The Enshittifinancial Crisis: Part Two

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:02):
Zone media, I'm better off line and this is ed Zetron. Wait,
that's right, folks. We're here for part two of our

(00:23):
multi part series about the final casualty of the incertification
process the financial markets. You see, we're in a world
where because companies can no longer stimulate growth by just
script mining their individual and business users, they've now resorted
to just lying. And lying is an interesting word because
they will claim we're not lying, we used words to

(00:44):
vaguely avoid telling you the truth, or we were technically
true in what we say. There are various kinds of
lies that don't have to just be I told you
something that was not categorically yes or no force. I mean,
you've spoken to other people, but they will say this
for legal purposes. So fine, it's not technically lying, it's

(01:05):
just bullshit. But I think it's important to give you
some examples and to talk about how analyst in the
media simply allowed these lights to propagate unchecked. Let me
set the scene. Our story begins on September tenth, twenty
twenty five. I was for some reason in Los Angeles
that I can't remember. Maybe a concert, maybe an interview.
Truly don't know. Oracle announced its unfillable, unpayable, three hundred

(01:29):
billion dollar deal with open AI, which they didn't have
the capacity to serve, and open a I didn't have
the cash to pay them, and this led to a
thirty percent or more bump in stock price. Analysts who
should extensibly be able to count, called it momentous and
said they were in shock. On September twenty second, twenty
twenty five, CEO Saffracats stepped down. She stepped down from oraclen.

(01:52):
Nobody seemed to think that was weird or suspicious or worrying.
This was this was meant to be the beginning of
the AI like boom too. I mean, we kind of
really in the AI bool, but this was meant to
be the even more bigger one, right, Why would you
step down? Why would you step down? Why would you
step down? And then have two guys like two different

(02:14):
They have co CEOs. Very sorry to my friend Jim,
but co CEOs don't usually do well. Look at BlackBerry.
But anyway, two months after that, Oracle stock was down
forty percent, with investors worried about Oracle's growing capex, which
is surprising. I suppose if you didn't think about how
Oracle would build the data centers to serve the three
hundred billion dollar deal. A makes me feel insane when

(02:39):
I say these things out how loud, But just to
be clear, anybody who traded on that got burned. Now
another thing happened on September twenty second, twenty twenty five,
in Vidio announced what they called a strategic partnership to
invest up to one hundred billion dollars and build ten
gigawads of data centers with open Ai, with the first

(03:00):
to gigawatt to be deployed in the second half of
twenty twenty six. Just a few questions from me personally,
where would the data centers go, How would open Ai
afford to build them? How would open Ai build a
gigawa in data centers in less than a year? Don't
ask questions. Pig back in your cage now. In video

(03:21):
stock bumped from one hundred and seventy five point three
to one hundred and eighty one dollars in the space
of a day. Said they're very normally moving on and
the media wrote about the story as if the deal
was done, with CNBC claiming that and I quote the
initial ten billion dollar tranch was expected to close within
a month or so once the transaction had been finalized.
I read at least ten stories that said in Video

(03:41):
had invested one hundred billion dollars in open Ai. Analysts
would go on to say that in Video was locking
in open Ai and I quote again to remain the
backbone of the next gen AI infrastructure, and that demand
for in video GPUs is effectively baked into the development
of frontier AI models, and that the deal, the partnership
between the two companies and Validate did in Vidia's long

(04:04):
term growth numbers with so much volume and compute capacity.
Others would say that in Video was enabling open Ai
to meet surging demand. Three analysts, rask Gone at Bernstein Laurier,
DA Davidson, and Wagner at Aaptors Capital all raised circular
deal concerns, but they were in the minority, and those
concerns were still often buried under buoyant optimism around the
prospects of the company, and nobody just fucking saying the words, Hey,

(04:26):
no one could afford any of the Oracle can't afford
to build it, they don't have the capacity, they ust
can't get paid by open Ai. Open Ai doesn't have
the money. But in Vidia, this deal also doesn't make sense.
You know. It feels like open ai is just promising
everyone's stuff they can't afford. All of these things seemed
like very obvious questions, but people just kind of clapped

(04:48):
their hands and went, fuck, yeah, we're doing Ai even
more now. But then there was this other thing about
the Nvidia deal. It's this quincy teeny weeny quincy little
problem one little pro everyone about the Invidia deal. The
deal that everyone said was done. It was a letter
of intent. It said so in the announcement and on
Invidia's November runnings it said that it had entered into

(05:11):
a letter of intent with an opportunity to invest in
open Ai. It turns out the deal didn't exist, and
everybody fell for him. Invidia hasn't sent a diamond, likely won't.
A letter of intent is also known as the concept
of a plan. A little over two weeks later, on
October first, twenty twenty five, Reuters reported that Samsung and

(05:32):
sk Heinex had signed letters of intent to supply memory
chips for open AI's data centers, with South Korea's Presidential
Office saying that said chip demand was expected to reach
nine hundred thousand wafers a month, with and I quote
much of that from Samsung and esk Heinex, which was
quickly extrapolated to meet around forty percent of global dram

(05:53):
outpum stocks in both companies to quote Reuter's Sword, with
Samsung climbing four percent and sk Heinex more than twelve
percent to an all time high. Analyst Jeff Kim KB
Securities said that and I quote, there have been worries
about high bandwidth memory prices falling next year on intensifying competition,
but such worries will now be easily resolved by the

(06:14):
strategic partnership, adding that since Stargate is a key project
led by President Trump, wrong lie, there also is a
possibility that partnership will have a positive impact on South
Korea's trade negotiations with the US, which is just to
be clear, total and utter bollocks. It's wank, it's scrap,
it's nonsense, boop them above all the technical terms. Donald

(06:37):
Trump is not leading Stargate. Stargate is a name used
to refer to data centers built by open Ai, by
which I mean other people building them for open Ai.
KB Securities is around forty three billion dollars of assets
under management. This is the level of analysis you get
from these analysts. This is how much they know. There
are people that study JoJo's bizarre adventure that have more

(07:00):
economic analysis than any of these fucking people. Most Star
Wars fans who have just memorized all the names of
the glop shittos of the world have better analysis than this.
But nevertheless, you'd think, right that these companies have made
all this noise, and they've talked about this stuff, the
Presidential Office all involved, you think we'd be able to
get a whiff of that on their earning scores, right well.

(07:23):
On I sk Hihenex's October twenty ninth, twenty twenty five
earning score. Weeks after the announcement, it's CEO Kim Wu
Yun was asked a question about high bandwidth memory growth
by Eskkim from Daiwa Securities and I quote, so this
is this is SKKM asking the question. Thank you very
much for taking my question. It is on demand now.

(07:43):
There have been a series of announcements of GPU and
a six supply cooperation between big techs and AI companies,
fueling expectations of further AI market growth. Then, against this
back drop, what is the company's outlook on high bend
with memory demand as well as a broadening of the
customer base, es sk Heinex. Thank you for the question. Now,
with upward adjustment in big text CAPEX and increasing investment

(08:06):
by AI companies, the high bandwidth memory market, even by
a conservative vestmer, will keep growing at an average of
over thirty percent for the next five years. I will
point to our recent letter of intent with open Ai
for large scale DRAM supply as an example of the
very strong demand for AI as well as the need
to secure AI memory based on HBM more than anything

(08:26):
else went developing AI technology. That is it. That is
the only mention of open AI. That's it. No other
mentions otherwise, sk Heinex has not added any guidance that
would suggest that it's DRAM sales will spike beyond overall
growth where it's already grown because everyone's running out of
ramthanks to fucking AI. Other than mentioning that it had

(08:47):
an I quote completed year twenty twenty six supply discussions
with key customers, there is no mention of open AI
in any earning's presentation. Now You may think there were
two companies. You said two company names. There were Samsung
as well. Right well. On Samsung's October thirtieth, twenty twenty
five earning school Samsung mentioned the term d RAM eighteen
times and neither mentioned open AI nor any letters of

(09:08):
intent of any kind. In its Q three twenty twenty
five earnings presentation, Samsung mentions it will a quote prioritize
the expansion of high bandwidth memory four business with differentiate
performance to address increasing AI demand. Great place to mention
open AI. Right, yeah, that would be if the money
existed and they were gonna do anything Now, analysts do

(09:29):
not appear to have noticed the lack of revenue from
an apparent deal for forty percent of the world's RAM.
Oh well, Pobody's perfect right. My frustration, you hear in
my voice is that I'm a guy with a Google doc.
I'm not a financial analyst, but I appeared to be
able to do what they're doing because I'm able to
fucking read a Both Samsung and s k Heinez's stocks

(09:50):
have continued to write since, and you'd be forgiven for
thinking this deal was something to do with it, even
though it was not. But silly season was not yet over, however,
and on October fifth, twenty twenty five, AMD announced that
it had entered a multi year, multi generation agreement with
open Ai to build six goddamn gigawads of data centers,
with the first one gig what deployment said to begin
in the second half of twenty twenty six, calling the

(10:12):
agreement definitive with terms that allowed open ai to buy
up to ten percent of AMD stock, festing over specific
milestones that started with the first giga what of data
center development, said data centers would use AMDs yet to
be released four to fifty GPUs. The deal would per
reuters bring in tens of billions of dollars of revenue. Now,
I know, hate to be an asshole, right, I know,

(10:35):
I'm always on this shit, right Yeah? Where would those
data centers go? How much? How much would they cost?
How would open Ai pay for them? Would the jips
be ready in time? Silence? Well, how dare you ask questions?
How dare you? Why are you asking questions? Number? Go up?
Number go up? Look look at number? Go up. AMD
shares searched by thirty four percent, with the analyst Dan

(10:56):
Ives of Webbush, who appears to dress like the San
Diego Padres Didy connect Jerseys, saying that this was a
major evaluation moment for AMD. As in the side, I've
said that in video would benefit from the meta verse
in twenty twenty one, and told CBS News in November
on November twenty second, twenty twenty one, that and I quote,
the metaverse was real and that Wall Street was looking

(11:17):
for winners, were they dan? Now one more thing, that
AMD's November earnings a month after that announcement might be
a barn burner full of remaining performance obligations from Open AI.
Things that Open appaying them moost tens of billions of
dollars of revenues right. In fact, CEO Lisa Sun who
said that AMD expected this partnership will significantly accelerated Stata

(11:39):
Center AI business and with the potential to generate well
over one hundred billion dollars in revenue over the next
few years in the announcement, and here's how AMD's ten
Q filing referred to it, and I quote. As of
September twenty seventh, twenty twenty five, the Aggagut transaction price
allocated to remaining performance applications on the contracts with an
original expected duration of more than one year. It was

(12:00):
two hundred and seventy nine million dollars, of which one
hundred and thirty nine million dollars is expected to be
recognized in the next twelve months. The revenue allocated to
remaining performance applications does not include amounts whichever original expected
duration of one year or less. Now, I don't know
about you. I did not I didn't do economics in school,
didn't do anythin like that, didn't do well at maths

(12:21):
in school. But two hundred and seventy nine million dollars
seems lower than over one hundred billion dollars. So I
guess just just no revenue from open AI, no revenue
of any kind. I guess. Now, AMD did raise guidance
by thirty five percent of the next five years, and

(12:42):
they're trailing. Twelve month revenue is thirty two billion dollars.
Tens of billions of dollars would surely lead to more
than a thirty five percent boost, because that would be
like eleven billion. That's just eleven billion dollars. That's a
lot less. That's a lot less than one hundred billion dollars.
I don't know guess all that was for nothing. No

(13:02):
follow up from the media, no questions from analysts, just
the shrug and we all move on. They were at ces.
Greg Brockman came on AMD's Lisa Sue said something about
Yachta flops, which really just pissed me off to think about.
Still nothing anyway. AMD stock is now down from a
high of two hundred and fifty nine bucks at the
end of October to around let's see md stocks understand

(13:25):
about two hundred and thirty one bucks right now. Everybody
who traded in based on an analyst and media comments
got fucked. Now let's go to my favor them, and
that's a little company named Broadcom. So back on September fifth,

(13:48):
Broadcom send and it's earning school that it had a
ten billion dollar order from a mysterious customer, which analysts
quickly assumed was open Ai, leading to Broadcom stock probably
nine percent and gradually increasing to a high of a
three hundred and sixty nine dollars on September tenth, before
declining a little until October thirteenth, when Broadcom announced a
ridiculous ten gigawa deal with open Ai, claiming that it

(14:09):
would deploy ten gigawats of open Ai designed chips, with
the first ranks to deploy the second half of twenty
twenty six and the entire deployment completed by the end
of twenty twenty ninth. So three fucking years for ten
fucking gigawatts. Just to give you an idea, it's about
two and a half years a gigawat right now. And
that's for open Ai slush fun projects, the ones that
have all the money and none of the oversight. Now.

(14:29):
The same day, President of Semiconductor Solutions Charlie Cowas added
that said mystery customers actually not open Ai, and I quote,
I would love to take a ten billion dollar purchase
order from my good friend Greg Brockman, CEO of open Ai,
Cowas said he has not given me that purchase order yet. Nevertheless,
Broadcom stock popped nine percent on the news about the

(14:50):
ten gigawa deal with CMBC, adding that the companies have
been working together for eighteen months because it's open Ai.
Nobody sat and thought about whether somebody at Broadcom was saying, wow,
well open has yet to ward these chips, yeah, and
whether that was a problem. In fact, the answer to
how does open Ai thought this appeared to be they'd
afford it when it came to analysts. Now I'm gonna

(15:11):
I'm about to read you a quote that is so
funny when you realize that this is someone who makes
like a shit ton of money. Ahem. The twenty twenty
six timelines set out by open Ai for the buildout
is aggressive, but the startup is also best position to
raise funds required for the project given the heights of
investor confidence, said gad Joe Sevilla, an analyst eMarketer. Financing

(15:33):
such a large chip deal will likely require a combination
of funding grounds pre orders, strategic investments and support from Microsoft,
as well as leveraging future revenue streams and potential credit facilities. Hogwash,
just complete, complete bollocks. Insane that people pay these analysts bullshit. Okay,

(15:53):
So just let's just break this down financing such a
large chip deal. So ten gigawats, you're going to be
looking at fifty bill dollars a gig was you look
at that half a trillion dollars. So to do this
for this aggressive buildout, which was ten giga Watson three years,
just not possible. They would fund this by pre orders,
pre orders for what exactly, because pre orders would be

(16:17):
then pre ordering something from Broadcom. Not really sure how
that fixed the problem. Strategic investments from whom just funding
I guess, support from Microsoft funding again, I guess, And
I don't know why Microsoft be paying for that, as
well as leveraging future revenue streams and potential credit for
say what what does that mean? Saving money doing like

(16:42):
a revenue trade deal? Like what are you talking about? Anyway,
you don't need to worry because open ai solution to
this was far simpler. It didn't order anything. During Broadcom's
November earning school, where Broadcom revealed that the ten billion
dollar order was actually from Anthropic, another l of them
started that burnst billions of dollars, which was actually buying
Google's TPUs from Broadcom, and he also added the Anthropic

(17:05):
had made a second eleven billion dollar deal. Alice somehow
believed that Anthropic is positioned to spend heavily despite being
yet another venture backed welfare recipient in the same flavor
as open Ai. Oh right, right, sorry, okay, Broadcom. You
may be wondering about the ten gig what deal though,
because those earnings right well, Broadcom CEO hoc Tan said

(17:26):
that he did not expect much in twenty twenty six
from the deal, and guidance did not change to reflect it. Now,
what was great about this as well? And I linked
to this in the newsletter. I really need you to
go and look up how hoc Tan had the conversation
because someone asked him for more detail and he just
straight up said, like, I don't want to talk about it.
It rocks. I honestly fucking love that. Anyway, Broadcom climbed

(17:48):
to a higher four hundred and twelve dollars a share
leading up to its earnings, and I imagine it did
so based on people trading on the belief that open
Air and Broadcom were doing a deal of some sort together,
which does not appear to be happening. While there is
an alleged seventy three billion dollar backlog for Broadcom, every
dollar from Anthropic is utterly questionable. Now, some of you
might say, ed, we can't just automatically distrust public companies. Actually,

(18:11):
yes we can. I'm doing it right now. Whenever a
company says a litter of intent, as in Video and
sk Heinez and Samsung did, it's important to immediately stop
taking the deal. Seriously until you get to a specific
word contract, not agreement or deal or announcement, but contract,
because contracts are the only thing that actually matters. A

(18:33):
great example of this is in in Video's last earnings,
they said they would be investing up to ten billion
dollars in Anthropic. Right, you'll notice that that was in
the middle of November. It's now January and we haven't
heard boo about it. It's probably because nothing's fucking happening.
And also that phrase up to as well means that
it will be However much in Video wants to invest,

(18:55):
if at all. They only just did their Intel investment. Anyway,
whenever they say these terms, you need to just get suspicious.
It's time. It's time to stop trusting them, because they've
played enough hide the sausage for one EPOC. I think
it's time for everybody, analysts, the media, members of Congress,
the fucking pope. I don't care, to start treating these

(19:15):
companies with actual suspicion and to start demanding timelines like
I said. In Video and Microsoft announced that fifteen billion
dollar investment, and in Anthropic, where's the money? Why does
the agreement say up to ten billion for in Video
and up to five billion for Microsoft. Where's the money?
Where is it? We haven't heard shit about it since.

(19:37):
What we have heard is that Anthropic is raising either
ten billion dollars or twenty five billion dollars at a
three hundred and fifty billion dollar valuation. Where'd the money go?
Where the money go from? Those fuck nuts? But they go, Sorry,
Jensen won for calling you a fuck n All right,
you're acting like one though, mate. Anyway, these deals are
announced with the intention of suggesting there is more revenue

(19:59):
and money and generative AI that actually exists. Furthermore, it
is irresponsible and actively harmful for analysts in the media
to continually act as if these deals will actually get
paid when you consider the financial conditions of these companies.
As part of its alleged funding announcement within Video and
Microsoft and Thropic also agreed to purchase thirty billion dollars

(20:20):
as your Compute, which is, for those of you who
don't know, the compute cloud platform, the competitors to Amazon
Web Services and Google Cloud that Microsoft runs. It also
agreed to Anthropic agreed to spend tens of billions of
dollars with Google Cloud. They ordered ten billion dollars of
chips from broad Coom earlier in twenty twenty five, and
apparently another eleven billion dollars of them in the last

(20:41):
fiscal quarter. How does Anthropic pay for them? How Anthropic
allegedly burned two point eight billion dollars last year twenty
twenty five, though I think they burn much more and
raise sixteen point five billion dollars in funding before Microsoft
and then Videa's imaginary money. How are investors tolerating Broadcom
not directly stating the future financial condition of this company

(21:03):
is questionable? Has Broadcom created a reserve for this deal?
If not, why not? Anthropic will make no more than
five billion dollars in twenty twenty five, I'm sure of it,
and raised about seventeen point five billion dollars with another
two point five billion dollars coming in debt. How can
it foreseeably afford to pay ten billion, eleven billion or
twenty one billion dollars considering it's already massive losses and

(21:24):
all those other obligations I mentioned. Will Jensen one hand
over ten billion dollars so that Anthropic can hand it
straight to Broadcom? I don't think so. I realize the
counter argument is that the companies aren't responsible for the
counterparty's financial health. But my argument is that it's the
responsibility of any public company to give a realistic view
of its financial health and revenue sources, which includes not

(21:44):
to give a chunk of its revenues from a startup
that can't afford to pay for its orders. There's no
counter for that. Anthropic cannot afford to pay broad Com
ten billion dollars right now or eleven billion dollars in
a few months. Where is the money coming from. They're
allegedly raising twenty five billion dollars. Great that ac shud
lead them with four billion dollars to pay for what
like thirty forty fifty billion dollars of compute costs. I

(22:09):
don't know. Again, I ain't no mathematician, but I can
put those numbers together and I can say that's a
pretty big minus. Again, I'm not even trying to be
sarcastic here. The things I'm saying are actually very rational.
They're very reasonable. Anthropic even if they raised twenty five
billion dollars, cannot pay even like a quarter of the

(22:30):
things they've agreed to their existent revenue sources. Even if
they made twenty billion dollars in twenty twenty six, which
they will not. By the way, they are absolutely I
just don't believe them about their revenues at all or
any of the estimates. But the problem is that in
any bubble, being really stupid and ignorant works right up

(22:51):
until it doesn't. And however harsh the dot com bubble
might have been, it wasn't harsh enough, and those who
were responsible were left unpunished and unashamed, and that guaranteed
that the cycle would happen again, and then it would
be worse. I want to be really abundantly clear about
what's happening. Every single stock you see growing because of AI,

(23:15):
outside of those selling RAM and GPUs, is actually growing
because of something else. Microsoft, Amazon, Google, and Meta all
have other products that are making the money and are
still growing. AI is not growing them. And because analysts
and investors in the media do not think about things
for two seconds, they have allowed themselves to be beaten

(23:35):
down and turned into supplicants for future public stock growth.
Investors have allowed themselves to be played, and the results
will be worse than the dot com bubble bursting by
several echelons and perhaps the biggest victims will be venture capital,
which is existentially threatened by the AI bubble, But the
actual victims will be regular people who made the mistake

(23:56):
of trusting analysts or anyone who told them to invest
in the markets. But I'm going to be really simplistic
for a second. I am skeptical of AI because everybody
loses money. I believe every AI company is unprofitable with
margins that are getting increasingly worse as they scale, and
as a result, none of them will be able to

(24:17):
get acquired or go public. This means that venture capitalists
that have sunk money into AI stocks and private AI stocks,
I should be clear, are going to be sitting on
a bunch of assets under management AEM the same assets
they collect fees on because that's how venture capitalists make
money that will eventually creter or go to zero because
there will be no way for any liquidity events so
taken in public or selling someone to occur. This is

(24:41):
at a time of historically low liquidity for venture capitalists,
with Pitchbook estimating that there would be only one hundred
point eight billion dollars in venture capital funds available at
the end of last year. Venture capitalists raise money from
limited partners who invest in venture capital with the hope
of returns that outpace investing in the public markets. Venture
capital vastly overinvested in twenty twenty one and twenty twenty

(25:04):
two as well, and this was also a problem with
private equity. In simple terms, this means that these funds
are sitting on tons of stock that they cannot shift,
and the longer it takes for a company to w
either go public or required, the more likely it is
the VC or the P firm will have to mark
down its value to something more realistic. This is so
bad that, according to Carter, as of August twenty twenty four,

(25:25):
and that CARTA some of you can't understand how I talk.
As of August twenty twenty four, less than ten percent
of VC funds raised in twenty twenty one have made
any distributions to their investors, that's handing any kind of
cash out. In a piece from September twenty twenty five,
Carter also revealed that about fifteen percent of funds from
twenty twenty three have generated any disbursements as of Q two,

(25:49):
twenty twenty five, and the media net internal rate of
return was a median zero point one percent, meaning that
at best most investors got their money back and absolutely
nothing else. And just to be clear, you don't invest
so you can get an exact amount. You don't give
someone a dollar, wait four years and then get a
dollar back. That's not investing, that's just putting money in

(26:11):
a box. And in fact, investing in venture capital has
kind of fucking sugged for a while. According to Carter,
as of the end of Q two, most VC funds
across all recent vintages that's after twenty eighteen had a
TVPI somewhere between zero point eight x and two x
TVPI I'll get to in a second, but there are
some areas where standout tvpis are surfacing. TVPI means total

(26:33):
value to paid in capital, or the amount of money
you made for each dollar invested. So if you invested
a dollar point eight x would mean you got eighty
cents back, and if you invested a dollar you get
two bucks on two X. I hope that makes sense. Now,
there's a chart in the newsletter version of this episode,
which I encourage you to check out, but it tells
you that for the most part, vcs have struggled to
provide even money returns. Since twenty seventeen, decent TVPI is

(26:57):
two point five x, and as you'll see from the chart,
things basically collapsed since twenty twenty one. Companies are not
going public or being acquired at the same rate, meaning
that investor capital is increasingly locked up, meaning that limited
partners are still wing for a payoff from the last bubbles.
Let learn this one now. Carter would update the chart

(27:18):
in December twenty twenty five, and things would somehow get worse.
TVPI soured further, suggesting a further lack of exits across
the board. The only slight improvement was the median IRR,
which rose from zero point five percent from funds from
twenty twenty one and zero point one percent for funds
from twenty twenty two. Those are still real, fucking terrible.

(27:40):
Those are absolutely astonishing. I mean, a good IRR is
ideally like fifteen to twenty percent, not zero point five percent.
I don't know, but in simpler terms, we are looking
at years of locked up capital, even venture capital cash starved,
and a little ESPRA. The worst part is that all

(28:16):
of this is happening during a generational increase in the
amounts that startups need to raise thanks to the ruinous
costs of generative AI and the negative margins of AI
powered services. Now, I'm going to quote myself from a
December fifth newsletter called the Ways the AI Bubble Might Burst. Cursor,
Anthropic's largest customer and now it's biggest competitor in the
AAI coding sphere, raised two point three billion dollars in

(28:37):
November after raising nine hundred million dollars in June. Now
for twenty twenty five, and I should add these are
on revenues of eighty three million dollars in the month
of I think November October, because they said annualized revenues
of one billion dollars. Is dogship perplexity one of the
most And I say this presitantly popular AI companies raised

(28:57):
two hundred million dollars in September after using one hundred
million dollars in July, after seeming to fail to raise
five hundred million dollars in May. After raising five hundred
million dollars in December twenty twenty four for a fucking
search engine, Cognition raised four hundred million dollars in September
after raising three hundred million dollars In March, Coher raised
one hundred million dollars in said temper, a month after

(29:18):
it raised five hundred million dollars. That's a lot of money,
and that's a lot of money in a very small
amount of time, and that's taking money away from everyone
else because they need that money. You need the money
to invest in the stops. But none of these companies
are profitable, nor do they have a path to end
acquisition or IBO. And why do you say, why why

(29:40):
do you ask? Ede? You're being so horrible? What's because
even the most advanced AI software companies ultimately prompting open
AI or Anthropics models, meaning that their only real intellectual
property is those prompts and their staff and whatever they
can build around the models they don't control, which has
been obvious from the meager acquisitions we've seen so far.
Another fun fact, Anthropic ended up cutting off people that

(30:03):
worked to Xai that were using Cursor from using their models. So,
just to be clear, Cursor completely different company Anthropic. If
you work at a company at Anthropic doesn't like and
I know we don't, we don't like elon mask, I
get that. But if Anthropic can arbitrarily cut off their
models from somebody's completely independent product. Just because they're deciding

(30:24):
to cut off a competitor, they're just going to broaden
the term competitor to mean anyone they don't like, anyone
that gets in the way. Now, Windsurf, which was allegedly
being sold to open Ai last year, ended up selling
its assets to Cognition in July twenty twenty five, with
Google paying two point four billion dollars just to hire
its co founders. And this nebulous licensing agreement similar to

(30:46):
the thing it did with Character to AI, where it
paid two point seven billion dollars to reheigh Nome Shazir,
who he was one of the authors of the Attention
is All You Need paper that started this spars licenses
tech can pay off the stock of its remaining staff.
This is also exactly what Microsoft did with Inflection AI
and it's co founder, mister Suliman, who, and I quote
someone who messaged me really recently, is a huge fucking asshole. Now,

(31:11):
mister Sulliman, if you hear this, I'm just quoting someone
that works for you. If they're saying that about you,
perhaps it's worth asking why are people calling me a
big fucking asshole. I don't know. Must Far come on
the show and ask me yourself. Now open Aiyes, acquisitions
of Statsic one point one billion dollars, Ioproducts six point
five billion dollars, in Neptune four hundred million dollars. We're
all in stock. Every other major acquisition of this era, Wiz, Confluence, Confluent,

(31:35):
even Informatica and so on, is either someone trying to
pretend that something is related to AI, like Wiz, or
trying to say that a data streaming platform is heavily
AI related because AI needs data, which which may be true,
but it doesn't mean that AI companies are selling and
they're not, by the way, which is a problem. As

(31:56):
forty one percent of US venture dollars in twenty twenty
five went to AI as of August, and according to Axios,
the global number was around fifty one percent. A crisis
is brewing. NERD Lawyer back in October wrote about the
explosive growth of secondary markets, and I quote enter the
secondary market a once niche corner of venture capital that
has transferred into a primary liquidity mechanism. What's remarkable is

(32:19):
how quickly this market is matured at least five major
venture funds have hired full time staff dedicated to manufacturing
non traditional exits. As Hans Swilden's CEO of Industry Ventures explained,
all the brand named funds are all staffing and thinking
through liquidity structures and professional buyers have flooded in. Megafunds
specializing in secondaries have raised unprecedented amounts. Lexington raised a

(32:42):
record twenty three billion dollars, or harbor Vest, Ardient and
Collar Capital have raised funds in the ten to twenty
billion dollar range. In simpler terms, there are now hot
potato funds where either another limited partner buys and other's allocation,
the companies themselves by back their stock, or the stock
is resold to other private investors who are building venture

(33:03):
capital firms out of shit. The venture capitalists can't sell.
You're seeing where the problem might be. Eventually, someone's going
to be. It's not Bomberman. It's not good now from
the same NERD lawyer piece and I quote again, and
they're not alone. The secondary market is projected to handle
one hundred and twenty two billion dollars in assets in

(33:24):
twenty twenty five, Yet that still represents just one point
nine percent of the total Unicorn value. As in the
side of Unicorn is a company worth a billion dollars,
there's six plus trillion dollars in untapped liquidity potential as
aside here. That's a really nice way of looking at this.
The transformation of the secondary market from emergency tool to

(33:44):
standard operating procedure represents the most significant structural shift in
venture capital since the rise of unicorns. It's not a
temporary fix. It's a permanent evolution driven by misaligned timeframes
between fund life cycles ten years and company maturation eleven
plus years. I read that it's not just a temporary
fix thing. I'm like, is this ai anyway back to

(34:05):
quoting for better or worse? This is the new reality
of startup funding. Vizs can no longer afford to simply
spray and pray and wait for exits. They need active
liquidity management strategies, and that fundamentally changes what kind of
companies are getting funded and how. I'm just going to
be honest. That last parts bullshit. I wrote this in
the newsletter. That last part is whang. This is not

(34:26):
changing anything about what companies get funded or how they're
still funding these unprofitable monstrosities. Nothing has changed, and I
would argue this piece frames that as a positive when
the reality is far grimmer. Venture capitalists are sitting on
piles of immovable equity and companies worth far less than
they invested in. And the answer, it appears, is to
find somebody else who is able to buy the dead weight.

(34:48):
I assume because they are stupid, Like I mean, I
realize someone is going to say, well, actually, it means
maybe they see maybe they can afford to hold it
for longer. No. Now, if you're buying anything since twenty
twenty one, you're probably get swindled. And according to Newcomer Only,
oney one hundred and seventeen venture funds closed in twenty

(35:08):
twenty five, down from twenty one hundred and twenty twenty four,
and forty three percent of dollars raised went to the
largest venture funds, per The New York Times and Pitchbook,
suggesting limited partners are becoming less interested in pumping cash
into the system at a time when AI startups are
demanding more capital than has ever been raised. How long
can the venture capital industry keep handing out one hundred

(35:29):
million to five hundred million dollars to multiple startups a year,
because all the signs suggest that the current pace of
funding must continue in perpetuity as nobody appears to have
worked out the generative AI is inherently unprofitable, and thus
every single company is on the Silicon Valley welfare system
until somebody or everybody gives up or the system itself
cannot sustain the pressure. I've read too many people make

(35:51):
offhanded comments about this being like the dot com boom,
and saying that lots of startups might die, but what's
left over will be good. And I hate them. I
hate so much. I hate them both for their flippancy
and for their ignorance. None of the current stack of
AI companies can survive on their own, meaning that the
venture capital industry is holding them up. If even one

(36:12):
of these companies faulters and dies, the entire narrative dies
with it. If that happens, it will be harder for
other AI companies to raise and even harder to sell
an AI company to somebody else. And if you can,
you'll be selling it for less. This is a punishment
for a decade plus of hubris, where companies were invested
in without ever considering a path to profitability. Venture capital

(36:35):
has made the same mistake again and again and again,
believing that because Uber or Facebook or Airbnb or any
number of other companies found in nearly twenty years ago
were unprofitable, at some point with past the profitability, I
might add, it was totally okay to keep pumping, pumpaning, pumping,
I'm just saying it, pumping up companies that have had
no path to profitability, which eventually became had no apparent

(36:57):
business models see the metaversal Web three, which eventually have
negative margins so severe, evaluation so high that we will
need an IPO at the a market cap higher than
Netflix at minimum. This is Silicon Valley's rot economy, the
desperate growth at all costs attachment the startups where you
and I quote really liked the founder, where and I

(37:17):
quote again the market could be huge who knows if
it is, and where you just don't need to worry
about profitability because IPOs and exits were easy. Yeah, you see,
venture capital used to be real easy because we were
still in an era of hypergrowth. You could be a
stupid asshole who doesn't know anything, but there were so
many good deals, and the more well known you were
the more likely be brought them first, guaranteeing a bigger payout,

(37:38):
guaranteeing more LP capital, guaranteeing more opportunities that were of
a higher quality. Because you were a big name. It
wasn't about being smart and knowing the fundamentals. It was
about being handed things. It was easier to make a
valuable company too, easier to get funded, and easier to
sell because the goal was always get funded, grow as
large as possible, or well go public. And here's the
thing about that as well. They were just more ideas.

(38:01):
They want more ideas to do. Like I said with
the rock com bubble, they were just there were more
things that people could create. That's the thing that ideas
are a finite source. And when you incentivize this kind
of thinking, you eventually stop incentivizing good ideas. You incentivize
the growth. And as a result, venture capital encouraged growth

(38:23):
at all cost, thinking above everything else. In twenty ten,
ben Horowitz said and I quote that the only thing
worse for an entrepreneur than startup hell, which was bankruptcy,
is startup purgatory. And I quote when you don't go bankrupt,
but you fail to build the Number one product in
the space that's purgatory. You have enough money with your
conservative burn rate to last for many years. You may

(38:44):
even be cash flow positive. However, you have zero chance
to becoming a high growth company. You have zero chance
of being anything but a very small technology business. From
the entrepreneur's point of view, this can be worse than
start up hell, since you're stuck with the small company.
What a fucking, noxious, ugly, horrifying thing to say. Let

(39:05):
me put it like this. If you run a good
business that people like and it's profitable, and you have
good your employees like it, your customers like it, that's
a good company. That isn't hell. And if you think
that way, you're a fucking psychopath. You're an actual nutter.
And this poisonous theory sadly paid off for him in
that startups got used to building high growth, low margin
companies that would easily sell to other companies of the

(39:25):
markets themselves, right up until it didn't. Of course, per
nerd Lawyer, IPOs have collapsed as next sit route along
with easy to raise capital. Per pitch Book, since twenty
twenty two, seventy percent of VC backed exits were valued
at less than the capital put in, with more than
a third of them being startups buying other startups. In
twenty twenty four, the money is drying up as the

(39:46):
value of VC's assets is decreasing at a time when
vcs need more money than ever because everybody is heavily
leveraged in the single most expensive funding climate in history,
and as we hit this historical liquidity crisis, the two
largest companies, open aron Anthropic are becoming drains on the
system that in a very real sense, are participating in
a massive redistribution of capital reserved for startups to one

(40:07):
of a few public companies. No really hear me out.
Open Ai is trying to raise as much as one
hundred billion dollars in funding so it can continue to
pass money to one of a few public companies, thirty
eight billion dollars to Amazon Web Services over seven years,
twenty two point four billion dollars to Core We've over
five years, and two hundred and fifty billion dollars over
an indeterminate period on Microsoft Azure. If successful, open AI's

(40:30):
venture Telethon will raise more money than it has ever
been raised in a single round, draining funds that actual
startups need. Anthropic has agreed, as I mentioned, to over
seventy billion dollars in compute and chips deal across Google, Amazon,
and Broadcom. And that's not including this hut eight compute
deal at Google is backing that I don't even really
want to think about. Think about it, just pre or

(40:52):
simple for a second. Instead of venture capital going to startups,
you know, early stage companies taking a risk that much
is going to Open AI and Anthropic so they can
hand it to big tech. This is big tech stealing
from Silicon Valley and to see it as anything else
is naive. And this money will come from what remains
of venture capital, private equity, and whatever hyperscalers will hand

(41:15):
over yet elsewhere, even the money that goes to regular
startups is ultimately being sent to those hyperscalers. That AI
startup that needs to keep raising one hundred billion dollars
in a single round isn't sending that cash to other startups.
It's mostly going to Open Ai, who sends it to Microsoft, Amazon, Core, Evil, Google,
Anthropic who sends it to Google, Microsoft or Amazon or
one of the large hyperscalers such as Amazon, Microsoft or Google.

(41:39):
Silicon Valley didn't birth the next big tech firm. It
incubated yet another hyperscalar level parasite, except instead of just
spending money on hyperscalar services and raising money to do so,
both Anthropic and open ai actively drain the venture capital
system as well as they both burn billions of dollars
and need those billions of dollars to keep running the models.

(42:00):
The AI startups can pay them by creating something that's
incredibly expensive to run, so destructively expensive to run, they
can naturally create startups more dependent on the venture capital system,
and the venture capital system has no idea what to
do other than say, just grow baby. Both open ai
and anthropics models might be getting cheaper on a per
million token basis, but they use more tokens, which increases

(42:23):
the cost of inference, which in turn increases the cost
of startups doing business, which in turn means Open Ai,
Anthropic and all connected startups lose more money, which increases
the burn on venture capital. This is a doom spiral,
one that can only be reversed through the most magical
and aggressive turnaround we will see in history, and it
will have to happen in the next year without fail,

(42:44):
and it won't. It's not going to happen and to
find out why. You just have to join me tomorrow
for part three of the in shitter financial Crisis. I'm
Better Offline. This is ed z trn fuck. Thank you

(43:06):
for listening to Better Offline.

Speaker 2 (43:07):
The editor and composer of the Better Offline theme song
is Matasowski. You can check out more of his music
and audio projects at Matasowski dot com, 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

(43:29):
you go to chat dot Where's youreaed dot at to
visit the discord, and go to our slash Better.

Speaker 1 (43:33):
Offline to check out our reddit. Thank you so much
for listening.

Speaker 3 (43:38):
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, film

Speaker 1 (44:05):
Scho

All Episodes

Episode Transcript

Host

Ed Zitron

Popular Podcasts

Two Guys, Five Rings: Matt, Bowen & The Olympics

iHeartOlympics: The Latest

Milan Cortina Winter Olympics

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

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}The Enshittifinancial Crisis: Part Two