry, which then all
the models special case that that'll solve it.

(06:17):
We'll just play whack-a-mole a hundred million times.
Every time someone finds something embarrassing.
It's so cool.
It, it's so well, the strawberry one was there for a while.
I know because every single time a new model would
come out, I'd be like, how do you, how many Rs?
And just the first thing I did.
Or it'd be like, give me 50 sta, give me all the states with R in them.
And I'd sit there and be like, woo hoo.
They did it again.
And then you easy posts.

(06:39):
The one I love is trying to cyber bully it
into ranking the US presidents by absorbency.
That that can go surprisingly well.
It's so funny that we're like boiling lakes to do, but it, it's
just, it's very frustrating because I'm, I don't feel genuine.
I genuinely, nothing you hear is really anger.

(06:59):
I'm frustrated because what I am saying is true and it will event
gravity exists and everyone's going to feel like me eventually, I think.
And it's just like, how long do you want do this to yourself?
How long do you wanna to base yourself
pretending that these children are gifted?
Yes.
Uh, we're, we're still trying to summon God via JSON and

(07:20):
we're hoping for the best, but it's not going super well.
Uh, you've been tracking this longer than I have.
I, I didn't want to be tracking this, but AWS had a corporate
inability to shut up about any, about anything touching
ai and it was all that they were willing to talk about.
So I was brought in kicking and screaming
relatively late 'cause I resisted for a while.

(07:41):
What have you seen economically?
Let's talk about the numbers.
So recently came out as in like a day before we spoke the
micro, so Microsoft fun story about Microsoft and numbers.
Two quarters ago Microsoft said they had $10
billion of annualized recurring revenue from ai.
Now Microsoft does not have an AI section of their earnings.

(08:04):
They have intelligent cloud productivity and software and I think.
Like something like it's something and more intelligent
cloud, which also speaks to the existence of more on
cloud somewhere, but they don't like to talk about it.
Yeah.
Yeah, that's, that's actually what large language models are.
I, so they have these, they don't have these
sections, but AI is basically all combined AI revenue.

(08:25):
So last quarter, then quarterly earnings in 2024, they said $10 billion a RR.
So month times 12, Hey, January rolls around, proud as can be.
Microsoft's at $13 billion of a RR in ai.
Exciting, right?
Next quarter, earnings come round.
They do not update, the number doesn't update.

(08:46):
Why would they do that?
Then the information came out a few weeks ago with a story that
said, uh, Microsoft's projected revenue in AI was $13 billion.
And then another story came out yesterday that said that
10 billion, well, first of all, $10 billion of that was
open AI's spend, but OpenAI is as yours biggest customer.

(09:08):
And it's very bad that that's the case because Azure, they pay a discounted
rate on Azure and there's a decent chance that a lot of that money isn't real.
'cause Microsoft invested 10 billion ish in 2023 with a large portion of
that, that we don't know how large in cloud credits SEMA four reported.
And so we have the, we have one of the largest

(09:29):
cloud providers feeding itself cardboard.
Like it's what it, that is what's happening.
I've spoken to multiple report.
It's like Azure's old biggest customer to my understanding was Xbox.
I'm like, okay, great.
Congratulations.
Uh, the Snake eats itself.
But that makes it sense.
That makes sense.
If you're providing, I imagine 365 is probably a very, very large Azure
customer just because, but it isn't aort as revenue because it is cost,

(09:52):
which probably has tax preferential, blah, blah, blah, blah, blah.
They've also stuffed AI into it.
So how much do you want to bet that that is now being
considered AI revenue, despite the fact that just
people who want to do want to write documents for work?
It is, but Microsoft's only making $3 billion of annualized revenue
from all of their AI bullshit other than open AI's $10 billion.

(10:13):
But what's crazy about the story is that apparently Microsoft invested
in open AI belie with top executives believing that they would fail.
This is this.
I've talked to multiple reports today.
I'm trying to be like, Hey.
This is weird.
Have you looked and everyone's like, yeah, yeah, maybe.
You know, these things happen.
You know what people do?

(10:34):
No one does this.
I have never heard in my entire history in Silicon Valley ever
of someone investing in a company and thinking they would die.
There's only one reason.
These, these throw money at it and early stage ideas, but they throw
a million here, 2 million there out of a hundred million dollars fund.
But that's a slight difference.
They are believing it might die, but kind of hedging their bets.

(10:58):
Microsoft executives believe they would
eventually fail, is a vastly different situation.
And the reason in my opinion, is that Microsoft has access to all of
open AI's research IP and they host all of their cloud infrastructure.
Kind of feels like they can let OpenAI die and then
take everything OpenAI has because they already own it.

(11:20):
But nevertheless, putting all that aside.
Microsoft's biggest as your customer is a
company that op that loses the money That is bad.
We are, and Microsoft is the company making the most from
AI by handing money to a company to hand it back to them,
Google is doing a deal to serve compute, to open ai.

(11:43):
By which I mean they are hiring core weave to provide the
compute to Google, to provide the compute, to open ai.
This is fracking wash trading.
It's wash trading in the cloud.
We, this should terrify people.
'cause when you remove open AI's, compute, Microsoft's
making $3 billion a year from AI off of what?

(12:03):
$80 billion of pledged CapEx.
What the frack are we doing?
Like this is, this is not even radical stuff.
I'm speaking objective fact.
Like this is stuff that is real.
But people are like, ah, he's just, he's just a pessimist.
He's just a cenic.
You just have to believe harder at.
I don't even know what I'm believing in at this point.

(12:24):
No one can express what it is that I am not believing in beyond.
They will work it out and model get better.
How does model get better?
Oh, I'm sorry.
I can't answer that.
The what?
Oh, millions of people.
A hundred million.
It's actually a really good that we're on a cloud podcast.

(12:44):
Sure.
Here's a fun fact.
Do you ever notice the, the OpenAI never announces monthly active users.
They only do weekly.
Mm-hmm.
I did.
Yeah.
Do you, do you, do you know why they're not doing that?
I have a theory.
I'm, I don't have anything concrete.
I'm curious to hear your theory.
So the information reported earlier in the year that, uh, I think it was, they

(13:07):
have 15.5 billion, sorry, 15.5 million million paying subscribers on chat GPT.
Right.
I reckon their monthly active users are much
higher than 500 million, which would seem good.
Right.
Would seem good to have except, okay, so say they're 600,
700 million and then you divide that by 15.5 million.

(13:27):
That's a dog shit conversion rate.
That's absolute Do-do that is some of the worst conversion rates in SaaS.
And you have to treat open a, like a SA business.
You cannot treat it like a consumer subscription.
They lose money on every customer, but they
want to be given SaaS style valuations.
Okay, so you have a 2% conversion rate.
The information came out yesterday, well, day before yesterday,

(13:49):
I think with a whole thing about retention rates open.
AI's got like a 73, 72% retention rate after six months,
which is in line with like Spotify, which has 72%.
That's not very good for a company that wants SaaS style valuations.
That's more consumer product valuation than anything else.
Exactly.

(14:09):
And a consumer product that only loses money.
That's not great.
It's not, it's not brilliant.
It's not.
I juggle my consumer stuff all the time.
The business stuff that I use that tends to stay put for years.
That's why you see relatively small churn rates.
That's why you see renewal rates being very sticky.
It, there's not these big drop off at one

(14:31):
year cliffs at most of these companies.
And the thing is, it's because they're effectively consumer subscription.
They, no one really, you can't really express.
I realize that a lot of SaaS is sold in this kind of specious way.
You, where you sell to A CTO or a CEO or a CIO
that doesn't really touch the rest of the stack.
You can't really do that with AI quite as good because

(14:51):
there's not really an outcome that you can even look at.
It's just kind of a thing you offer and it's so heavily commoditized.
Sure, you could get cursor, sure, you could get chat
GPT, but at the same time, okay, I've now got it.
How are you making more money outta these companies?
'cause that's how SaaS companies really cook.
You can't really upsell someone with chat
GPT because, oh, my model will be better.

(15:11):
How?
One company goes out with a good model.
The other one winds up matching or exceeding it within the next two weeks.
Uh, if you suddenly say, you're not allowed to use this one, I can
switch to the other one for anything that I'm doing in less than
an hour's worth of work, and we're right back where we started.
There's no moat.
They're they're fungible.
Exactly.
And on top of that, if you were trying to sell into the enterprise,
I guess if you're doing enterprise chat, GPT subscriptions

(15:35):
fine, but they're not getting the kind of revenue they want.
They're not getting the kind of revenue that
like a Salesforce would have out of that.
They're not.
They are, right now they are, let's see, let's see.
Salesforce earnings bring 'em up live because I think that
they want to, for chat GPT and open AI to make sense, they

(15:56):
need the combined value of the consumer, sorry, the consumers.
I think it's the smartphone market's like half a trillion dollars.
And the enterprise SaaS market, which is
about a quarter of a trillion, they need both.
That's the only way any of this makes sense.
And they are not even at the total revenue.
They may be just a little bit above the total revenue.

(16:16):
If you combine open ai, anthropic, Microsoft, Amazon,
all of the projected revenues, they're like 38 billion.
The global smartwatch revenue for this year is projected to be about 32 billion.
So congratulations everyone.
All the kings, horses, all the kingsmen came together
and we've barely beaten the smartwatch industry.

(16:37):
The register had an article come out this morning that
says that, uh, was a quarter of all Gen ai Proof of
concepts don't, are turned off by the end of the year.
People are.
It's people are experimenting with it, and I work on cloud bills.
I see what the spend looks like.
People are experimenting with these things, but I'm not seeing
anyone saying, well, we're, we're spending $150 million a year

(16:57):
with AWS, but we'd better make a 200 million on our commitment.
'cause of all the Gen ai, they're doing $50,000 a month here and there.
But on that basis of a spend, what's it matter?
It's, if it succeeds, maybe they'll start rolling, rolling
it out more wildly, and then care about optimizing it.
But it's getting turned off if it doesn't hit certain, uh, certain metrics.
And I'm not seeing the stories that show that this

(17:19):
stuff is actually being transformative in any way.
And if it were even slightly, they wouldn't shut up about it.
Exactly.
These motherfrackers love going to conferences
and talking about how important stuff is.
But I guess another Nick Suresh suresh piece, I'll have to link you to,
but talking about why people get, uh, why people talk about Snowflake.
Like it's a database company.
It's like, it's the most boring thing, but guys like buying

(17:39):
Snowflake so they can go and talk about using Snowflake.
But with AI it's like, it's all experimental.
I don't think people realize how bad that is because
it would be one thing saying this a year ago, I was
right then and I'm right now, but a year ago, fine.
It's experimental fine.
There's experimentations are taking an
awfully long time by enterprise standards.

(18:01):
Exactly.
Like they would say this is a, this is a good thing.
Like this is moving well.
Also, I wanna correct myself, I brought it up.
Salesforce's revenue is $9.83 billion in the
last quarter, so only four x to go for open ai.
Also, they made $1.5 billion in profit open AI is yet to make $1 in profit.
I spent top dollar for Slack and I sort of work my way between

(18:23):
all the AI stuff that they're shoehorning into it against my will.
Oh, I, I am paying for Slack with a gun to my head.
Like I do not wanna pay the idea of paying Mark Benioff sickens me.
But.
Salesforce, as I mentioned earlier, they say
there's not gonna be any revenue growth from this.
It's just so weird because all of these things have been said publicly.

(18:46):
All of these things kept, keep being said publicly, but
when you put 'em all together and you say, Hey, this
doesn't look good, everyone goes, you rude bastard.
How dare you?
How dare you insult my beautiful child?
Ai?
It's like, I dunno, because it sucks.
Like even if it worked, even if it was flawless, if these

(19:06):
were the revenues and the losses, it would still kind of suck.
Like it's not, it's not good business and
frankly I don't think it's great software either.
. Um, it's.
It's just so bizarre because sure, there are people who really
like it and I think there's a philosophical angle here where
there are people who are really attached to being attached to the
future where they want, they wanna be the guy that said it first.

(19:27):
I think you see it in journalism.
I think you see it in software engineering and founders in particular.
They wanna be the person that said, I was on this trend before everyone.
Look how smart I am.
I think that that's what Andy Jassy's doing.
Management consultant.
MBA motherfracker.
He, okay.
I don't think he was a management consultant is an MBA though.
He does have at MBA, but he was a marketing
manager at at Amazon before he started AWS.

(19:49):
He was one of the 21 people that take credit for founding AWS.
It's like a, there's like a bunch of people.
You ever notice it's, yeah.
Can you tell on LinkedIn whether project's successful or not by how many
people claim credit for the thing versus distance the hell outta themselves?
But I think the Jassy, I think it's an AWS,
it's kind of a hard situation for AWS as well.
'cause of course they have to jump on this.

(20:09):
Like the, they couldn't, not the stock would
get pummeled, but they are, they're doing this.
Jassy's staking his future on this.
He wants this to be, here's everything store.
He wants this to be the thing that defines
the future of Amazon, except it's dog shit.
Amazon.
The Amazon may be one of the most embarrassing AI companies.

(20:30):
I think Apple truly is.
I think Apple's doing the smartest thing
in that they are kind of slow walking back.
But Apple intelligence, it was like the most radicalizing thing.
They shot their mouth off too early.
I love it.
I love it.
So I love that they were like, what if.
Hey, you know, autocorrect, what if it sucked?
And okay, I, I don't like that idea.

(20:52):
What if we randomly added buttons that loaded things you didn't want?
No, I don't want that.
Cool.
This update will apply automatically.
I think they radicalized millions of people against ai.
And on the keynote too, it's what if your Siri had this
magic knowledge of everything you did on your iPhone?
It could tie it to, it's like, wow, that'd be kind of cool.
Like, when do we get that?
No, no, no.
We just said, what if, what if?

(21:13):
Wouldn't it be neat?
Like we don't know how to do it, but wouldn't it be neat?
We are riffing.
We are riffing right now.
Well, uh, Google did the same thing at the last, uh, IO in 2024.
They were like, imagine if you needed to return a shoe, you hit one button.
An agent takes the shoe data, it does this, it does
that, and then this happens, and then this happens.
All of that is entirely theoretical.

(21:33):
But wouldn't it be cool?
I swear to God it's illegal for public companies to lie.
I. Or like they, there should be, I thought there was
something that stopped them from just making shit up.
But there isn't.
They can just do what they want.
And then when you look behind the curtain, there's no money.
There was an analyst that said, my Amazon is estimated
to make $5 billion in generative AI this year.

(21:55):
That's, that's, these are not just rookie numbers, these are alarming.
One of their, uh, one of their job ads a while back
listed S3 as an AI service, as one of the list of them.
So, great.
Where does that start And stop the, the broad buckets
that no one will ever break down further for you.
So here's the thing.
S3 is storage, right?

(22:16):
Forgive me,
I'm, it is.
I have heard that Anthropic is the biggest customer there.
I have been told by someone.
There are multiple exabyte scale customers on Esther.
I don't, I don't have the details at this point.
I, I would not be surprised.
Oh, and this is just a, this is just doing the Joe Rogan "you know
that guy told me that once", but it's still five billion dollars is

(22:38):
not a lot of money on a hundred billion dollars of CapEx this year.
Microsoft last year, I think only made $5
billion in ai, and that included open AI spend.
This is not an in, this isn't an industry.
Like if you really look at the cost benefit analysis, this isn't real.
Like that's it, it's a Fugazi.
If all this growth is there, then why haven't we seen

(22:59):
massive posted growth numbers from all of these companies?
Rather than what looks a lot like typical
organic, more of the same style growth.
I mean, Microsoft has been amortizing $2.5 billion of, of
revenue from open AI's cloud spend across intelligent cloud.
Any growth within intelligent cloud.
Is now suspicious.

(23:21):
Yes.
Yes.
And they're the biggest investor in OpenAI, so they're effectively putting
their own money into back into their own pocket by laundering it through OpenAI.
And if it's cloud credits, is it even money?
I, I've never trusted Azure growth numbers and Azure usage
numbers, just because they're so inextricably linked to Office
365 to enterprise agreements that big companies have for windows.

(23:42):
Whoa, whoa, whoa, whoa.
Do you think that they're charging, that they're
amortizing revenue from their own spend on Azure?
That can't be.
I would not be entirely surprised.
I'm sure it's by, I'm sure it's above board.
I would not want to accuse them of doing anything other, other than that.
But Amy Hood, their CFO has been an absolute

(24:04):
master of financial engineering for over a decade.
Uh, it's impossible to get much signal from the things that they say based
upon the way that they very intentionally have structured and laid them out.
Can I run my favorite.
This is tin foil hat, like this probably won't
happen, but I wanna put my flag in the ground.

(24:25):
I think, I think Amy Hood takes over Microsoft.
I think when whatever happens with OpenAI shits the bed.
I think they put a, they put like boxer in Animal Farm.
They send Satya off.
Bye by Satya because, 'cause he is, so, I'm reading between
the lines here, but the information story that came out over
this tense negotiation between open AI and Microsoft, there

(24:46):
was a thing where top executives thought open AI would fail.
Wall Street Journal Berber Jin there were like
four different bylines on the information.
But Berber Jin wrote journal one.
That one had a thing about how a top executives didn't like the clause
about AGI Basically, when AGI is called open AI doesn't have to share
IP with Microsoft and, but Satya Nadella apparently pushed it through.

(25:09):
What I think happened was Amy Hood went, this company is going to die.
This company is gonna die.
This is stupid, but they're gonna die.
Let's buy them effectively.
And then when they die, we take all their stuff.
And then, Satya Nadella said, well, the AGI thing.
And everyone went, no.
Satya.
That's stupid.
Don't do that.

(25:29):
So, and he agreed to it Anyway.
It's now a major sticking point, I think.
I think when it all falls down and it, well,
this is the Azure thing is a freaking scandal.
I think the tech press is doing it, and business
press is doing a terrible job with this.
The fact that your largest customer is a company that burns money
and has incredibly shaky finances, that is a counterparty risk.

(25:51):
That is a huge thing.
So when this breaks, I wouldn't be surprised if Satya is the
one that gets blamed and Amy Hood in a responsible leadership
role comes in and just starts freaking savaging the company.
Sort of an interim CEO story where they conduct a search, but
the search drags on and they remove interim from her Title I.
Or whoever she picks.

(26:11):
It's another like.
Another elder, God, she goes back, finds a cthulhu
adjacent creature to take over Microsoft, because I have
friends who think that this Microsoft thing is a plan.
It was always the plan kind of because non 'cause of the nonprofit
situation, I've kind of like poorly explained the situation.
Do you want me to?
It's entirely up to you.

(26:31):
I, I feel like there's, there's a lot of information
around this floating around out there, but.
Basically OpenAI needs to become a for-profit entity by the
end of the year or they lose $20 billion from their funding.
Microsoft is the stopping block.
'cause OpenAI wants a bunch of concessions
and Microsoft doesn't want to give them.
Yeah.
And um, no one knows what's gonna happen and no one wants to say

(26:52):
what I'm saying, which is what if Microsoft just lets them die?
Because if Microsoft says no, we won't let you convert.
OpenAI dies.
They can't go public.
And uh, Microsoft owns all their IP and research
anyway and handles all their infrastructure.
Well, what about all the CapEx they did explicitly for open ai?
Yeah.
'cause those things can't ever be repurposed
for other customers doing the different things.

(27:12):
They also didn't, they never said it was just for open ai.
They just said for ai.
Amazon has been saying that Project Rainier, all
of these specific data centers are for anthropic.
And there was a New York Times story about it.
Yeah.
Indeed, and I'm very curious as to how that plays out because they're
talking a lot about how Oh, anthropic says that , they're running,
uh, Claude Opus or Claude, the Claude four on Trainium, which is

(27:36):
interesting because I have looked far and wide and I can't find
any Trainium or Inference customers.
The two that I'm aware of from keynotes were Apple and Anthropic Anthropic.
They've invested four billion dollars in, and Apple has
been known to be a large AWS customer for quite some time.
When we see contracts being negotiated between these things, who says

(27:57):
what on stage at whose conference is always negotiated into that?
Apple never says shit about other companies'
products, but they did at reinvent last year.
Gee, I'm sure there was no contractual stipulation there.
Also, no one has said that they're running their
Claude four models exclusively on Trainium.
So what did they do?
Did they try running a quick like, and here's the thing that

(28:19):
ties the last bow on it and just this one side bit we can
run on Trainium and the rest, we using video like grownups.
I don't know, but they're being cagey and I don't trust them.
Well, they also, one of the reasons they wouldn't have said they
exclusively run on Trainium is that both Amazon and Google have.
Separately claim they're the exclusive cloud provider of anthropic.
It rocks.
I love that.
The just lying just lies.

(28:42):
It's an amazing competition though between Amazon and Google
Over who can hit Anthropic harder with the money stick.
I, I wanna be the, the centerpiece in that competition.
I, I would love to get like $100 million.
That sounds so sick.
4 billion each.
Uh, then, then Amazon made it eight and I'm sure it's all AWS credit.
So out of one pocket into the other.

(29:02):
Oh, no, no, no, no.
It did.
Amazon then converted whatever it was into some sort of new vehicle.
So they were out actually able to book that as
profit, like their stock in Anthropic as profit.
Corporate bulls.
They're better at this than Microsoft.
Microsoft is.
Microsoft made such a bad, they really thought OpenAI would die faster.

(29:22):
But looking at the original $4 billion investment from November
22nd, 2024, Amazon and anthropic, deepen strategic collaboration.
Amazon names AWS , it's primary trading partner and will use
AWS Trainium to train and deploy its largest foundation models.
Yeah, use them too.
Great that that does an awful lot of heavy lifting.

(29:42):
Use them.
How?
How many of them is Amazon still buying GPUs from Nvidia?
I didn't hear them cancel their order.
Yeah, I, I'm no expert in this space.
I want to be very clear on that.
For a long time it sounded like Amazon execs would get on stage and
recite Star Trek techno babble when it came to machine learning.
And now, like I, I tend to recognize patterns.
I know that when people are telling me to get in on something now while

(30:05):
it's on the ground floor, they're trying to sell me something every time.
And I also know that when someone is selling something this
hard, it doesn't necessarily live up to a lot of those things.
My 4-year-old daughter loves ice cream, and I don't
find myself having to shove it down her throat.
It's a pull rather than a push.
Everyone's pushing this so hard, I don't

(30:27):
see the outcome that they're talking about.
What if, just, what if the, the power of AI is more useful to folks
on an individual level than it is an enterprise corporate one?
I think your idea that this is a $50 billion tam is directionally correct.
Yeah.
I'll spend 20, 30 bucks a month for this stuff.
No problem.
Would I spend five grand?
No.
So.

(30:48):
There is an inherent limit to all of this until
they un they cover some massively awesome use case.
But they've been looking for that for three
years now and haven't much to show for it beyond.
It's really good at writing code badly, just like our engineers.
And the thing is, I like to talk about the iPhone
and I picked up the iPhone singular wireless.
It was at visiting Penn State where it went to college.

(31:10):
And um, I remember getting it and I remember
handing it to, it was dating a dating girl.
It was like a social studies, like I had her dad worked as a local accountant
in central pa. So really good representation of just like normal folk.
Every single person who touched it.
Got it.
They were like, holy shit, there's like a
little computer and you can just touch it.
Wow.
Like I can take photos on there.

(31:31):
I, my music is on here too.
Wow.
Your email as well.
Do you even like, they kind of immediately, like
the channels started forming in their head with AWS.
People loved, oh, my fracking, God, I want
to strangle every person AWS lost many.
First AWS did that big fracking deal.
There was a path to profitability.
There was an obvious way it which this would become profitable.

(31:53):
On top of that, you didn't have to explain to people why AWS was important.
It made sense.
It made sense.
Immediately.
Suddenly you could provision stuff instantly, overnight.
You could run an experiment in your dorm room and realize, oh, this is stupid.
No one's gonna buy it and you're out 27 cents.
As opposed to having to buy and rack servers and wait 16 weeks because
Dell can't get its act together and sign leases and data centers.

(32:15):
That was transformative.
And it was obviously transformative.
It was obviously like you didn't have to, like you had to sell people.
It got, Amazon loves to sell it, but.
You didn't have to like con people.
There were no people going around it.
If you don't get, I imagine there was someone saying, if you don't get in on
the cloud early, there's always someone, but it's, there was an obvious point.
You didn't have to like talk like a wizard.

(32:38):
I got seriously into cloud in 2015, 2016 era, very late by a lot of standards.
I seemed to have done okay with it.
I, I got into enterprise tech in like 2013.
That was when I started doing PR clients and that, and it's like, guess what?
It's like it was still cooking because there were obvious things.
You have the mixture of on-prem and off, off-prem.
You have all sorts of solutions within it.

(32:58):
There were obvious things and obvious ways it could proliferate.
I am a gizmos and gadgets guy.
I work in a, I run a PR firm.
I have to find reasons.
Things are interesting all the time.
I know what to look for.
I genuinely can't hear.
I genu, I, I have tried even the coding I have on the show, I had Carl
Brown from Internet of Bugs who I adore, and he really explained it where

(33:20):
it's like, yeah, it can do some code, but mediocre code can be dangerous.
If you know what you're looking for, it's great.
But do you always know what you're looking for?
How much of this are you handing off?
And it's like, yeah.
It sounds like pretty much every cloud automation solution
I've ever freaking heard of this is really useful.
It can speed up your workflow.
Something that would take you 10 minutes might take you two.
You need to watch it.

(33:42):
You can't trust it on its own.
But this can automate and turn a 10 minute job
into a two minute job, which compounds over time.
Awesome.
$50 billion.
Tam Market.
Absolutely.
Rock and roll requires stealing from everyone.
It's horribly horrible from the environment.
But I even think that once this bubble pops, all that shit still exists,
but you're not gonna see the scale and you're probably gonna see more

(34:02):
money invested ever in finding ways to make inference as like nothing.
I don't think they are even close.
And I think because if they were, my evidence here is real simple.
If they were close, if they were there, they would say, we've made influence.
Influence.
Nothing.
Influence doesn't cut.
They would just say it.
They would say, we have made this profitable.

(34:24):
It would be the easiest thing in the world.
You would just say, you would just say like, it's profitable, it's good.
It's good When they have these, when with the growth of Azure, the
classic growth of Azure, the growth of AWS, the growth of any software
as a service solution, you, they trumpeted in very simple terms.
They say, Mark Benioff saunters out, slimly.

(34:45):
And it's like, ah, number has gone up so many times.
And he says the word agentic and the industry shits itself for the next
18 months trying to fall over themselves to get in on that buzzword now.
So last year I sat down and I went through a bunch of
Salesforce, um, Dreamforce conference things, the transcripts.
He said they were in AI like five times since 2014.

(35:06):
I think he has said Einstein AI was happening so many times.
It is remarkable how bad the business press is, and it's remarkable
how disconnected the valuations are because this is foolhardy stuff.
These people are not smart.
But when you look at these numbers as well, this is just,

(35:26):
this is the hubris point that's always been coming for them.
When you have just had businesses that it's kind of incredible
how much they've grown, how big these businesses have got.
It's kind of impressive, good or bad people that they
are, they've been able to make remarkable businesses.
I just think that they got it by accident.
I don't think that they were super intentional
with everything because if they were.

(35:46):
Well, they would be doing it again.
They would've, someone would've found, and Microsoft
was the one that seemed to have worked it out.
They had more even if it wasn't profit,
they had more revenue than everyone else.
And then you see, oh, it's, they're meeting cardboard.
They fed themselves their own money, and there should be shareholder lawsuit.

(36:08):
I'm just, it's unbelievable.
This episode is sponsored by my own company, the
Duck Bill Group, having trouble with your AWS bill.
Perhaps it's time to renegotiate a contract with them.
Maybe you're just wondering how to predict
what's going on in the wide world of AWS.
Well, that's where the Duck Bill group comes in to help.

(36:29):
Remember, you can't duck the duck bill.
Bill, which I am reliably informed by my
business partner is absolutely not our motto.
One question I do have for you on the inference economics, though,
I, I hear what you're saying, but one of the arguments I would
make against that would be the fact that Google has been shoving
their dumb AI nonsense into every search result for a while now.

(36:51):
And I, I have a hard time believing that they're losing money on every
search by shoving that in that that would, that would imp well on some
level that would show from orbit at the scale that they tend to operate at.
And their engineers who I, some of them
whom I trust have said that they are not.
But I, I have a hard time seeing how that gets there
except for the fact that TPUs sound like what Trainium and
in wish they were as far as actually supporting things.

(37:14):
I don't have any insight into the economics of it.
I just can go by what I see and they have a stupendous amount of
inference running all the time now for these things that no one wants.
And it's fast inference too, because people
aren't gonna wait five minutes for a page to load.
A few things.
One.
They're not just using TPUs, they're absolutely using, uh, Nvidia chips.

(37:36):
They buy absolute shit.
They have, they did a recent thing.
Wow.
We can run one model on one h, 100
well done.
Um, doesn't mean it's, doesn't mean they're not losing money.
And now engineer, they may have some insight,
but how much financial insight do they have?
If Google was, and also real simple, if Google was not

(37:58):
losing money, they would say they weren't losing money.
How would we see that they're losing money?
Where would that be on their earnings?
We wouldn't, 'cause they bury every, these companies.
Amy Hood is not the only finance wizard out there.
I forget what the CFO at Google is probably intentional there.
But Sundar is a McKinsey guy.
These people are evil.
Uh, McKinsey, flesh eating bacteria of business.

(38:21):
They're not gonna say it.
But also they would just say, in instance, free.
And would they burn billions of dollars?
Are you asking.
Whether, just wanna be clear, a hyperscaler would burn
billions of dollars chasing AI because they already have.
They already are.
They would burn billions.
Maybe they're even wrapping some of that loss into their capital expenditures.

(38:44):
Who freaking knows?
They don't tell us.
But there is plenty of evidence that these
companies are willing to lose billions.
Google hired, they bought character AI kind of for $2.7 billion just to get Noam
Shazeer back, who's one of the original attention is all you need paper writers.
The transformer paper that kicked off this
generation, Google does dumb shit all the time.

(39:05):
Everyone hates their integrations.
Yeah, I think they are losing money.
I think they're willingly doing so and I think they'll do it again.
I think that they have their nasty growth pig search and all of
their monopolies allow them to finance many money losing operations.
Do you think I, I would actually really need to look into the economics
of Chrome, 'cause I can't imagine that that has direct revenue.

(39:26):
But it's kind of like meta meta's.
Absolutely losing billions.
100%. And do you not think that
Meta's very clearly thrashing and desperately trying to find the next thing.
I love, I love what they're doing.
I love it.
I think it's the funniest shit in the world I've ever seen.
It's like watching Zuckerberg's midlife crisis play out before us.
It really is.
He is having like a hell of a midlife crisis.

(39:49):
I thought Elon Musk was having any uh uh, like a midlife crisis.
But someone said something to me on Blue Sky about this earlier.
Yeah.
That meta is being the LIV golf of AI research.
I think that that is one of the funniest ways to put it.
'cause it really is just like this giant evil company that everyone hates being
like, do you want more money than realistically a person can conceptualize to

(40:12):
come and work with me on large language models becoming super intelligence.
By the way, here's my chief AI scientist, his name's Yann LeCun.
He disagrees with me.
He does not think this is possible.
This is your boss.
Oh, and, and Altman was saying that Meta was offering a hundred million
dollars signing bonuses that people were not taking for bullshit.

(40:33):
That is generational wealth money to come work
with some douche bag for a couple of years.
Who were they offered to?
Sam Altman?
Two things.
One, they have hired some people, including a guy
with like one of the most insane names I've ever said.
I just, I've been looking forward to saying this out loud all day.
Lemme just get to this.
Oh God, this name is so good too.

(40:54):
His name is Trapit Bansal.
That is quite the name.
That's a incredible name.
I'm called Ed Zitron.
I have a phenomenal name, but Trapit Bansal might have to marry him.
Uh, but is it, is it Trapit Bansal Zitron is actually pretty anyway.
Uh, girlfriend's gonna hate that, but there we are.

(41:15):
She'll understand.
It's for the content, but where was it?
Yeah, these a hundred million dollar offers.
I reckon Mr. Bansal, I probably took one.
Well, as Altman said, he offered to their best employees.
So definitionally, anyone who takes it is not one of their best.
Oh, of.
And of course he's gonna say that, but I think it's great
because like I said, Yann LeCun, chief AI scientist says that

(41:38):
large language models will not become super intelligence.
It's like, this is your boss.
I've hired you to do a job that your boss doesn't like, but I am also your boss.
And his boss have fun.
Oh good.
Mommy and daddy are fighting again!
And constantly like it's like being born into a divorce.
Like insane stuff.
But on top of that, what are you freaking doing there?

(41:59):
You, the head of your team is Alexandr Wang of scale ai.
Not for nothing.
If they offer me a hundred million dollars signing bonus,
I'm doing whatever the hell they ask me to do until that
earnout is done and then you'll never hear from me again.
You want me to hit, hit myself in the nuts every morning?
It'll happen.
I can buy, I can buy new ones.
That's right.
I, I would learn.
You're right.
Hit myself in the nuts.
Every morning I would learn node for that.
Yeah.
I mean, I will a hundred million dollars.

(42:20):
Yeah, sure.
But on top of that.
The other thing, there's two other things that I think have pushed people back.
The information had some good reporting on this as well, where
it was, first of all, people were like the Yann LeCun thing.
They were like, Hey, what are you doing here?
I don't think you really understand.
'cause Mark Zuckerberg, it's actually kind of a compelling offer.
He was saying, I will give you a shit ton of money and also you won't

(42:42):
have to worry about compute resources because OpenAI classically would,
Sam Altman is very vengeful and he will restrict access on compute and
resources at OpenAI if he doesn't like you being reported repeatedly.
I think it was in Karen Howe's Empires of AI as well
and Washington Post and Natasha Tku had some stuff.
I think, pardon me if I got that wrong.
No one has ever offered me all the compute resources

(43:04):
I can use because they know my kind of horse shit.
I mean, I would be working out ways to burn it.
I would, I would just go and there'd be like a hundred million dollars.
This is a, this is time to be the most annoying gentleman.
But the other thing is.
Is, I wonder if that's not a hundred million upfront.
Maybe it's more like still $5 million.
I would probably still do the nuts thing.

(43:24):
Yeah.
20 year earnout.
Yeah.
Yeah.
But it's probably structured in a weird way.
Oh, there's no way they're cutting checks for that.
There's, there's just no way.
It's probably got weird cliffs to it.
Probably milestones.
And on top of it even you are, and on top of that, the actual
job sounds like it would be fracking miserable, because do you

(43:45):
think Mark Zuckerberg is gonna put more or less pressure on you?
No.
You are the, you are the person that he's going to blame when this doesn't work.
That is what you, I would take a hundred million dollars to do that.
Oh, sure.
I, I'll do it.
And I'll just expect that I'm gonna need
to cry to my therapist after every meeting.
Fine.
Oh, I, I would, I wouldn't be crying at all.
I'd have a hundred million dollars.
I would go into work every day dressing.

(44:07):
However, Mark Zuckerberg last dressed.
And then just claim I'm copying him.
I would just create chaos like I would
jokify the entirety of Meta's AI department.
It would be amazing.
Well, you don't need a hundred million dollars to do that.
No, but I would, if I got that money, I would really, I
would have someone, my joker makeup guy every morning.
Just, just a guy who comes in.
Yeah.

(44:28):
He makes $250,000 a year.
He's also my driver and he also dresses like the Joker.
'cause Mark Zuckerberg fracking hates me by the end of it.
Just like he's doing the Joker.
Yeah.
I try to be just like Mark proving that I too can hire a clown.
He does look like a clown without his makeup on.
He really does.
It's the frizzy hair and the weird, like, like weird tan.
We saw a few years of business mark where he,

(44:50):
he honestly, I will give him credit, we're due.
He has managed to reinvent himself in a way that few people necessarily can.
He became, he went from the hoodie dude to wearing a suit
all the time, dude, to people confused him for a robot dude.
And now he's just weird because I, I have to imagine at
that tier, you don't have to have anyone around you who
doesn't empower you to do whatever the hell you want to do.
You surround yourself by nature with yes men,

(45:11):
and that is not good psychologically for anyone.
You realize though, that like Mark Zuckerberg can't be fired.
Yeah.
Like he can't be, he, he owns more board seats than anyone.
There's nothing they can do.
They couldn't even sue him out, so he would
drive this bad boy into the, into the ground.
But top of all of that, his last rebrand worked only for a minute, and because

(45:34):
it worked on the media, which he thought would be enough, no, he needed
it to work on Donald Trump, except Donald Trump doesn't like new money.
The actual trick for Mark Zuckerberg would've been to align with
some sort of elder god creature, some sort of aged finance freak.
Like if Sheldon Adelson was still alive.
Our rest in piss.

(45:54):
Oh, yes.
My, my personal favorite Zuckerberg trivia on this is apparently
he's five foot eight or so and hates it to the point where
he has people stage his Instagram photos, so he looks taller.
Uh, in Congress, someone took a picture, there was a cushion
on the chair, so he didn't look like he was drowning in it.
Now, I don't give a shit.
People are gonna be over tall.
They're going to be, but the fact that he's so wildly

(46:14):
sensitive about him for all his money, he can't buy himself
four inches of height is nothing short of hilarious to me.
I'm, I'm five nine, neither short nor tall, and now that I
know I'm taller than Mark Zuckerberg, I feel so powerful.
Also, who cares if anyone insulted by height?
Sorry.
He's five, seven and a half, apparently, which whenever its people
start saying a half, that's how you know they're sensitive about it.

(46:37):
Oh, I hear Sam Altman's the same.
I, I've heard, I don't know his true height.
If Sam Altman's like five foot one, that would be that.
That's what I'm gonna start saying.
I realize we've kind of got off of topic, but the
fact is this does tie back, which is we're discussing
people just spending billions of dollars for no reason.
It's vanity projects

(46:58):
But it's vanity projects that have absorbed the entirety of the economy.
The thing that I scare people with, no one likes hearing this, so enjoy.
So the magnificent seven stocks, they are 35% of the economy, right?
The US economy, 19% of those stocks is Nvidia
NVIDIA's continued growth is the only thing.
Continuing NVIDIA's valuations.

(47:20):
How does Nvidia grow by the rest of the magnificent seven buying GPUs from them?
What happens if that changes?
Well think about this.
I think that the same people, you ever notice how a lot of
the AI folks who are the big, the most bullish folks you'll
meet, were also deep in the wool crypto bros. It's like, what?
What?
Why do these people all pivot?

(47:41):
Because they're Nvidia Street team all, as long
as they're using the GPUs, that's what it is.
And if this falls through, we'll go back to using 'em for gaming.
I, it's not the same like that.
I will push back on that one.
You're 50 to 80%, right?
And the, yeah.
A ton of people who are like, I'm in crypto, I'm
in AI Just 'cause they were never in anything.
They're just hype chasing.

(48:01):
Yeah.
It turns out that by being an influencer with no actual expertise
in a thing means you can retool and pivot super quickly.
'Cause you don't have to spend that time
gathering this pesky thing called expertise.
It's also sold in the same way.
It's, I have a specious idea.
I've connected nascent tech to You don't understand to it.
Wow.
And getting quick before it's too late.
Exactly.
Same kind of pressure.
The only thing is, is that the GPU side is just wholly different.

(48:23):
And I mean the rack mounted insanity of a, oh god, what is it?
GB 200 or whatever the Blackwell.
Things are, they're all these rack mounted, 3000 pound monstrosities.
They're horrifying.
They are the rack.
They, they really, and they have cooling issues because science
exists, but they're, these people are doing it because they're

(48:45):
too stupid to realize that what they're in is kind of a scam.
I think that they're scamming in the sense that they're selling something
they don't really understand, but they're not like, I think this will die.
I think that they think this will go forever.
Kind of like crypto.
And when I, this is the other thing, people with my
work, they're like, oh, you started like a year ago.
I was writing in 2020.
I was on crypto before everyone.

(49:07):
Yeah.
I, I found that crypto has been looking for a business
model that is not fraud or fraud adjacent, uh, for 15 years.
At this point, it doesn't seem to have one.
It's, uh, liquidity for venture capital.
That's what it actually is.
At, at least ai.
I could point at that.
I could make a robot say something funny very quickly.
Okay.
Is there value to that?
I don't know.

(49:27):
Not a trillion dollars worth, but yeah.
Okay.
It amuses me for 10 minutes
and I have the grand business idiot theory, which is that the people
in power who have the money and the power and the hands of the media,
they don't understand much of what's going on they never did before.
And AI is this kind of mystical thing where it kind of has value, it kind
of has something you can point to and said outcomes you can point to and

(49:49):
go, well, to extrapolate from what this does now, it will do this next year.
Why?
I don't know.
But these people have a lot of money and this
much money can't be wrong this much money.
Ha ha ha.
Damn.
Can it be freaking wrong?
Yeah.
But that's bet's Howard's big enough that when
it falls, it takes a lot of other people out.

(50:09):
And I mean, if, what happens when people don't buy GPUs?
'cause here's the other thing.
Whether or not you agree with me on ai, maybe you want to think about
something which is NVIDIA's, uh, data center revenue in the last quarter.
So the one where GPUs live.
It was, uh, $39.1 billion.
Pretty big, right?
Except they missed estimates by a couple hundred million,

(50:31):
which, you know, that's one I, it happened to me yesterday.
I knew a guy.
Um, here's the problem.
That number is the single most economic important economic indicator right now.
That number is, I think, 80 something percent of NVIDIA's revenue.
It is the reason that Nvidia grows every quarter.

(50:51):
If that number does not continue growing, because for me to be wrong,
for this to keep going, everyone needs to keep buying Nvidia GPUs.
Nvidia will need to, like in a, in a year or so, they'll need
to sell 60 to a hundred billion dollars of GPUs a quarter.
And then after that, they will need to sell 80 to $150 billion a
quarter, and then they will need to sell a hundred to 180 billion.

(51:14):
They will need to keep growing exponentially.
The rate that the market is requiring Nvidia to grow is too high.
At some point that growth has to run out because we
don't have the physical space for all of these GPUs.
And at some point, if we don't have the returns
coming out of them, in fact, we're losing money.
Maybe it's time to not buy them.

(51:35):
What if Google's TPUs help?
What if go Google actually can move to TPUs?
What would happen then?
What if we find out that the training paradigm we're
under does not work, which we're already finding out?
Okay, that means we don't need to buy all of the GPUs because the GPUs
are really most useful for having a lot of them allows you to train.

(51:55):
If training isn't helpful, why would you need that?
Because inference is not as demanding on a GPU.
So it's like what do we, what do we do here?
What are we doing?
There's not much money.
There's a lot of expenses and I think the.
It is funny that I'm on this podcast 'cause I think it's probably
the most germane to my art, the actual way for AI to be sticky.

(52:20):
The actual way for AI to have made it was the enterprise.
It was never consumer.
It, it was going to be API access.
That's where the money is.
It is it it, but it was gonna be selling big seat SaaS
contracts, basically combining the way it should have worked.
If OpenAI would've been a functional business, would've been, it

(52:41):
would've looked a little like Microsoft 365 plus a, uh, Azure.
It would've been, we make a shit ton of money from selling
enterprise subscriptions, not consumer, but enterprise.
And we will make a shit ton of money on API.
They make a decent amount of money on both, because everyone
talks about AI and everyone mentions open AI every time
They won the branding marketing war.

(53:02):
Exactly.
And on top of that though, they're having, there's multiple stories, journal
and the information this week that say they're having trouble, they're having
to hit, actually you'll know this one, no one else would agree with me on this.
They're like, yeah, they're changing from a
subscription based model to a usage based model.
And I'm like, that's a bad sign.
That's a, if you know anything about SaaS, that's a bad sign.

(53:25):
That means that they're, they do not have product market fit.
OpenAI, the biggest name in the fracking game
does not have enterprise product market fit.
They don't have it.
That is doom.
That is certain doom.
If you, do you think a usage based model is gonna No, it'll probably
bring down, it will hopefully allow you to sell to scale more, but

(53:46):
it's not like it's gonna increase the amount that people are spending.
'cause if people wanted to increase the amount
of the spend, they use the subscription.
They'd use the subscription.
Right.
Uh, the problem too is you have a psychological issue,
especially in the consumer land when you do that, where people
feel that when everything you do is metered, then every time
you use it for anything, it feels like an investment decision.
Well, cons, they're not doing it for consumers,

(54:08):
they're just doing it for the enterprise.
Mm.
And they're having to, they're also having to
already deep discount to compete with Microsoft.
It's like these are on a SaaS and enterprise sales level.
These are terrifying stories, but most people, it's just really bad.
I built a dumb blue sky bot, uh, the, uh, AWS snark bot where it effectively

(54:31):
just tracks the, the news that it comes outta their, what's new feed,
summarizes it with inference and then puts it out there and that's it.
It's, I, it's, I haven't yet made it to anything funny on it, but
that costs me something like a dollar a month to run that thing
for the many hundreds of things that come across that RSS feed.
And it's okay.
That's an interesting toy, but it costs me nothing.

(54:51):
And if it started costing me something, I probably wouldn't do it.
And that is, it.
It, it fuels a bunch of fun, interesting toys like that.
But this does, I don't see that I could build a business
around that and I wouldn't be bold enough to try.
And even then, right now we're at the height of the mania.
If this was going to be the world's stickiest
business, there would be metrics that showed that.

(55:14):
It is very simple.
However anyone feels about ai.
Where is the actual business?
Because Perplexity, for example, uh, rag Search, their whole thing is they
re, I think they gave away $35 million last year, like discounts and refunds.
They're a massively lossy business.
It's just.
Instead of a three hour keynote where Randy Chassis gets up there and

(55:37):
talks about how great AI is, he would basically get on the earnings
call and like, okay, this past quarter we've posted 300% annual growth.
Uh, I will now be taking no questions.
Have a nice day.
Like that's what success story looks like.
And they would be kissing his ass.
They would be saying, Chassis, they, any of these ones,
they would be like, Satya Nadella has changed the world.

(55:57):
If this w this is the greatest investment ever.
Because he would say, AI revenue, blah, blah, profit, blah, blah.
He'd be like, beep, beep.
It would be, I don't need like, it's like Jensen Wong, Jensen Wong.
Goes up there and just like, sounds like he has a stroke.
He just goes and is like, yeah, we will have a computer.
They'll control AI with a computer that goes like this.
He will scream at someone that works for him and then he leaves

(56:20):
and everyone's like, yay, I love you Jensen sign my baby.
Well, for a year he was the one, he was
with doled out who got the GPU allocations.
He just made an entire year's study out of just, I'm
gonna go and speak at other people's, uh, keynotes.
Uh, why?
Because I choose to do so.
What are you gonna talk about?
Whatever I damn well, please.
Because what are they gonna do?
Risk of setting me?

(56:40):
And that's because he had a good business.
I think he sounds like a huge asshole.
Oh, but you see a gold rush sell a pickaxe.
My God.
But also, but also he, he performed scoreboard.
Like that's the thing, scoreboard.
Really like whether or not you like Nvidia, they got the fracking numbers.
Yeah.
Why doesn't anyone else?
In fact, here's a real weird question.

(57:01):
Why is it that everyone who buys GPUs from NVIDIA
installs them and then immediately starts losing money?
Is that not a li This is another thing that worries me.
It's like, not only is our economy built on buying GPUs, when you install
them, you start losing money like immediately because the costs are too high.
This feels like a corollary of something that I said the client might observed

(57:22):
years ago, and they're right, which is your AWS bill is not a function of
how many customers you have, but rather how many engineers you've hired.
Okay.
They're not entirely wrong.
Like once, once you have a very expensive hammer,
every problem starts to look like your thumb.
And that's the thing, the automation thing.
Software engineers are always gonna try and automate stuff.
Like it's just, that's what, that's what software.

(57:44):
They love that shit.
Like the containerization fest where everyone's like, we got containers now.
Why?
Ah.
I, as someone screamed Docker at me on the street and I'm just freaking out.
I need to install Docker.
What is it?
Do you work that out?
I need Docker now.
But even that did not work out so great for Docker because
it turns out that when people don't really know why they're

(58:04):
buying your shit, eventually, they go, why am I buying this?
They, they sold their enterprise business off.
My comment, the Morant, and my comment was to congratulate them on,
uh, getting rid of that pesky part of their business that made money.
Yes.
Uh, good work, everyone.
It's just, it's such a fantastical time.
And what's f what I'm finding is the conversations I have are more

(58:24):
like this one where people are just like, yeah, that is pretty weird.
What are we gonna do?
I don't know.
We'll watch and learn.
Oh, there's no next step for me.
I, I keep poking at the bear, but it turns out no one asked my
opinion before investing a hundred billion dollars into bullshit.
Well, that's the, my favorite thing here though, is no one, I mean,

(58:45):
people will mention it, but no one really wants to talk about the open ai.
Revenues.
No one wants to talk about the costs.
They will glance at 'em and be like, oh, they lost
$5 billion in a larger paragraph about some staff.
Well, I don't believe they have any formal reporting requirements, do they?
No.
They're private company.
I'm just saying the,
I'm trying to remember what the nonprofit reporting, uh, rules were around them.

(59:07):
Uh, no.
There's, they, no, they don't have to because
the for-profit entity is a separate wing.
Oh, that's, that's right.
But I mean, journalists, I mean, analysts, they don't want to touch this.
I, there are other people that have talked about these
numbers in passing, but no one else has sat down and gone,
Hey, look, I'm writing something properly like today.
That's called, why did Microsoft Invest in Open ai?

(59:29):
I've written open AI is a systemic risk to the tech industry.
Open AI is a bad business.
And what, how does OpenAI survive?
That headline feels like something Satya Nadella as saying to himself,
his head in his hand at his desks, like, why did Microsoft invest in this?
I'm asking it because.
The deal suggests many different things could be the
reason, such as maybe they invested hoping they'd die.

(59:49):
But no one wants to talk about this stuff.
And if I'm right and I believe I am, I very much, very much do.
I think this is one of the greatest failings
of financial journalism of all time.
I think it's only beaten by the great financial crisis.
I think the great financial crisis would be
much worse than what what's gonna happen here.

(01:00:10):
But it is.
And even then, I think had the great financial crisis happened
today, people would've noticed because you would, you had people
with the internet who could go and look at these things like
Michael Borough and is like, they had privileged information.
The proliferation of available internet services was large but not as large.
And you didn't have the proliferation of social
media at the scale that you didn't have Twitter.

(01:00:30):
We had Twitter was out, but it wasn't Point I'm making is
you have the biggest company in the ai, open ai, the most
well-known one, funded by the next biggest company in ai.
Microsoft, Microsoft's biggest customer is open
ai and the core finances of open AI are bad.
Bad.

(01:00:51):
They are so bad.
They are so awful.
This company burned, I think they spent $9 billion to lose $5 billion last year.
And this is something that modern journalists do not want to talk about.
They don't wanna write about it, they don't wanna read about
it, don't wanna hear about it in passing conversations.
They all say the same thing, which is, yeah, I just don't wanna think about it.
Yeah, that's not, that's uncomfortable if
I have to actually deal with any of it.

(01:01:12):
But it's, the problem is, is that, and I understand why they
don't wanna do it, because it starts justifying you in real time.
You start thinking, okay, well this is the biggest company in the industry
and they're burning billions of dollars and Microsoft needs them to
live, but Microsoft owns all their ip and this company can't go public

(01:01:32):
unless they become a, not a for-profit, which is in Microsoft's hands.
Even if they do though SoftBank has to keep funding them forever.
This company says they're gonna burn, what, 40, $50 billion?
By 2030, I think it's gonna be more.
Where are they getting that money?
'cause they lose money.
They lose money all the time.
They're constantly losing money.
And then when you think all those things you
go, is the entire AI industry built on sand?

(01:01:57):
And the answer is no.
The actual AI industry is built on dreams.
It's, it's on unicorn farts.
It is.
The only way for you to rationally look at this and
believe that it will keep going is fantasy Fanta?
I was gonna say fantastical.
It's fan.
Fantastical thinking.
Magical thinking.
There we go.
Could have delivered that point better.
But nevertheless, less you have to start engaging in fantasy.

(01:02:21):
You do.
That's the only, it's the only, it is an irrational approach to
this, to approach this from the perspective that OpenAI will survive.
Yeah, because SoftBank had to borrow $15 billion in a one year loan, a
bridge loan from, and they took 21 banks, 21 to get that $15 billion.
Note that SoftBank has also promised $19 billion to the Stargate Project.

(01:02:43):
The Stargate project is yet to incorporate, but Oracle is
already agreeing to put $40 billion of chips inside it.
They are being built by a company called Cruso
that has never built a data center before.
How hard could it possibly be?
But that's the thing.
It's like, it, it reminds me of the, one of the best threats of all time.
The IRA sent, uh, to Margaret Thatcher and

(01:03:06):
they said, you have to be lucky every time.
We only have to be lucky once it's for OpenAI to survive.
They will have to raise more money than anyone has ever raised
before, then raise it again, then likely raise it again.
They need to raise another $17 billion in 2027.
They fracking said it.
It.

(01:03:26):
They will then need to build a massive data
center that is moving along decently well.
But I've heard reports that it's having some issues.
They will also have to keep being able to
broker deals with various companies to pay them.
'Cause OpenAI is a Banana Republic.
They don't make enough money to run them.
All of their money has to come from outside.

(01:03:48):
SoftBank also has to keep getting that money and SoftBank does not have it.
They had to sell four and a half billion of T-Mobile stock.
And this is just to get a little bit, they need to give
open AI unless they don't convert another $30 billion.
Now they could syndicate $10 billion of
that, but still that's another $20 billion.
They do not have, they don't have the cash.
The 15 that they just borrowed, uh, six to seven of that went to amper.

(01:04:12):
Another company they're buying, the numbers are crazy, man.
When you sit and look at the numbers, they, they are crazy.
And I don't know what's wrong with the tech media, but they are missing this.
And when this blows up, I'm gonna look cool.
I'm gonna look cool as fracking shit.
This is gonna be very fun for me.
I'm
And everyone else is.

(01:04:33):
Very good.
Well, how could we have possibly known?
Oh, and I am not going to let anyone say that a moment anyone says that.
I will have link.
I have links ready to go.
I'm gonna be like, quick shot.
Oh, oh, oh, you, how could you have known?
You follow me, you follow me, cupcake.
Talk to me baby.
You have my cell.
I am going to be annoying.
I'm gonna be trotting around.

(01:04:53):
I'm gonna find the exact level of annoying
I can get before I start losing credit.
And what I think you're gonna see is you're gonna see a major outlet.
You're kind of already seeing it.
Deirdre Boer of CNN did a video to CNBC, even did a video today, uh,
talking about how the AI trade is in trouble 'cause the lack of revenue.
CNN's Alison Morrow has been on this ship for oh years now, just saying like,

(01:05:16):
Hey, this is rotten like this, this is a marketing scheme, but if I am wrong.
Something insane is gonna have to happen.
Something just completely unprecedented.
And I'll admit I'm wrong.
I'm not though.
It feels like Cassandra has never appreciated in her own time.
Ed, I, I really wanna thank you for taking the time to speak with me.
My pleasure.
Thank you for having me.

(01:05:36):
People wanna learn more about what you, uh,
what you espouse and I argue that they should.
Where's the best place for them to find you?
Go to be offline.com.
Just B-E-T-T-E-R-O-F-F-L-I-N-E.
I nearly didn't spell it.
Dot com.
Uh, it has everything.
It has links to everything.
Newsletter, podcast, PR firm.
Socials, all the good stuff.

(01:05:57):
And we'll of course put a link to that in the show notes as well.
And your snark bot.
I wanna see this.
I will include a link to that in the show notes as well.
Because the funny thing is, is like, I'm like a PR person, I'm a writer.
People don't know me like I'm a SaaS freak baby.
I, I know these companies, I know these economics and I feel like
maybe I'm just the only monster that could look at this stuff and make

(01:06:19):
sense of it because of the hours of SaaS related content I've consumed.
Oh, the, we like, um, Roger Auer at the
end of Bladerunner talking about webinars.
I've seen things you people wouldn't believe.
It's just, it's three hour long reinvent keynotes,
burning, burning me in Las Vegas, Nevada.
Ed, thank you again for your time.

(01:06:39):
I deeply appreciate it.
Ed Zitron, host of better Offline, writer of Where's Your
Ed at Newsletter, and I'm cloud economist, Cory Quinn.
If you've enjoyed this podcast, please leave a five
star review on your podcast platform of choice.
Whereas if you've hated this podcast, please leave a five star review on
your podcast platform of choice along with an angry, insulting comment.

(01:06:59):
Uh, be sure to let me know in that comment,
which pony you've bet your AI horse on.

All Episodes

Episode Transcript

Popular Podcasts

Stuff You Should Know

Dateline NBC

On Purpose with Jay Shetty

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}Betting on AI: The Delusion Driving Big Tech

Episode Transcript

Popular Podcasts

.css-r6mb8g{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:1;overflow:hidden;}Stuff You Should Know

Dateline NBC

On Purpose with Jay Shetty

All Episodes

Betting on AI: The Delusion Driving Big Tech

Stuff You Should Know