October 9, 2024 • 46 mins
As of last week, OpenAI is now worth $157 billion - yet below the hood is a far darker story. In this episode, Ed Zitron explains the cold, hard truth - that OpenAI is a terrible business that burns billions of dollars, and its failure to scale its cloud business tells a dark tale about the wider generative AI industry.
---
LINKS: https://www.tinyurl.com/betterofflinelinks
Newsletter: https://www.wheresyoured.at/
Reddit: https://www.reddit.com/r/BetterOffline/
Discord: chat.wheresyoured.at
Ed's Socials:
https://www.instagram.com/edzitron
https://bsky.app/profile/zitron.bsky.social
https://www.threads.net/@edzitron
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:02):
A media Hello, and welcome to Better Offline. I'm your
host ed zich Tron. What as ever, please check the
episode notes for links to the things that I'm talking
(00:22):
about now. Frequent listeners know that I'm immensely skeptical of
generative AI, both as a technology but also as a business.
It's expensive, it's unreliable, it doesn't actually do that much,
and the real world use cases range from kind of
helpful to unhelpful to not really useful in any way,
shape or form, and definitely not useful enough to justify
(00:45):
hundreds of billions of dollars in spending. And I know
somebody is gonna email and say, but ed, my mate's
granny's uncle's dog uses chat GBT to brainstorm or something
else equally flimsy. And I really need you to put
all of this in context. Open AI loses five billion
dollars a year, and basically every big tech company is
blown past their emissions targets, with few signs that they'll
(01:06):
ever bother to try and meet them again, all in
pursuit of the piddliest or most convoluted use cases in history.
It's a goddamn fast and that's why generative AI is
such a regular topic on this show. I feel like
the tech industry is experiencing a kind of collective madness,
a delusion that's seeing companies like Microsoft and Google, but
both their companies and our planet on a technology that
(01:28):
continually fails to deliver on well anything I want to
say here, And I actually wrote down this supposedly massive potential,
but what was the potential? Ever? Can anyone actually tell
me what this was meant to do? Other than agi? Anyway,
Eventually I see this all falling apart, and I think
there's going to be a clamorous event when everything does
(01:49):
actually unravel. I return so often to Generative AI because
I truly fear the damage that this confrontation will cause,
and I'm concerned about the damage that's happening right now.
If I'm honest, And while my tone's off in acerbic
and I don't think you'd have it an any other way,
I don't really take much joy in watching these things burn. Sure,
I admit it's pretty funny thinking about satching a Delea
(02:11):
getting fired or sunded up a shy getting fired, or
both of them looking stupid, But it's not the same
thing Miglee's kind of tempered. When I stop and think
about what all of this means or will mean, I
should say, for the companies trying to get funding right now,
or for the companies that will try and raise money
after the bubble bursts, and for the tech workers that
will get laid off, either because Microsoft and Google want
(02:33):
to put more money into data centers and GPUs, or
because their generative AI bets have gone tits up and
they need to save money some other way. And so
I hope you'll forgive me for a somewhat more somber
tone to this episode. I want to again make a
clear headed analysis of where we are today and where
I think we're going. And the reason I'm retreading this
path is because over the last month or so, we've
(02:55):
seen some really really alarming signs about the accelerating crisis
in generative AI, whispers that the party will soon come
to a halt. Now, you may remember a few episodes
back that I started talking about pale horses of the
AI apocalypse, signs that things were falling apart, like layoffs
at AI companies, price increases or decreases, internal discord at
these big AI companies, speciously impressive yet kind of flimsy
(03:19):
product announcements and so on and so forth. Since then,
I've kind of been proven right several times. Multiple pale
horses have emerged. There was a big, stupid magic trick
to impress investors in the form of open aised rush
launch of its one and that model, by the way,
was code named Strawberry. Rumored pricing increases which hat GBT
in the future. Now they're looking at forty four dollars
(03:40):
a month by twenty thirty, and they're looking at potentially
thousands for models in the future, and then layoffs at
Scale AI, which is one of the biggest players in
AI training data, if not the biggest. And these are
all signs that things are beginning to fall apart. And
I think it's important to explain how precarious things are
right now despite all of the money stashing around, and
how dangerous the power of magical thinking really is. I
(04:04):
want to express my concerns about the fragility of this
movement and the obsessiveness and directionless that brought us here,
and I want us all to kind of do better now.
Don't mean you the listener, but if you remember the
media that's written something vague about AI and you've kind
of carried water for Sam Mortman. It is time to stop.
It was already time to stop some time Aguary, I
should just be clear, but now really is the time
(04:26):
to stop. But this week's a two parter, and I'll
be explaining things well in two parts. Obviously. First of all,
we're going to talk about open ai, the most recent
funding round, and how bad their business is and exactly
how worried you should be about them now. The second
part is going to be about something I'm calling the
subprime AI crisis, where I see a growing bubble of
(04:46):
companies integrating generative AI at prices well heavily subsidized by
big tech and heavily discounted by companies like open ai
and Anthropic. Should either of these companies get desperate or
I don't know, actually want to make more money than
they spend, they're inevitably going to have to raise the
prices to match their actual costs, which will likely make
the existence of many AI startups completely untenable. You know,
(05:10):
kind of like when the teaser interest rates on subprime
mortgages that expired during the Financial crisis, and people suddenly
started losing their houses. In any case, these are going
to be meaningful, yeah, kind of brutal episodes, in part
because while it's enjoyable to watch big tech burn or
get embarrassed, there's a real human cost, like I said,
and the damage to our environment is already being done
(05:31):
and might not even stop when Generative AI does. And
whether Microsoft and Google and the other big generative AI
backers slowly wind down their positions or cannibalize their companies
to keep open AI and Anthropic alive, I'm convinced that
the end result is still going to be the same.
I fear tens of thousands of people will lose their
jobs and much of the tech industry will suffer as
they realize that the only thing that can grow forever
(05:53):
is a cancer. I'm going to paint you oblique picture
not just for the big AI players, but for tech
more widely, and for the people to work at these companies,
and tell you why I think the conclusion to this
sordid saga, as brutal and damaging as it will be,
is coming sooner than you think. Let's begin, as have
(06:14):
explained in agonizing detail in the past in my newsletter
Open AI will have to continue to raise more money
than any startup has ever raised in history, in perpetuity
just to survive. They're going to burn five billion dollars
this year, and yes, that's after they make their revenue
and costs are only set to increase over the next
few years they develop bigger models. Their business as it
(06:35):
stands is completely untenable, and I'm going to explain why
in a bit. Nevertheless, that's why open Ai, the ostensible
nonprofit that may soon become a for profit who bloody
knows at this point, raise the funding round at evaluation
of one hundred and fifty seven billion dollars, bringing in
a reported six point six billion dollars in cash, along
(06:55):
with opening a four billion dollar revolving credit line from
banks like JP morgan See and a bunch of others
that should know better after the acquisition of Twitter. Investors
in the funding ground included Josh Kushner's Thrive Capital in
Video and of course Microsoft, who were somehow undeterred by
the abrupt exiffs open ai cto mirror Marati days before
the round closed. Also joining the frame was soft Bank,
(07:18):
the Japanese investment company with a drag record of making
some of the worst bets in history, and they dumped
five hundred million dollars into open Ai, and that alone
should start getting people a little worried. You see, soft
Bank's founder, Masayoshi Son is what we in the business
call the dumbass. I realized this is a very broad
brush to paint somebody with. But let me explain. Soft
(07:41):
Bank's vision fund dumped sixteen billion dollars into we Work,
which by the end of last year, by the way,
was worth forty four million dollars, and nearly a billion
dollars into wire Card, which turned out to be an
elaborate fraud. With its leadership. Are the facing criminal charges
or becoming fugitives from justice in Russia. Soft Bank has
become effective synonymous with burning cash on terrible bets and
(08:02):
lost over thirty two billion dollars in the last three years.
The problem that open Ai faces is that they need
an absolute shit ton of money, and generally the only
companies that need that much are the ones that may
not have a sustainable business model. Open Ai demanded a
minimum investment of two hundred and fifty million dollars from investors,
and while regular American vcs might have that much, and
(08:25):
indeed had in the case of Costler and Kochume Management,
there aren't enough of them to fill out a six
point six billion dollar funding round. And because open Ai
needed that much, not wanted needed, their only real choices
were to go to Masayoshi Son, who's a man who
put tons of money into stupid things and believes he
was put on this earth to make digital superintelligence and
(08:45):
I have a link to that. By the way. It's insane,
all I mean. There's also another choice, you know, the
far dodgier investors in the Middle East m mg X,
one hundred billion dollar investment fund backed by the United
Arab Emirates to invest primarily in AI and semiconductor companies.
This should also be a big warning sign that things
(09:06):
are going poorly, because absolutely nobody raises from the UAE
of the SuDS because they want to. They're the place
you go if you need a lot of money and
you're not confident anybody else is going to give it
to you. Other aspects of this deal are also a
bit worrying, one of them being that one of the
founding partners of MGX is the Sovereign Wealth Fund of
Abu Dhabi, which brought in around five hundred million dollars
(09:29):
into Anthropic in twenty twenty three. Kind of a conflict
of interest. Do you think any of these people actually
goddamn care though they're all working to the same rot economy.
Nonsense fucking Anyway, as I mentioned previously, open Ai now
has access to that four billion dollars in debt from
banks in that revolving credit facility too, and that's also
not brilliant for a company that just burns cash. I
(09:52):
don't know what the interest rate on it is, but
I can't imagine it's great. Generally, revolving credit lines are
worse interest rates, So that's great, I guess. But open
ai is desperate. Fundraising comes down to one one thing,
one really obvious thing. They need so much money. They
needed like ten billion dollars here, and they kind of
(10:13):
got it in this wonky way. But they need that
money to survive because, like I said, they lost five
billion in twenty twenty four, and that figure is likely
to increase as more complex models demand more compute and
more training. Data and Anthropic CEO Dario ama Day predicted
that future models may cost as much as one hundred
billion dollars to train and open ai. By the way,
(10:35):
they may have just raised the biggest round in history.
They're going to have to raise another round again soon.
And I'll explain why. While open ai succeeded here, it
was only after a fairly arduous process where they'd already
tried to raise a hundred billion dollar valuation earlier in
the year, and that specifically turned off investors because of
the huge price tag. And there's this, by the way,
(10:58):
to quote the information, there's a growing concern over the
overvaluation of generative AI companies. Oh you fucking think do
you think that there's a concern here? Well, after this podcast,
you're going to be really, really concerned. I guarantee anyway
to get the round done, open ai committed to converting
itself from a nonprofit to a for profit entity. Failure
to do so within two years will see their funding
(11:19):
converted into debt, which by the way, will be the
kiss of death for a company that only ever loses money.
The deal, per Reuter's, also includes provisions that would allow
investors to adjust the valuation or claw back their investment
should the transition from nonprofit to for profit fail. And
I mean this is worrying because I don't think there's
(11:39):
any historical precedence of a company, a nonprofit at least,
that had this much money then converted. And what complicates
that as well is converting a business of this scale
from a nonprofit to a for profit. Well, it involves
transferring assets, and because any assets previously donated to the
public benefit, as Alexander Reed, partner at Baker Holsteller, is
quote as saying The Wall Street Journal cannot be converted
(12:02):
into a private benefit without compensating the public for the loss.
Every asset, which includes things like patterns and other intellectual property,
would have to be paid for. Well, it's hard to
say for certain, it's likely this deal and the process
of untangling open ai from its nonprofit beginnings is really
only going to add to open AI's cash problem. And
(12:22):
it's not like that was a good problem to deal
with in the beginning. And now I will do a
thorough review and analysis of open ai because I want
you to see exactly how stupid things have become in
the tech industry and how ridiculous all of this writing
this boiled my blood, by the way, because when you
see how lossy this company is, when you really ingest
(12:43):
how bad open ai is and how stupid it is
that we've let this happen, and we meaning the tech industry.
You then will think on your own lives and be like, wow,
you think I could pay my rent like late, or
like ask a company to pay my rent because my
job involves me just burning piles of cash. No, you
wouldn't be able to lose money for a business. You
(13:06):
and I we have to operate normally, not sam Oltman,
though sam Oltman doesn't have to be normal. Sam Altman,
like all of the gilded assholes of the tech industry,
is allowed to live in this fantasy land based on
marketing high Now, before we go into it, though, I'd
like to enjoy one of the following lovingly crafted advertisements
(13:26):
that in no way seem out of place from anything
I'm saying. And we're back, So to put it bluntly,
open ai is an absolute dog of a company. On
(13:48):
September twenty seventh, The New York Times reported that open
ai would lose five billion dollars in twenty twenty four,
a number which the Information had estimated back in July,
and the open Ai expected to raise the price of
chat GPT plus its premium product to twenty two dollars
a month by the end of twenty twenty four and
are remarkable forty four dollars a month by twenty thirty. Interestingly,
(14:09):
and worryingly, the Times also confirmed another hypothesis of mine
that and I quote fundraising material also signaled that open
ai would need to continue to raise money over the
next year because its expenses grew in tandem with the
number of people using its products. In simpler terms, open
ai may have raised six point five or six point
six billion in funding literally a week ago, but they
(14:31):
need to raise more, probably the same amount or more
I'm going to say, within the next six months. And
The Times also reports that open ai is making internal
revenue estimates I would describe as this is a technical term,
absolutely fucking ridiculous. Open aiy's monthly revenue hit three hundred
million dollars in August, and the company expects to make
(14:53):
three point seven billion dollars this year. The company will,
as mentioned, loose five billion dollars anyway, and I'm going
to keep minding you of that. Yeah. Open ai says
that it expects to make eleven point six billion dollars
in twenty twenty five and an astonishing one hundred billion
dollars by twenty twenty nine, a statement that is so
egregious that I'm surprised it's not some kind of financial
(15:14):
crime to say it. For some context, by the way,
Microsoft makes about two hundred and fifty billion dollars a year,
Google about three hundred billion dollars a year, and Apple
about four hundred billion dollars a year. So yeah, I
guess by twenty thirty that's how big open ai. Well
you've well, you fucking talk. It drives me insane. And also,
by the way, all of those companies make more money
(15:35):
than they spend. Open ai, on the other hand, spends
two dollars and thirty five cents to make one dollar.
If you remember one thing from this podcast, it's that
every dollar they make they have to spend two dollars
and thirty five cents to get. It drives me insane. People,
we don't have to live in these conditions. How do
(15:56):
I do this? How do I get to burn money?
What the no? Seriously, open ai loses money on every
single transaction, every single time that somebody uses chet, GPT
or plus or connects one of their models, and while
it might make money selling premium subscriptions, I severely doubt
those subscriptions are turning a profit and they I really
(16:17):
do think losing money even on their power users. And
as I'll get into the next episode, I believe there's
also a SUBPRIMEI crisis brewing because open ai is API
services which let people integrate their models into products, are
currently priced to attract customers in scale, and increasing these
prices to match the actual cost will likely make this
product unsustainable for many businesses currently relying on these discount
(16:40):
of rates, and that's if they have any usage. But
I'll also get to that later. Now, if you anything
like me, you have somebody who's told you that open
ai is a growth business and that it will just
turn the knob to make it self profitable, much like
Amazon Web Services, which is Amazon's cloud computing product, did
years ago. Now, I just want to be clear. If
(17:01):
you're listening and you think that you're just wrong, you're
just completely wrong. You live in a fancyland. Wake the
fuck up. I'm sick and tired of hearing this crap.
Open ai is nothing like Amazon Web Services. First of all,
open ai owns none of its infrastructure as everything everything
is run on Microsoft's as Your Cloud. Secondly, Microsoft, also
(17:22):
as part of an open ai funding around from twenty nineteen,
has full access to all of open AI's pre agi research,
which is all of it, by the way, and full
license to sell and integrate all of their technology. They
have complete free reign over open aiy's intellectual property. Thirdly,
Amazon Web Services did not immediately start with a deficit
(17:44):
so great that it required raising more money than's ever
been raised in the history of tech. Where still open
AI's technology doesn't get cheaper as it scales. In fact,
it does the opposite. This is not like other businesses.
You want to compare this to stop saying this to people.
It's ridiculous. And I know why people say this, and
(18:04):
I am ranting, and you just have to forgive me.
I know that people don't want to believe that this
much money can be wrong. It can be. It is
currently being wrong. You are watching and listening to it
being wrong. This is not like Google Cloud. It is
not like the early days of the Internet. It is
not like Amazon Web Services. It's nothing like them at all.
(18:25):
There is no historical example like this, not even Uber
Uber at least had a business model. In the worst
year of its life, Uber lost I think six point
six billion dollars or something like that. So about one
point six billion dollars in open ai lost in twenty
twenty four. Right, you want to know what year that was?
Twenty twenty the year when people couldn't go outside, and
(18:47):
there are many years after they probably should have stayed indoors. Two,
But that's a different podcast. The point I'm making is,
if you've got a comfortable little thing in your soul
that's saying this is just like things I've seen before,
they'll turn it around. They absolutely And now I'm going
to break down exactly how untenable open ai is with
a few statements for starters. Open ai is trying to
(19:09):
hit eleven point six billion dollars of revenue by the
end of twenty twenty five, and to do that it
will have to triple its revenue and its current cost
of revenue, so two thirty five to make a dollar,
open ai will spend twenty seven billion dollars to hit
eleven point six billion dollars, even if open ai somehow
harves their costs, and by the way, in the next
(19:29):
year they're going to be building a brand new model,
which will cost them an absolute shit ton of money.
They will still lose two billion dollars in twenty twenty
five to hit these revenues. However, I must be clear,
these costs are going up. They need this company to
grow by three hundred percent and the only way they're
going to do that is by making a new model.
And the only thing that's really growing at this company
(19:51):
is they're free users, and those free users lose the
money every single time they use it. And even if
they added a two dollars price increase to chat GPT
plus and a similar price on their business plans like
teams and enterprise, these things aren't really going to significantly
move the needle without also adding a bunch of growth
on top. An open Aiy's latest GPT four model it
(20:14):
cost them one hundred million dollars to train, and more
complex models like the Orian one they're allegedly coming up
with and I imagine oh one which will get to
next episode, well, they're going to cost them an absolute
shit ter more to build. And also, by the way,
the information estimated back in July the open AI's training
costs would be about three billion dollars in twenty twenty four.
The costs are not coming down. People. Now here's another problem.
(20:38):
Buzzy tech companies need exciting things to get investors in,
customers jazzed and opening eye. Well, they haven't really had
anything exciting or truly important since the launch of GPT
three point five, and even their latest reasoning model, like
I said, I'll get to it next episode, it's not
been particularly impressive. That model, by the way, is much
(20:58):
more expensive to run and uses something called chain of
thought reasoning, which I'll get into. But nevertheless, it takes
more power to run because they're doing way more calculations
and they didn't even have a use case when they
announced it. What is going on with this company? But
putting that aside, open AI's products are also becoming increasingly commoditized,
(21:19):
with Google, Meta, Amazon, and even Microsoft building their own
large language models and other models to compete. Where still
these models are using effectively identical training data, which they're
also running out of, which makes their outputs and by extension,
the technology itself kind of similar. But most worrying of all,
and I'm really going to drum on this one later.
(21:39):
Open AI's cloud business, So the one where they connect
their models to other businesses and then the businesses sell
products to customers using those models technology. It's small, It's
really small. It's small to the point that it suggests
there's a fundamental weakness in the generative AI industry. It's
extremely worrying that the biggest player in the game, oh
(22:00):
only makes a billion dollars less than thirty percent of
its revenue from selling access to its supposedly innovative technology.
And fundamentally, I can't really find any compelling evidence that
suggests that open ai is going to be able to
sustain this growth. In fact, I really can't find any
historical comparison for this company. And I also feel like
(22:23):
open AI's growth is already stumbling. But if you don't
want to stumble through life bereft of joy or growth
or happiness, I really recommend you buy one of the
following things. And if you can't buy it, download it.
If you can't download it, go into a shop and
demand that they start stocking it. And if they'll just
give it to you, demand that you can give them money.
Listen to the advertisers. All right, we're back, So for
(22:58):
the next part of this episode, I'm going to begin
to exactly how open ai makes money, which is going
to require me to lay out its various businesses and products,
and it's going to have a lot of numbers. Trust me,
it's worth it, because this shit is insane. So, according
to the New York Times, open ai expects chat GPT
to make about two point seven billion dollars in revenue
in twenty twenty four, with an additional billion dollars in
(23:20):
revenue coming from other businesses using open AI's technology. Roughly
seventy three percent, or two point seven billion dollars of
open AI's revenue comes from selling premium versions of chat
jupt called plus Teams and Enterprise. Chat Jupt Plus is
marketed to individuals for twenty bucks a month, offering faster
response times, priority, access to new features, and capability is
(23:41):
not found in the free product, like image generation, which
you can get in so many different other places. Importantly,
open ai can use anything you'd do as training data
unless you explicitly opt out. Open Ai also sells access
to a team's product, which costs twenty five dollars a
user a month if pay annually, so three hundred bucks
a year per user and thirty dollars a user a month.
(24:02):
If paid monthly from this point on, your date is
excluded from that used to train open AI's models. By deform.
Open ai sells enterprise subscriptions that include an expanded context
window for longer prompts, meaning you can give more detailed instructions,
admin controls, and enhance support an ongoing account management. It
isn't clear how much this costs, but I found a
(24:22):
Reddit thread from a year ago that suggests it's about
sixty dollars a user a month, with a minimum of
one hundred and fifty seats on an annual contract. I
also believe open ai offers discounts for customers who buy
in bulk very standard software as a service business model.
A further billion dollars, or twenty seven percent of open
AI's revenue comes from open ai licensing its models and
(24:43):
services via its API. One thing you notice when you
look at this price page is that there's this huge
variety of models and APIs available, and that there's massive
amounts of variation in pricing too. I look into these
in depth in my newsletter, which by the way, is
called open ai is a bad business and it's worth
reading if you want to know, if not, just for
the fact that the pricing in its various services is
(25:04):
also super messy, but you don't need to know that
right now. Open Ai also makes about two hundred million
a year selling access to its models through its Microsoft
according to Bloomberg, where open Ai takes a twenty percent
cut of all revenue, and by the way, that's not
on top of the billion. This means that open ai
only makes about eight hundred million dollars a year by
selling access to its API. That two hundred million is
(25:26):
on top I want to add, how worrying this is
both open Ai and the larger generative AI market. If
open Ai, the most prominent name in all of generative ai,
is only making a billion dollars a year from selling
the shovels for the gold rush, what does that say
about the growth trajectory for this company all the actual
(25:47):
usage of generative AI products. That's the question for later
in the show. The first let's talk dollars. So as
it stands, open Ai makes the majority, like I said,
more than seventy percent of its revenue from selling access
to premium versions of chat GPT. A few weeks ago,
the Information Report that chat gpt plus had more than
ten million paying subscribers, and that it had one million
(26:08):
more than we're paying for higher priced plans for business teams.
As I've laid out previously, this means that open ai
is making about two hundred dollars million dollars a month
from consumer subscribers. But business teams is a very vague.
It isn't obvious what it means. It could mean the
team plan, it could mean the enterprise plan, and you
(26:29):
may think, well ed, couldn't you be horribly wrong? Couldn't
it be that they're actually ripping' they they got a
million enterprise subscribers. Wouldn't that be good? Wrong? Oh? The
New York Times has my back here. Based on their reporting,
we can actually get a little more specific. Chat gpt
plus is ten million customers, making open Ai around two
point four billion dollars a year. As ten million users
(26:51):
spending about twenty bucks a month, which equates two hundred million,
might apply by twelve two point four billion dollars. Dad
I swear I was listening in Mathson Times. This means,
by the way, that business users make up about three
hundred million dollars a year in revenue. Of twenty five
million dollars a month, which isn't really great. While ten
million paying customers might seem like a lot. Chat GPT
(27:13):
is effectively to generative AI what Google is to search.
Ten million people paying for this kind of table stakes,
and the idea that this company can triple that number
in a year is absolutely ridiculous. Open ai has been
covered by effectively every media outlet in the world. It's
mentioned in almost every single conversation about AI, even when
it's not about generative AAI, and has the backing and
(27:33):
marketing push of Microsoft and pretty much the entirety of
Silicon Valley. Chat GPT has over two hundred million weekly users,
and The New York Times reports that open ai has
and I quote, three hundred and fifty million people using
their services each month as of joom though it's unclear
if that includes those using the API. I would guess
that's chat GPT users collectively. This means that open ai,
(27:56):
the most popular company in the industry, in the most
prevalent industry in the industry, talked about to everyone and everywhere,
can only convert about three percent of its free users
into paying customers. Now, by the way, Eager listeners may go, well,
three twenty percent conversion, that's not bad, right, right, wrong. Generally,
free products don't cost this much to run. Chat GPT
(28:19):
itself is extremely expensive. Whether it's free or whether it's premium,
it's the same thing. So yeah, those three hundred and
fifty million people, they're more of like a parasite than anything.
The ninety seven percent who won't give them the credit card.
It's not great. And I think that lack of conversion
might be because chat GPT and chat GPT Plus are
(28:39):
really similar products. Chat GPT Plus that you use the
product more often and have access to numer models, but
there's no obvious new thing you can do. As a result,
you can't really upsell to plus unless you found someone
who's already worked out the use case for themselves. And
open Ai remains piss poor at actually marketing this product,
mostly because of the limitations of what a large language
(29:01):
model can do. As a result, most people diddling with
chat GBT will get what they need to out the
free version. I'd also argue that those willing to pay
for a Plus subscription are more likely to use the
platform way way more than free users, and because every
prompt on chat, GBT loses the company money. It's reasonable
to believe that paying users would be far more of
a burden on the system. While there's a chance to
(29:23):
open an eye, could have a chunk of users that
aren't particularly active. One cannot run a business based on
selling stuff you hope that people won't use. And no, no,
it's nothing like a gym or insurance. Stop it. When
I wrote the newsletter and I said that, I got
so many people who thought they were the cleverest people
at all school. Oh, it's like a gym. It's like, sure,
it's not the same. Stop it. You're not clever. No
(29:44):
one's impressed, And I should be clear. There's also a
reason why enterprise customers are generally more desirable than private individuals.
As with any other consumer centric subscription product, regular customers
are far more likely to cut their spending when they
don't feel like they're getting value from the product, or
when the household budget demands it, or there's economic problems.
Just look at Netflix. They were the biggest name while
(30:06):
they still are the biggest name in streaming, and they
lost a million customers in twenty twenty two because of
the cost of living crisis. These are real things that
will happen to open ai if they're not already happening.
Chat GPT plus is likely for many people, kind of
a lifestyle product, and the problem is that when people
lose their jobs, or inflation hikes happen, or cheaper things
(30:27):
come along that do much the same thing, these products
are the first to get slashed from the budget. And
that's before any arbitrary or silly or desperate price increase
is done by a company that isn't the good business. Honestly,
it's kind of remarkable that open ai found ten million
people to actually pay for chat gpt, But how do
you grow that to twenty million or forty million people.
(30:48):
I think they need like thirty million by the end
of twenty twenty five. How does that happen? Very worrying anyway.
At present, open ai makes about two hundred and twenty
five million dollars a month. That's two point seven billion
a year. By selling these premium subscriptions. To hit eleven
point six billion dollars in twenty twenty five, open ai
would have to increase revenue from chat GPT customers by
(31:09):
three hundred and ten percent. If we consider and forgive me.
I'm going to do some maths at you. The current
ratio of plus subscriptions, the teams and enterprise subscriptions that
eighty eight point eight nine percent is GPT plus versus
the business ones at eleven point eleven percent. Open Ai
would need to find eighteen point two nine million paying users,
and that's at the new price point of twenty two
(31:29):
bucks a month, while also retaining every single one of
the current subscribers to Chat GPT plus, who would also
need to renew at the same price point to hit
seven point four billion dollars or about six hundred and
sixteen million dollars a month. It would also have to
make an additional nine hundred and thirty three million dollars
in revenue from its business or enterprice clients, which again
(31:50):
would require open Ai to more than triple their users
and retain the current ones to triple these users. To
actually do this, Chat GPT has to meaningfully change, It
has to do so soon, or disclose multiple meaningful, powerful
use cases that are so impressive that eighteen million new
people agreed to give them twenty two dollars a month.
That's an incredible, and I would say insane goal and
(32:13):
one that I do not think this company is capable
of achieving. It would require making chat GPT something far
more compelling than it already is, in ways that I
can't even imagine. Just as the company lost multiple members
of its senior leadership team, it doesn't look good. There
needs to be a meaningful change. Now. I know I've
(32:35):
already given you a lot of worrying things about open
aiye about their very top heavy, subscription driven business, Why
it isn't going to grow, why they need to grow
it to grow faster than any one's ever grown, why
they need to stop losing so much money. But there's
something else to be worried about. There's something even flimsier
about this business, and it's actually something that's disastrous. It's disastrous.
(32:58):
I'm serious. It is simply disastrous. How little of open
AI's revenue comes from providing other companies that means to
integrate generative AI into their systems. It is astonishingly bad.
I cannot be clear enough here. Before I wrote this script,
before I wrote the newsletter that led to it, I believed,
in my heart of hearts that how OpenAI made their
(33:20):
money was from providing access to their models. That makes sense, right,
because this company's talked about everywhere. Their technology is meant
to be innovative. Everyone's meant to use it, right, that
makes sense. Surely most of their money would come from
people getting in on the revolution and then integrating the
revolution into their products, and then people using the revolutionary thing. Right. Wrong.
(33:44):
First of all, if open ai only makes a billion
dollars a year selling API access and thus let you
integrate the models into your products, it suggests that even
the biggest company in generative ai cannot find enough customers
to make its business viable. Secondly, it's yes, there's a
remarkably small amount of demand for GENERATIVEAI integrations. Or consider
(34:04):
another way that the companies connecting to open ai aren't
really making it very much money at all. If the
GENERATIVEAI revolution were really here, surely open ai, the household
name in large language models, would be rolling in cash
from their ABI business, not leaving it at less than
thirty percent of their and your revenue. This isn't the case.
(34:25):
It's so strange. So at the twenty twenty four open
Ai Dev Day event, it's developer conference, which took place
in October, first, the company said that over three million
developers are building apps using open AI's infrastructure. That works
out to about three hundred and thirty three dollars a developer,
and that's far less money than other companies which make
money by providing a service through their API actually make. Twilio,
(34:46):
a company that makes its money sending SMS, messages and
push notifications for companies, over the past quarter, made about
a billion dollars in revenue. That's what OpenAI made from
renting out its models and APIs over the past year.
Twilio also made roughly four billion dollars over the last
four quarters, which is more than open AI's projected revenue
for the entirety of twenty twenty four. As I mentioned before,
(35:10):
open ai makes two hundred million dollars of its one
billion dollar revenue from Microsoft reselling its models a twenty
percent cut. That suggests that Microsoft two is making about
a billion dollars from open aised models. So in the
event that open ai and Microsoft are making about a
billion dollars in annualized revenue by providing access to open
AI's models. Again, these are estimates based on current growth trajectories.
(35:33):
It suggests that there's only two billion dollars in annual
revenue coming from both of these companies combined, and that's
without wait is open ai making less money from open
AI's models than Microsoft is. Oh my god, this business
is a stinker. And this also suggests that generative AI
(35:54):
as a technology doesn't have a product market fit. According
to a survey by Andrews and Hory It's a VC
earlier in the year and I quote, the twenty twenty
three market share of closed source models was estimated at
eighty to ninety percent, with the majority of that share
going to open ai and open source ones would include
things like metas Lama model, which is kind of open source.
(36:15):
That's a whole other thing anyway. Another survey from IoT
Analytics published late last year suggested that the number might
look a little different, with thirty nine percent of the
market share going to open Ai and thirty percent going
to Microsoft. Assuming that the latter numbers are true or
even close to true, this suggests that the generative AI
market is really small. If open Ai, which dominates with
(36:38):
I'd wager about thirty percent of the market, is only
making a billion dollars a year from selling access to
its models at this stage. In this massive hype bubble.
There might not even be ten billion dollars of annual
revenue from companies integrating generative AI into their products. That's tiny,
and this should be where all the money is. If
this stuff is the future of everything, why is the
(36:59):
revenue stream so painfully weak. Open ai is making so
little selling access to their models, and it suggests that
despite this hype cycle, there either is an interest in
integrating these products from developers, or when these integrations are
actually in there, consumers aren't really into them or using them. Remember,
these products are charged on usage, and so it's possible
(37:19):
for generative AI to be integrated into a service but
not actually drive much revenue for open Ai as a
result of users not really caring. While chat gpt as
brand recognition, companies integrating open AI's models into their products
are far more indicative of the long term health of
both the company and the industry itself. Because if open
ai can't convince people to integrate and use this shit,
do you think others are succeeding? I mean, think about it.
(37:43):
Where have you seen some weird chatbot up here in
your life? Where have you seen an LM poked into
something you use? Have you been like, oh good, I
can't wait, or have you just kind of tried to
ignore it. Maybe you've interfaced with it. It sucks. There
are some decent products, like there are email clients that
summarize emails Microsoft teams Apparently people really like the summarization.
(38:03):
But we are meant to be describing the future here.
We're meant to be describing something exciting and sexy. Not huh.
This summarized the meeting for me interesting. But looking at
these numbers, it's hard to imagine how open ai will
more than triple revenue in the next fifteen months to
hit eleven point six billion dollars in sales. Furthermore, at
its current burn ray, open ai is currently spending, like
(38:26):
I said, two dollars and thirty five cents to make
a buck, meaning that eleven point six billion dollars in
revenue could cost as much as twenty seven billion dollars
to actually make. And as I previously mentioned, what costs
could foreseeably come down, All signs point to them increasing.
GPT four costs one hundred million dollars to train, and
more complex future models will cost hundreds of millions or
(38:47):
billions to train, as I mentioned the Information said that
training costs were would look like three billion dollars in
twenty twenty four, and I think it's fair to assume
that the new models are just as costly, if not
more so. Ye there's something deliciously ironic about all of
this that, despite a clear lack of user interest in
generative AI, open Ahi's global marketing push has succeeded in
(39:08):
making lots of people intrigued enough to try a completely
free service that only loses the money. While it's usually
great news that a product has three hundred or more
million free users, every book time somebody uses a service
like chet GPT, it loses the company money. And in
the case of chet GPT, good lord, they must be
losing so much. The Information estimated in July the open
(39:30):
Ai will spend around four billion dollars in server costs
in twenty twenty four to run chat GPT and host
other companies running their services using GPT in its other models,
effectively meaning that every dollar of revenue is immediately eaten
by the costs of acquiring it, and that's before your
factor in more than fifteen hundred people, as many as
seventeen hundred people now work at open Ai, which is
(39:50):
another one point five billion dollars or more in costs,
and other costs, by the way, are on top of that,
they've got real estate taxes, stock grunts. This is all
very bad and while open ay could potentially reduce costs,
they've shown no proof that they can. And they tried once,
well at least one got out, the so called iraqis
(40:12):
model from last year that they tried to show that
was meant to be more efficient if failed to launch.
That's not a good sign, is it. But there's one
more problem. There's still more problems, and this one, well,
this one was caused by their fundraising. You see, they
just raised six point five six point six billion dollars
in capital one hundred and fifty seven billion dollar valuation.
(40:33):
This means that all future rounds have to be at
that valuation or higher. A lower valuation, which is called
the down round, would make current investors quite pissy and
send a very loud signal to the market that the
company is having trouble because nobody thinks they're worth what
they used to be, which in turn would overwhelmingly suggest
open AI's only way to survive is to raise its
next valuation at a two hundred billion dollar valuation and
(40:56):
require yet another giant raise. I would say at least
ten billion. For context, the biggest IPO valuation in US
corporate history was Alababa, which debuted on the NASC with
a market cap of nearly one hundred and seventy billion dollars.
That figure is more than double the runner up Facebook,
which had a value of eighty one billion dollars. And
I want to give you some history on that one too.
(41:19):
Facebook's initial OPO was really bad because people were concerned
about mobile users and not monetizing them. Well, they won
that one, but the market's pulverized Facebook for that. Do
you think that they're going to be Oh yeah, so
your company loses five or more billion dollars a year
and you have no path of the profitability. Yes, put
you up on a Nasdaq sounds perfect. Fuck that. No,
(41:40):
they're not going to do that. How is this company
going to do? How is this company going to IPO?
And yet there's more problems too. I don't know how
openay is meant to convert itself from its weird nonprofit
structure into a for profit company. I don't know if
it's possible, it won't be easy. And what's crazier is
and there's so many little things like this in this
(42:00):
story where you're like, people are going to look back
on this and think, Wow, we were goddamn stupid. At
least I hope they did. But open ai has two
years from close to convert from a nonprofit to a
for profit or their funding will convert into debt. I
think it's six to nine percent interest. Hey, this isn't good. Also,
at some point open ai is going to have to
(42:21):
work out a way to go public, like I said,
because otherwise, why did people invest? Why did people bother?
They could do future secondary sales, And at some point,
if that's all they're going to be able to do
in secondary sales referring to selling private stock to another individual, well,
I mean at that point it's a Ponzi scheme. You're
just pulling in money to hand out money to other people. Eventually,
(42:45):
it's not good these This company has no past to profitability.
It doesn't have one. And regardless of what happens with
everything I'm mentioning, they still need to raise more funds.
Chat GPT's free version is an actual poison on open
AI's system. It's a marketing channel that burns billions of
dollars to introduce people to a product that only ten
million people will actually pay for. And open AI's future
(43:07):
depends largely on its ability to continue convincing people to
use it. So how does this continue? How does open
ai survive? I don't think it can. Open ai is
a disaster in the making, and behind it, it's a nastier,
shittier disaster, a lack of fundamental strength in the generative
AI market writ Large. If open ai can only make
(43:28):
a billion dollars as the leader in this market, and
really only eight hundred million because the other two hundred
million comes from Microsoft, it suggests that there's not really
developer or user interest in generative AI products writ Large.
Perhaps it's the hallucination problem, where it authoritatively states something
that isn't true. Well, maybe it's just that generative AI
(43:48):
isn't something that produces interesting interactions with a user. While
you could argue that somebody could work out a really
cool product, it's time to ask why Amazon, Google, Meta,
open Ai, Apple and Microsoft have failed to make one
one in the last two years? Where is it? Where's
the killer app? And no, it's not early days. Shut up.
(44:12):
More money than anything has ever raised has gone into
these companies. More attention has gone into these companies than
any other hyperscalar movement ever, even the metaverse. Where's the product? Man?
Where is it? And while this bubble can continue coasting
for a little while longer, nothing about the open ai
story looks good. This company's lost. They're bleeding money with
(44:33):
every single interaction with the customer, and they're flogging serve
software that's the best kind of useful and worst actively
harmful to the environment. Unless something significantly changes, like a
massive scientific breakthrough in energy or compute efficiency, I just
don't see how open ai makes it two more years.
And worse still, if my hypothesis about the wider market
(44:55):
is true, there might just not be a viable business
in make and selling large language models. While Meta and
open Ai might be able to claim hundreds of millions
of users on these services, I don't see any evidence
that these people will make them any money, and in fact,
I only see evidence they'll lose it. And if I'm right,
we're watching vcs in companies like Microsoft burn tens of
(45:16):
billions of dollars to power the next generation of products
and nobody really gives a shit about And in the
next episode, I'm going to go into the much bigger problem,
the subprime AI crisis, where everyone building on these platforms
is inevitably going to get rug pulled when Open AI
and Anthropic and other companies raise their prices, because if
they don't raise their prices, how are they going to
(45:38):
be able to afford to keep going. I don't bloody no,
but we'll get into it next episode. Thank you for
listening to Better Offline. The editor and composer of the
Better Offline theme song is Matosowski. You can check out
more of his music and audio projects Matttersolski dot com,
(46:01):
M A T T O s O W s Ki
dot com. You can email me at easy at Better
offline dot com or visit Better Offline dot com to
find more podcast links and of course my newsletter. I
also really recommend you go to chat dot Where's youreaed
dot at to visit the discord, and go to our
slash Better Offline to check out our reddit. Thank you
(46:22):
so much for listening.
Speaker 2 (46:24):
Better Offline is a production of Cool Zone Media. For
more from Cool Zone Media, visit our website cool zonemedia
dot com, or check us out on the iHeartRadio app,
Apple Podcasts, or wherever you get your podcasts.
A media Hello, and welcome to Better Offline. I'm your
host ed zich Tron. What as ever, please check the
episode notes for links to the things that I'm talking
(00:22):
about now. Frequent listeners know that I'm immensely skeptical of
generative AI, both as a technology but also as a business.
It's expensive, it's unreliable, it doesn't actually do that much,
and the real world use cases range from kind of
helpful to unhelpful to not really useful in any way,
shape or form, and definitely not useful enough to justify
(00:45):
hundreds of billions of dollars in spending. And I know
somebody is gonna email and say, but ed, my mate's
granny's uncle's dog uses chat GBT to brainstorm or something
else equally flimsy. And I really need you to put
all of this in context. Open AI loses five billion
dollars a year, and basically every big tech company is
blown past their emissions targets, with few signs that they'll
(01:06):
ever bother to try and meet them again, all in
pursuit of the piddliest or most convoluted use cases in history.
It's a goddamn fast and that's why generative AI is
such a regular topic on this show. I feel like
the tech industry is experiencing a kind of collective madness,
a delusion that's seeing companies like Microsoft and Google, but
both their companies and our planet on a technology that
(01:28):
continually fails to deliver on well anything I want to
say here, And I actually wrote down this supposedly massive potential,
but what was the potential? Ever? Can anyone actually tell
me what this was meant to do? Other than agi? Anyway,
Eventually I see this all falling apart, and I think
there's going to be a clamorous event when everything does
(01:49):
actually unravel. I return so often to Generative AI because
I truly fear the damage that this confrontation will cause,
and I'm concerned about the damage that's happening right now.
If I'm honest, And while my tone's off in acerbic
and I don't think you'd have it an any other way,
I don't really take much joy in watching these things burn. Sure,
I admit it's pretty funny thinking about satching a Delea
(02:11):
getting fired or sunded up a shy getting fired, or
both of them looking stupid, But it's not the same
thing Miglee's kind of tempered. When I stop and think
about what all of this means or will mean, I
should say, for the companies trying to get funding right now,
or for the companies that will try and raise money
after the bubble bursts, and for the tech workers that
will get laid off, either because Microsoft and Google want
(02:33):
to put more money into data centers and GPUs, or
because their generative AI bets have gone tits up and
they need to save money some other way. And so
I hope you'll forgive me for a somewhat more somber
tone to this episode. I want to again make a
clear headed analysis of where we are today and where
I think we're going. And the reason I'm retreading this
path is because over the last month or so, we've
(02:55):
seen some really really alarming signs about the accelerating crisis
in generative AI, whispers that the party will soon come
to a halt. Now, you may remember a few episodes
back that I started talking about pale horses of the
AI apocalypse, signs that things were falling apart, like layoffs
at AI companies, price increases or decreases, internal discord at
these big AI companies, speciously impressive yet kind of flimsy
(03:19):
product announcements and so on and so forth. Since then,
I've kind of been proven right several times. Multiple pale
horses have emerged. There was a big, stupid magic trick
to impress investors in the form of open aised rush
launch of its one and that model, by the way,
was code named Strawberry. Rumored pricing increases which hat GBT
in the future. Now they're looking at forty four dollars
(03:40):
a month by twenty thirty, and they're looking at potentially
thousands for models in the future, and then layoffs at
Scale AI, which is one of the biggest players in
AI training data, if not the biggest. And these are
all signs that things are beginning to fall apart. And
I think it's important to explain how precarious things are
right now despite all of the money stashing around, and
how dangerous the power of magical thinking really is. I
(04:04):
want to express my concerns about the fragility of this
movement and the obsessiveness and directionless that brought us here,
and I want us all to kind of do better now.
Don't mean you the listener, but if you remember the
media that's written something vague about AI and you've kind
of carried water for Sam Mortman. It is time to stop.
It was already time to stop some time Aguary, I
should just be clear, but now really is the time
(04:26):
to stop. But this week's a two parter, and I'll
be explaining things well in two parts. Obviously. First of all,
we're going to talk about open ai, the most recent
funding round, and how bad their business is and exactly
how worried you should be about them now. The second
part is going to be about something I'm calling the
subprime AI crisis, where I see a growing bubble of
(04:46):
companies integrating generative AI at prices well heavily subsidized by
big tech and heavily discounted by companies like open ai
and Anthropic. Should either of these companies get desperate or
I don't know, actually want to make more money than
they spend, they're inevitably going to have to raise the
prices to match their actual costs, which will likely make
the existence of many AI startups completely untenable. You know,
(05:10):
kind of like when the teaser interest rates on subprime
mortgages that expired during the Financial crisis, and people suddenly
started losing their houses. In any case, these are going
to be meaningful, yeah, kind of brutal episodes, in part
because while it's enjoyable to watch big tech burn or
get embarrassed, there's a real human cost, like I said,
and the damage to our environment is already being done
(05:31):
and might not even stop when Generative AI does. And
whether Microsoft and Google and the other big generative AI
backers slowly wind down their positions or cannibalize their companies
to keep open AI and Anthropic alive, I'm convinced that
the end result is still going to be the same.
I fear tens of thousands of people will lose their
jobs and much of the tech industry will suffer as
they realize that the only thing that can grow forever
(05:53):
is a cancer. I'm going to paint you oblique picture
not just for the big AI players, but for tech
more widely, and for the people to work at these companies,
and tell you why I think the conclusion to this
sordid saga, as brutal and damaging as it will be,
is coming sooner than you think. Let's begin, as have
(06:14):
explained in agonizing detail in the past in my newsletter
Open AI will have to continue to raise more money
than any startup has ever raised in history, in perpetuity
just to survive. They're going to burn five billion dollars
this year, and yes, that's after they make their revenue
and costs are only set to increase over the next
few years they develop bigger models. Their business as it
(06:35):
stands is completely untenable, and I'm going to explain why
in a bit. Nevertheless, that's why open Ai, the ostensible
nonprofit that may soon become a for profit who bloody
knows at this point, raise the funding round at evaluation
of one hundred and fifty seven billion dollars, bringing in
a reported six point six billion dollars in cash, along
(06:55):
with opening a four billion dollar revolving credit line from
banks like JP morgan See and a bunch of others
that should know better after the acquisition of Twitter. Investors
in the funding ground included Josh Kushner's Thrive Capital in
Video and of course Microsoft, who were somehow undeterred by
the abrupt exiffs open ai cto mirror Marati days before
the round closed. Also joining the frame was soft Bank,
(07:18):
the Japanese investment company with a drag record of making
some of the worst bets in history, and they dumped
five hundred million dollars into open Ai, and that alone
should start getting people a little worried. You see, soft
Bank's founder, Masayoshi Son is what we in the business
call the dumbass. I realized this is a very broad
brush to paint somebody with. But let me explain. Soft
(07:41):
Bank's vision fund dumped sixteen billion dollars into we Work,
which by the end of last year, by the way,
was worth forty four million dollars, and nearly a billion
dollars into wire Card, which turned out to be an
elaborate fraud. With its leadership. Are the facing criminal charges
or becoming fugitives from justice in Russia. Soft Bank has
become effective synonymous with burning cash on terrible bets and
(08:02):
lost over thirty two billion dollars in the last three years.
The problem that open Ai faces is that they need
an absolute shit ton of money, and generally the only
companies that need that much are the ones that may
not have a sustainable business model. Open Ai demanded a
minimum investment of two hundred and fifty million dollars from investors,
and while regular American vcs might have that much, and
(08:25):
indeed had in the case of Costler and Kochume Management,
there aren't enough of them to fill out a six
point six billion dollar funding round. And because open Ai
needed that much, not wanted needed, their only real choices
were to go to Masayoshi Son, who's a man who
put tons of money into stupid things and believes he
was put on this earth to make digital superintelligence and
(08:45):
I have a link to that. By the way. It's insane,
all I mean. There's also another choice, you know, the
far dodgier investors in the Middle East m mg X,
one hundred billion dollar investment fund backed by the United
Arab Emirates to invest primarily in AI and semiconductor companies.
This should also be a big warning sign that things
(09:06):
are going poorly, because absolutely nobody raises from the UAE
of the SuDS because they want to. They're the place
you go if you need a lot of money and
you're not confident anybody else is going to give it
to you. Other aspects of this deal are also a
bit worrying, one of them being that one of the
founding partners of MGX is the Sovereign Wealth Fund of
Abu Dhabi, which brought in around five hundred million dollars
(09:29):
into Anthropic in twenty twenty three. Kind of a conflict
of interest. Do you think any of these people actually
goddamn care though they're all working to the same rot economy.
Nonsense fucking Anyway, as I mentioned previously, open Ai now
has access to that four billion dollars in debt from
banks in that revolving credit facility too, and that's also
not brilliant for a company that just burns cash. I
(09:52):
don't know what the interest rate on it is, but
I can't imagine it's great. Generally, revolving credit lines are
worse interest rates, So that's great, I guess. But open
ai is desperate. Fundraising comes down to one one thing,
one really obvious thing. They need so much money. They
needed like ten billion dollars here, and they kind of
(10:13):
got it in this wonky way. But they need that
money to survive because, like I said, they lost five
billion in twenty twenty four, and that figure is likely
to increase as more complex models demand more compute and
more training. Data and Anthropic CEO Dario ama Day predicted
that future models may cost as much as one hundred
billion dollars to train and open ai. By the way,
(10:35):
they may have just raised the biggest round in history.
They're going to have to raise another round again soon.
And I'll explain why. While open ai succeeded here, it
was only after a fairly arduous process where they'd already
tried to raise a hundred billion dollar valuation earlier in
the year, and that specifically turned off investors because of
the huge price tag. And there's this, by the way,
(10:58):
to quote the information, there's a growing concern over the
overvaluation of generative AI companies. Oh you fucking think do
you think that there's a concern here? Well, after this podcast,
you're going to be really, really concerned. I guarantee anyway
to get the round done, open ai committed to converting
itself from a nonprofit to a for profit entity. Failure
to do so within two years will see their funding
(11:19):
converted into debt, which by the way, will be the
kiss of death for a company that only ever loses money.
The deal, per Reuter's, also includes provisions that would allow
investors to adjust the valuation or claw back their investment
should the transition from nonprofit to for profit fail. And
I mean this is worrying because I don't think there's
(11:39):
any historical precedence of a company, a nonprofit at least,
that had this much money then converted. And what complicates
that as well is converting a business of this scale
from a nonprofit to a for profit. Well, it involves
transferring assets, and because any assets previously donated to the
public benefit, as Alexander Reed, partner at Baker Holsteller, is
quote as saying The Wall Street Journal cannot be converted
(12:02):
into a private benefit without compensating the public for the loss.
Every asset, which includes things like patterns and other intellectual property,
would have to be paid for. Well, it's hard to
say for certain, it's likely this deal and the process
of untangling open ai from its nonprofit beginnings is really
only going to add to open AI's cash problem. And
(12:22):
it's not like that was a good problem to deal
with in the beginning. And now I will do a
thorough review and analysis of open ai because I want
you to see exactly how stupid things have become in
the tech industry and how ridiculous all of this writing
this boiled my blood, by the way, because when you
see how lossy this company is, when you really ingest
(12:43):
how bad open ai is and how stupid it is
that we've let this happen, and we meaning the tech industry.
You then will think on your own lives and be like, wow,
you think I could pay my rent like late, or
like ask a company to pay my rent because my
job involves me just burning piles of cash. No, you
wouldn't be able to lose money for a business. You
(13:06):
and I we have to operate normally, not sam Oltman,
though sam Oltman doesn't have to be normal. Sam Altman,
like all of the gilded assholes of the tech industry,
is allowed to live in this fantasy land based on
marketing high Now, before we go into it, though, I'd
like to enjoy one of the following lovingly crafted advertisements
(13:26):
that in no way seem out of place from anything
I'm saying. And we're back, So to put it bluntly,
open ai is an absolute dog of a company. On
(13:48):
September twenty seventh, The New York Times reported that open
ai would lose five billion dollars in twenty twenty four,
a number which the Information had estimated back in July,
and the open Ai expected to raise the price of
chat GPT plus its premium product to twenty two dollars
a month by the end of twenty twenty four and
are remarkable forty four dollars a month by twenty thirty. Interestingly,
(14:09):
and worryingly, the Times also confirmed another hypothesis of mine
that and I quote fundraising material also signaled that open
ai would need to continue to raise money over the
next year because its expenses grew in tandem with the
number of people using its products. In simpler terms, open
ai may have raised six point five or six point
six billion in funding literally a week ago, but they
(14:31):
need to raise more, probably the same amount or more
I'm going to say, within the next six months. And
The Times also reports that open ai is making internal
revenue estimates I would describe as this is a technical term,
absolutely fucking ridiculous. Open aiy's monthly revenue hit three hundred
million dollars in August, and the company expects to make
(14:53):
three point seven billion dollars this year. The company will,
as mentioned, loose five billion dollars anyway, and I'm going
to keep minding you of that. Yeah. Open ai says
that it expects to make eleven point six billion dollars
in twenty twenty five and an astonishing one hundred billion
dollars by twenty twenty nine, a statement that is so
egregious that I'm surprised it's not some kind of financial
(15:14):
crime to say it. For some context, by the way,
Microsoft makes about two hundred and fifty billion dollars a year,
Google about three hundred billion dollars a year, and Apple
about four hundred billion dollars a year. So yeah, I
guess by twenty thirty that's how big open ai. Well
you've well, you fucking talk. It drives me insane. And also,
by the way, all of those companies make more money
(15:35):
than they spend. Open ai, on the other hand, spends
two dollars and thirty five cents to make one dollar.
If you remember one thing from this podcast, it's that
every dollar they make they have to spend two dollars
and thirty five cents to get. It drives me insane. People,
we don't have to live in these conditions. How do
(15:56):
I do this? How do I get to burn money?
What the no? Seriously, open ai loses money on every
single transaction, every single time that somebody uses chet, GPT
or plus or connects one of their models, and while
it might make money selling premium subscriptions, I severely doubt
those subscriptions are turning a profit and they I really
(16:17):
do think losing money even on their power users. And
as I'll get into the next episode, I believe there's
also a SUBPRIMEI crisis brewing because open ai is API
services which let people integrate their models into products, are
currently priced to attract customers in scale, and increasing these
prices to match the actual cost will likely make this
product unsustainable for many businesses currently relying on these discount
(16:40):
of rates, and that's if they have any usage. But
I'll also get to that later. Now, if you anything
like me, you have somebody who's told you that open
ai is a growth business and that it will just
turn the knob to make it self profitable, much like
Amazon Web Services, which is Amazon's cloud computing product, did
years ago. Now, I just want to be clear. If
(17:01):
you're listening and you think that you're just wrong, you're
just completely wrong. You live in a fancyland. Wake the
fuck up. I'm sick and tired of hearing this crap.
Open ai is nothing like Amazon Web Services. First of all,
open ai owns none of its infrastructure as everything everything
is run on Microsoft's as Your Cloud. Secondly, Microsoft, also
(17:22):
as part of an open ai funding around from twenty nineteen,
has full access to all of open AI's pre agi research,
which is all of it, by the way, and full
license to sell and integrate all of their technology. They
have complete free reign over open aiy's intellectual property. Thirdly,
Amazon Web Services did not immediately start with a deficit
(17:44):
so great that it required raising more money than's ever
been raised in the history of tech. Where still open
AI's technology doesn't get cheaper as it scales. In fact,
it does the opposite. This is not like other businesses.
You want to compare this to stop saying this to people.
It's ridiculous. And I know why people say this, and
(18:04):
I am ranting, and you just have to forgive me.
I know that people don't want to believe that this
much money can be wrong. It can be. It is
currently being wrong. You are watching and listening to it
being wrong. This is not like Google Cloud. It is
not like the early days of the Internet. It is
not like Amazon Web Services. It's nothing like them at all.
(18:25):
There is no historical example like this, not even Uber
Uber at least had a business model. In the worst
year of its life, Uber lost I think six point
six billion dollars or something like that. So about one
point six billion dollars in open ai lost in twenty
twenty four. Right, you want to know what year that was?
Twenty twenty the year when people couldn't go outside, and
(18:47):
there are many years after they probably should have stayed indoors. Two,
But that's a different podcast. The point I'm making is,
if you've got a comfortable little thing in your soul
that's saying this is just like things I've seen before,
they'll turn it around. They absolutely And now I'm going
to break down exactly how untenable open ai is with
a few statements for starters. Open ai is trying to
(19:09):
hit eleven point six billion dollars of revenue by the
end of twenty twenty five, and to do that it
will have to triple its revenue and its current cost
of revenue, so two thirty five to make a dollar,
open ai will spend twenty seven billion dollars to hit
eleven point six billion dollars, even if open ai somehow
harves their costs, and by the way, in the next
(19:29):
year they're going to be building a brand new model,
which will cost them an absolute shit ton of money.
They will still lose two billion dollars in twenty twenty
five to hit these revenues. However, I must be clear,
these costs are going up. They need this company to
grow by three hundred percent and the only way they're
going to do that is by making a new model.
And the only thing that's really growing at this company
(19:51):
is they're free users, and those free users lose the
money every single time they use it. And even if
they added a two dollars price increase to chat GPT
plus and a similar price on their business plans like
teams and enterprise, these things aren't really going to significantly
move the needle without also adding a bunch of growth
on top. An open Aiy's latest GPT four model it
(20:14):
cost them one hundred million dollars to train, and more
complex models like the Orian one they're allegedly coming up
with and I imagine oh one which will get to
next episode, well, they're going to cost them an absolute
shit ter more to build. And also, by the way,
the information estimated back in July the open AI's training
costs would be about three billion dollars in twenty twenty four.
The costs are not coming down. People. Now here's another problem.
(20:38):
Buzzy tech companies need exciting things to get investors in,
customers jazzed and opening eye. Well, they haven't really had
anything exciting or truly important since the launch of GPT
three point five, and even their latest reasoning model, like
I said, I'll get to it next episode, it's not
been particularly impressive. That model, by the way, is much
(20:58):
more expensive to run and uses something called chain of
thought reasoning, which I'll get into. But nevertheless, it takes
more power to run because they're doing way more calculations
and they didn't even have a use case when they
announced it. What is going on with this company? But
putting that aside, open AI's products are also becoming increasingly commoditized,
(21:19):
with Google, Meta, Amazon, and even Microsoft building their own
large language models and other models to compete. Where still
these models are using effectively identical training data, which they're
also running out of, which makes their outputs and by extension,
the technology itself kind of similar. But most worrying of all,
and I'm really going to drum on this one later.
(21:39):
Open AI's cloud business, So the one where they connect
their models to other businesses and then the businesses sell
products to customers using those models technology. It's small, It's
really small. It's small to the point that it suggests
there's a fundamental weakness in the generative AI industry. It's
extremely worrying that the biggest player in the game, oh
(22:00):
only makes a billion dollars less than thirty percent of
its revenue from selling access to its supposedly innovative technology.
And fundamentally, I can't really find any compelling evidence that
suggests that open ai is going to be able to
sustain this growth. In fact, I really can't find any
historical comparison for this company. And I also feel like
(22:23):
open AI's growth is already stumbling. But if you don't
want to stumble through life bereft of joy or growth
or happiness, I really recommend you buy one of the
following things. And if you can't buy it, download it.
If you can't download it, go into a shop and
demand that they start stocking it. And if they'll just
give it to you, demand that you can give them money.
Listen to the advertisers. All right, we're back, So for
(22:58):
the next part of this episode, I'm going to begin
to exactly how open ai makes money, which is going
to require me to lay out its various businesses and products,
and it's going to have a lot of numbers. Trust me,
it's worth it, because this shit is insane. So, according
to the New York Times, open ai expects chat GPT
to make about two point seven billion dollars in revenue
in twenty twenty four, with an additional billion dollars in
(23:20):
revenue coming from other businesses using open AI's technology. Roughly
seventy three percent, or two point seven billion dollars of
open AI's revenue comes from selling premium versions of chat
jupt called plus Teams and Enterprise. Chat Jupt Plus is
marketed to individuals for twenty bucks a month, offering faster
response times, priority, access to new features, and capability is
(23:41):
not found in the free product, like image generation, which
you can get in so many different other places. Importantly,
open ai can use anything you'd do as training data
unless you explicitly opt out. Open Ai also sells access
to a team's product, which costs twenty five dollars a
user a month if pay annually, so three hundred bucks
a year per user and thirty dollars a user a month.
(24:02):
If paid monthly from this point on, your date is
excluded from that used to train open AI's models. By deform.
Open ai sells enterprise subscriptions that include an expanded context
window for longer prompts, meaning you can give more detailed instructions,
admin controls, and enhance support an ongoing account management. It
isn't clear how much this costs, but I found a
(24:22):
Reddit thread from a year ago that suggests it's about
sixty dollars a user a month, with a minimum of
one hundred and fifty seats on an annual contract. I
also believe open ai offers discounts for customers who buy
in bulk very standard software as a service business model.
A further billion dollars, or twenty seven percent of open
AI's revenue comes from open ai licensing its models and
(24:43):
services via its API. One thing you notice when you
look at this price page is that there's this huge
variety of models and APIs available, and that there's massive
amounts of variation in pricing too. I look into these
in depth in my newsletter, which by the way, is
called open ai is a bad business and it's worth
reading if you want to know, if not, just for
the fact that the pricing in its various services is
(25:04):
also super messy, but you don't need to know that
right now. Open Ai also makes about two hundred million
a year selling access to its models through its Microsoft
according to Bloomberg, where open Ai takes a twenty percent
cut of all revenue, and by the way, that's not
on top of the billion. This means that open ai
only makes about eight hundred million dollars a year by
selling access to its API. That two hundred million is
(25:26):
on top I want to add, how worrying this is
both open Ai and the larger generative AI market. If
open Ai, the most prominent name in all of generative ai,
is only making a billion dollars a year from selling
the shovels for the gold rush, what does that say
about the growth trajectory for this company all the actual
(25:47):
usage of generative AI products. That's the question for later
in the show. The first let's talk dollars. So as
it stands, open Ai makes the majority, like I said,
more than seventy percent of its revenue from selling access
to premium versions of chat GPT. A few weeks ago,
the Information Report that chat gpt plus had more than
ten million paying subscribers, and that it had one million
(26:08):
more than we're paying for higher priced plans for business teams.
As I've laid out previously, this means that open ai
is making about two hundred dollars million dollars a month
from consumer subscribers. But business teams is a very vague.
It isn't obvious what it means. It could mean the
team plan, it could mean the enterprise plan, and you
(26:29):
may think, well ed, couldn't you be horribly wrong? Couldn't
it be that they're actually ripping' they they got a
million enterprise subscribers. Wouldn't that be good? Wrong? Oh? The
New York Times has my back here. Based on their reporting,
we can actually get a little more specific. Chat gpt
plus is ten million customers, making open Ai around two
point four billion dollars a year. As ten million users
(26:51):
spending about twenty bucks a month, which equates two hundred million,
might apply by twelve two point four billion dollars. Dad
I swear I was listening in Mathson Times. This means,
by the way, that business users make up about three
hundred million dollars a year in revenue. Of twenty five
million dollars a month, which isn't really great. While ten
million paying customers might seem like a lot. Chat GPT
(27:13):
is effectively to generative AI what Google is to search.
Ten million people paying for this kind of table stakes,
and the idea that this company can triple that number
in a year is absolutely ridiculous. Open ai has been
covered by effectively every media outlet in the world. It's
mentioned in almost every single conversation about AI, even when
it's not about generative AAI, and has the backing and
(27:33):
marketing push of Microsoft and pretty much the entirety of
Silicon Valley. Chat GPT has over two hundred million weekly users,
and The New York Times reports that open ai has
and I quote, three hundred and fifty million people using
their services each month as of joom though it's unclear
if that includes those using the API. I would guess
that's chat GPT users collectively. This means that open ai,
(27:56):
the most popular company in the industry, in the most
prevalent industry in the industry, talked about to everyone and everywhere,
can only convert about three percent of its free users
into paying customers. Now, by the way, Eager listeners may go, well,
three twenty percent conversion, that's not bad, right, right, wrong. Generally,
free products don't cost this much to run. Chat GPT
(28:19):
itself is extremely expensive. Whether it's free or whether it's premium,
it's the same thing. So yeah, those three hundred and
fifty million people, they're more of like a parasite than anything.
The ninety seven percent who won't give them the credit card.
It's not great. And I think that lack of conversion
might be because chat GPT and chat GPT Plus are
(28:39):
really similar products. Chat GPT Plus that you use the
product more often and have access to numer models, but
there's no obvious new thing you can do. As a result,
you can't really upsell to plus unless you found someone
who's already worked out the use case for themselves. And
open Ai remains piss poor at actually marketing this product,
mostly because of the limitations of what a large language
(29:01):
model can do. As a result, most people diddling with
chat GBT will get what they need to out the
free version. I'd also argue that those willing to pay
for a Plus subscription are more likely to use the
platform way way more than free users, and because every
prompt on chat, GBT loses the company money. It's reasonable
to believe that paying users would be far more of
a burden on the system. While there's a chance to
(29:23):
open an eye, could have a chunk of users that
aren't particularly active. One cannot run a business based on
selling stuff you hope that people won't use. And no, no,
it's nothing like a gym or insurance. Stop it. When
I wrote the newsletter and I said that, I got
so many people who thought they were the cleverest people
at all school. Oh, it's like a gym. It's like, sure,
it's not the same. Stop it. You're not clever. No
(29:44):
one's impressed, And I should be clear. There's also a
reason why enterprise customers are generally more desirable than private individuals.
As with any other consumer centric subscription product, regular customers
are far more likely to cut their spending when they
don't feel like they're getting value from the product, or
when the household budget demands it, or there's economic problems.
Just look at Netflix. They were the biggest name while
(30:06):
they still are the biggest name in streaming, and they
lost a million customers in twenty twenty two because of
the cost of living crisis. These are real things that
will happen to open ai if they're not already happening.
Chat GPT plus is likely for many people, kind of
a lifestyle product, and the problem is that when people
lose their jobs, or inflation hikes happen, or cheaper things
(30:27):
come along that do much the same thing, these products
are the first to get slashed from the budget. And
that's before any arbitrary or silly or desperate price increase
is done by a company that isn't the good business. Honestly,
it's kind of remarkable that open ai found ten million
people to actually pay for chat gpt, But how do
you grow that to twenty million or forty million people.
(30:48):
I think they need like thirty million by the end
of twenty twenty five. How does that happen? Very worrying anyway.
At present, open ai makes about two hundred and twenty
five million dollars a month. That's two point seven billion
a year. By selling these premium subscriptions. To hit eleven
point six billion dollars in twenty twenty five, open ai
would have to increase revenue from chat GPT customers by
(31:09):
three hundred and ten percent. If we consider and forgive me.
I'm going to do some maths at you. The current
ratio of plus subscriptions, the teams and enterprise subscriptions that
eighty eight point eight nine percent is GPT plus versus
the business ones at eleven point eleven percent. Open Ai
would need to find eighteen point two nine million paying users,
and that's at the new price point of twenty two
(31:29):
bucks a month, while also retaining every single one of
the current subscribers to Chat GPT plus, who would also
need to renew at the same price point to hit
seven point four billion dollars or about six hundred and
sixteen million dollars a month. It would also have to
make an additional nine hundred and thirty three million dollars
in revenue from its business or enterprice clients, which again
(31:50):
would require open Ai to more than triple their users
and retain the current ones to triple these users. To
actually do this, Chat GPT has to meaningfully change, It
has to do so soon, or disclose multiple meaningful, powerful
use cases that are so impressive that eighteen million new
people agreed to give them twenty two dollars a month.
That's an incredible, and I would say insane goal and
(32:13):
one that I do not think this company is capable
of achieving. It would require making chat GPT something far
more compelling than it already is, in ways that I
can't even imagine. Just as the company lost multiple members
of its senior leadership team, it doesn't look good. There
needs to be a meaningful change. Now. I know I've
(32:35):
already given you a lot of worrying things about open
aiye about their very top heavy, subscription driven business, Why
it isn't going to grow, why they need to grow
it to grow faster than any one's ever grown, why
they need to stop losing so much money. But there's
something else to be worried about. There's something even flimsier
about this business, and it's actually something that's disastrous. It's disastrous.
(32:58):
I'm serious. It is simply disastrous. How little of open
AI's revenue comes from providing other companies that means to
integrate generative AI into their systems. It is astonishingly bad.
I cannot be clear enough here. Before I wrote this script,
before I wrote the newsletter that led to it, I believed,
in my heart of hearts that how OpenAI made their
(33:20):
money was from providing access to their models. That makes sense, right,
because this company's talked about everywhere. Their technology is meant
to be innovative. Everyone's meant to use it, right, that
makes sense. Surely most of their money would come from
people getting in on the revolution and then integrating the
revolution into their products, and then people using the revolutionary thing. Right. Wrong.
(33:44):
First of all, if open ai only makes a billion
dollars a year selling API access and thus let you
integrate the models into your products, it suggests that even
the biggest company in generative ai cannot find enough customers
to make its business viable. Secondly, it's yes, there's a
remarkably small amount of demand for GENERATIVEAI integrations. Or consider
(34:04):
another way that the companies connecting to open ai aren't
really making it very much money at all. If the
GENERATIVEAI revolution were really here, surely open ai, the household
name in large language models, would be rolling in cash
from their ABI business, not leaving it at less than
thirty percent of their and your revenue. This isn't the case.
(34:25):
It's so strange. So at the twenty twenty four open
Ai Dev Day event, it's developer conference, which took place
in October, first, the company said that over three million
developers are building apps using open AI's infrastructure. That works
out to about three hundred and thirty three dollars a developer,
and that's far less money than other companies which make
money by providing a service through their API actually make. Twilio,
(34:46):
a company that makes its money sending SMS, messages and
push notifications for companies, over the past quarter, made about
a billion dollars in revenue. That's what OpenAI made from
renting out its models and APIs over the past year.
Twilio also made roughly four billion dollars over the last
four quarters, which is more than open AI's projected revenue
for the entirety of twenty twenty four. As I mentioned before,
(35:10):
open ai makes two hundred million dollars of its one
billion dollar revenue from Microsoft reselling its models a twenty
percent cut. That suggests that Microsoft two is making about
a billion dollars from open aised models. So in the
event that open ai and Microsoft are making about a
billion dollars in annualized revenue by providing access to open
AI's models. Again, these are estimates based on current growth trajectories.
(35:33):
It suggests that there's only two billion dollars in annual
revenue coming from both of these companies combined, and that's
without wait is open ai making less money from open
AI's models than Microsoft is. Oh my god, this business
is a stinker. And this also suggests that generative AI
(35:54):
as a technology doesn't have a product market fit. According
to a survey by Andrews and Hory It's a VC
earlier in the year and I quote, the twenty twenty
three market share of closed source models was estimated at
eighty to ninety percent, with the majority of that share
going to open ai and open source ones would include
things like metas Lama model, which is kind of open source.
(36:15):
That's a whole other thing anyway. Another survey from IoT
Analytics published late last year suggested that the number might
look a little different, with thirty nine percent of the
market share going to open Ai and thirty percent going
to Microsoft. Assuming that the latter numbers are true or
even close to true, this suggests that the generative AI
market is really small. If open Ai, which dominates with
(36:38):
I'd wager about thirty percent of the market, is only
making a billion dollars a year from selling access to
its models at this stage. In this massive hype bubble.
There might not even be ten billion dollars of annual
revenue from companies integrating generative AI into their products. That's tiny,
and this should be where all the money is. If
this stuff is the future of everything, why is the
(36:59):
revenue stream so painfully weak. Open ai is making so
little selling access to their models, and it suggests that
despite this hype cycle, there either is an interest in
integrating these products from developers, or when these integrations are
actually in there, consumers aren't really into them or using them. Remember,
these products are charged on usage, and so it's possible
(37:19):
for generative AI to be integrated into a service but
not actually drive much revenue for open Ai as a
result of users not really caring. While chat gpt as
brand recognition, companies integrating open AI's models into their products
are far more indicative of the long term health of
both the company and the industry itself. Because if open
ai can't convince people to integrate and use this shit,
do you think others are succeeding? I mean, think about it.
(37:43):
Where have you seen some weird chatbot up here in
your life? Where have you seen an LM poked into
something you use? Have you been like, oh good, I
can't wait, or have you just kind of tried to
ignore it. Maybe you've interfaced with it. It sucks. There
are some decent products, like there are email clients that
summarize emails Microsoft teams Apparently people really like the summarization.
(38:03):
But we are meant to be describing the future here.
We're meant to be describing something exciting and sexy. Not huh.
This summarized the meeting for me interesting. But looking at
these numbers, it's hard to imagine how open ai will
more than triple revenue in the next fifteen months to
hit eleven point six billion dollars in sales. Furthermore, at
its current burn ray, open ai is currently spending, like
(38:26):
I said, two dollars and thirty five cents to make
a buck, meaning that eleven point six billion dollars in
revenue could cost as much as twenty seven billion dollars
to actually make. And as I previously mentioned, what costs
could foreseeably come down, All signs point to them increasing.
GPT four costs one hundred million dollars to train, and
more complex future models will cost hundreds of millions or
(38:47):
billions to train, as I mentioned the Information said that
training costs were would look like three billion dollars in
twenty twenty four, and I think it's fair to assume
that the new models are just as costly, if not
more so. Ye there's something deliciously ironic about all of
this that, despite a clear lack of user interest in
generative AI, open Ahi's global marketing push has succeeded in
(39:08):
making lots of people intrigued enough to try a completely
free service that only loses the money. While it's usually
great news that a product has three hundred or more
million free users, every book time somebody uses a service
like chet GPT, it loses the company money. And in
the case of chet GPT, good lord, they must be
losing so much. The Information estimated in July the open
(39:30):
Ai will spend around four billion dollars in server costs
in twenty twenty four to run chat GPT and host
other companies running their services using GPT in its other models,
effectively meaning that every dollar of revenue is immediately eaten
by the costs of acquiring it, and that's before your
factor in more than fifteen hundred people, as many as
seventeen hundred people now work at open Ai, which is
(39:50):
another one point five billion dollars or more in costs,
and other costs, by the way, are on top of that,
they've got real estate taxes, stock grunts. This is all
very bad and while open ay could potentially reduce costs,
they've shown no proof that they can. And they tried once,
well at least one got out, the so called iraqis
(40:12):
model from last year that they tried to show that
was meant to be more efficient if failed to launch.
That's not a good sign, is it. But there's one
more problem. There's still more problems, and this one, well,
this one was caused by their fundraising. You see, they
just raised six point five six point six billion dollars
in capital one hundred and fifty seven billion dollar valuation.
(40:33):
This means that all future rounds have to be at
that valuation or higher. A lower valuation, which is called
the down round, would make current investors quite pissy and
send a very loud signal to the market that the
company is having trouble because nobody thinks they're worth what
they used to be, which in turn would overwhelmingly suggest
open AI's only way to survive is to raise its
next valuation at a two hundred billion dollar valuation and
(40:56):
require yet another giant raise. I would say at least
ten billion. For context, the biggest IPO valuation in US
corporate history was Alababa, which debuted on the NASC with
a market cap of nearly one hundred and seventy billion dollars.
That figure is more than double the runner up Facebook,
which had a value of eighty one billion dollars. And
I want to give you some history on that one too.
(41:19):
Facebook's initial OPO was really bad because people were concerned
about mobile users and not monetizing them. Well, they won
that one, but the market's pulverized Facebook for that. Do
you think that they're going to be Oh yeah, so
your company loses five or more billion dollars a year
and you have no path of the profitability. Yes, put
you up on a Nasdaq sounds perfect. Fuck that. No,
(41:40):
they're not going to do that. How is this company
going to do? How is this company going to IPO?
And yet there's more problems too. I don't know how
openay is meant to convert itself from its weird nonprofit
structure into a for profit company. I don't know if
it's possible, it won't be easy. And what's crazier is
and there's so many little things like this in this
(42:00):
story where you're like, people are going to look back
on this and think, Wow, we were goddamn stupid. At
least I hope they did. But open ai has two
years from close to convert from a nonprofit to a
for profit or their funding will convert into debt. I
think it's six to nine percent interest. Hey, this isn't good. Also,
at some point open ai is going to have to
(42:21):
work out a way to go public, like I said,
because otherwise, why did people invest? Why did people bother?
They could do future secondary sales, And at some point,
if that's all they're going to be able to do
in secondary sales referring to selling private stock to another individual, well,
I mean at that point it's a Ponzi scheme. You're
just pulling in money to hand out money to other people. Eventually,
(42:45):
it's not good these This company has no past to profitability.
It doesn't have one. And regardless of what happens with
everything I'm mentioning, they still need to raise more funds.
Chat GPT's free version is an actual poison on open
AI's system. It's a marketing channel that burns billions of
dollars to introduce people to a product that only ten
million people will actually pay for. And open AI's future
(43:07):
depends largely on its ability to continue convincing people to
use it. So how does this continue? How does open
ai survive? I don't think it can. Open ai is
a disaster in the making, and behind it, it's a nastier,
shittier disaster, a lack of fundamental strength in the generative
AI market writ Large. If open ai can only make
(43:28):
a billion dollars as the leader in this market, and
really only eight hundred million because the other two hundred
million comes from Microsoft, it suggests that there's not really
developer or user interest in generative AI products writ Large.
Perhaps it's the hallucination problem, where it authoritatively states something
that isn't true. Well, maybe it's just that generative AI
(43:48):
isn't something that produces interesting interactions with a user. While
you could argue that somebody could work out a really
cool product, it's time to ask why Amazon, Google, Meta,
open Ai, Apple and Microsoft have failed to make one
one in the last two years? Where is it? Where's
the killer app? And no, it's not early days. Shut up.
(44:12):
More money than anything has ever raised has gone into
these companies. More attention has gone into these companies than
any other hyperscalar movement ever, even the metaverse. Where's the product? Man?
Where is it? And while this bubble can continue coasting
for a little while longer, nothing about the open ai
story looks good. This company's lost. They're bleeding money with
(44:33):
every single interaction with the customer, and they're flogging serve
software that's the best kind of useful and worst actively
harmful to the environment. Unless something significantly changes, like a
massive scientific breakthrough in energy or compute efficiency, I just
don't see how open ai makes it two more years.
And worse still, if my hypothesis about the wider market
(44:55):
is true, there might just not be a viable business
in make and selling large language models. While Meta and
open Ai might be able to claim hundreds of millions
of users on these services, I don't see any evidence
that these people will make them any money, and in fact,
I only see evidence they'll lose it. And if I'm right,
we're watching vcs in companies like Microsoft burn tens of
(45:16):
billions of dollars to power the next generation of products
and nobody really gives a shit about And in the
next episode, I'm going to go into the much bigger problem,
the subprime AI crisis, where everyone building on these platforms
is inevitably going to get rug pulled when Open AI
and Anthropic and other companies raise their prices, because if
they don't raise their prices, how are they going to
(45:38):
be able to afford to keep going. I don't bloody no,
but we'll get into it next episode. Thank you for
listening to Better Offline. The editor and composer of the
Better Offline theme song is Matosowski. You can check out
more of his music and audio projects Matttersolski dot com,
(46:01):
M A T T O s O W s Ki
dot com. You can email me at easy at Better
offline dot com or visit Better Offline dot com to
find more podcast links and of course my newsletter. I
also really recommend you go to chat dot Where's youreaed
dot at to visit the discord, and go to our
slash Better Offline to check out our reddit. Thank you
(46:22):
so much for listening.
Speaker 2 (46:24):
Better Offline is a production of Cool Zone Media. For
more from Cool Zone Media, visit our website cool zonemedia
dot com, or check us out on the iHeartRadio app,
Apple Podcasts, or wherever you get your podcasts.
Host
Ed Zitron
Popular Podcasts
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
The Bobby Bones Show
Listen to 'The Bobby Bones Show' by downloading the daily full replay.