There is a disconnect in the

profitability accounting.

The companies that are selling the picks

and shovel are recognizing revenues and

profits immediately. The hyperscalers

and others who are spending those very

same dollars are capitalizing those

costs.

>> These chips are so tight as we speak

>> Mhm.

>> that the prices rental prices for GPUs,

which are really old, like 6, 7, 8 years

old, are going up in price as we speak.

>> I really take a jaundiced eye

on these forecasts of just immense need

for compute at today's prices. Um it

might happen.

But history tells us that that these

kind of insane exponential growth rates

tend to get constrained by the real

world.

>> These people in the memory world or in

the semiconductor world are dramatically

different from the Silicon Valley

people. They're cautious.

>> Extremely cautious, right?

>> Bull markets put a premium on forecasts

and and bear markets put a discount on

reality.

>> Where's to find value in the AI boom on

the long side and the short side? This

is the question that I asked two great

investors earlier this month when I had

the privilege of hosting Jim Chanos and

Val Zlatev. Jim is a legendary short

seller renowned for his short positions

in Chinese real estate stocks, Wirecard,

and of course Enron. Val manages a

multi-billion dollar hedge fund with a

sterling track record for alpha

generation in long short investing in

semiconductors and tech hardware

specifically. This was out of panel I

hosted for MacroMinds Symposium, a

mission-driven for institutional

investors to support student education.

This year's symposium raised money for

three beneficiaries, NYC First,

Opportunity Music Project, and 100 Women

in Finance. Other panelists included

giants in the industry such as Apollo's

John Zito and BlackRock's Rick Rieder.

I'm very grateful to MacroMinds and its

founder Dean Curnutt for allowing me to

be part of it. I'll include more

information in the description as well

as at the end of the interview where

I'll also share some closing thoughts.

Let's get into it.

>> Please welcome Jim Chanos, Vals Lyatif,

and Jack Farley.

>> [music]

>> Thank you everyone for being here. We

got a very special conversation talking

about investing in AI and semiconductors

on the long side, on the short side as

well. Of course, Jim Chanos of Chanos

and Company and Vals Lyatif of Analog

Century Capital Management. I want to

start just your overall outlook on

artificial intelligence and the

build-out that goes with it. Jim, let's

start with you and then Val.

>> Well, as as Rick Reeder said in the in

the past

previous panel, I mean, it is

dominating everything in the financial

markets right now. It it's it's really a

unique concept when it comes to

particularly the equity markets, but

increasingly the credit markets as well.

The one macro comment I'll make since

we're

we're at the macro minds is

is I think people should be a little bit

careful about extrapolating

much broader impacts to global economic

growth than earnings growth.

Um we took a look at the the decade that

preceded the introduction of Netscape

in late '95, early '96

to the decade, the 10 years that had

happened after Netscape, the

post-internet era, and it was not

penalized by the GFC. It ended in '96

'06 '07.

And US economic growth was virtually the

same in the decade before the

introduction of the internet

versus after the internet was

introduced. And interestingly, corporate

profitability,

which

I would have thought would have had a

meaningful increase with productivity

did not increase its growth rate at all.

It was 6% per year, which is the

long-term average in the decade before

the internet versus the decade after the

internet. Now, of course, there's a lot

of dynamism in that and that's

a long-winded way of saying there's

going to be a lot of dynamism dynamism

in the winners and losers

in the AI economy, but whether it'll

contribute to overall

economic growth and/or a long-term

increase in corporate profitability

remains to be seen.

>> Well,

so I'm not a macro investor and I'm not

going to argue about the overall effect

on the economy, but from a micro

perspective, when

we listen to the companies that we talk

to the companies that we invest in both

on the long side and the short side,

what we've seen is that

uh the effects of AI on the actual

individual businesses are they pretty

well seen.

And many of the CEOs of the companies

that we cover quite excited about the

impact so far.

Um and it's very simple to actually just

look at head counts over the last three

or four years compared to the operating

profits of these companies and you will

see that the head count is barely

budged. For some of them it has

declined. Meanwhile, the operating

profits have dramatically increased. And

I'm not even just talking about company

memory companies that have increased

prices, but I'm just talking about a

very wide slew of over 500 companies

within hard tech. That's my universe is

hard tech.

Um so the actual impact is already felt

as we speak and it's fairly meaningful.

To the extent that gets transitioned

from

the early adopters, the technology

companies that are really adopting it

because they kind of eat their own stuff

that they're selling.

Um to the rest of the economy to be

seen, right? We shall see how the whole

thing evolves.

Uh Uh,

but immediate impact is there. And in

terms of the going forward,

I would agree with that there will be

many debates. Uh, there will be many ups

and downs. This is not a situation that

everybody has agreed on.

You go on Twitter, you go on podcasts,

uh, AI lives on these forums,

and you will see numerous bears in

addition to the bulls. So, this is not a

one-sided argument.

There's probably just as many bears as

bulls on the AI argument, which I think

is extremely healthy. Uh, I love it that

there are many, many bears on AI because

it creates a

pause,

uh, stepping back, thinking, considering

it as opposed to just going raw raw into

it, which was kind of more like the

'90s, which was kind of a one-sided

situation.

>> But if you really want to be scared,

I'll tell you that I'm net long AI

versus my short.

>> [laughter]

>> So, that'll terrify you all.

>> And

>> True, I am.

>> So, Joe, I think you're you're net long

uh, via via the index. So, what are you

short on? So, you're you're not short

the semis or or at least in size

relative to the S&P. What are you short?

>> So, we're I mean

we, you know, I'm going to make one

preface comment be- before I I go into

that.

One of the other things you have to also

keep in mind, and we did see this

parallel in the late '90s,

is when you get these type of CapEx,

technology CapEx booms, and there's no

doubt in anyone's mind, bulls and bears

agreed we're in a CapEx boom

in high tech,

there is a disconnect in the

profitability accounting

between the companies that are selling

the picks and shovels, in this case the

chips and data center equipment, and

construction companies, what have you,

and the companies that are spending

those dollars.

Uh, the companies like Nvidia and and

uh,

GE Renova and and uh, Verdiv and what

have you, that are building out this

giant uh capital-intensive

business called AI

are recognizing revenues and profits

immediately.

The hyperscalers and others who are

spending those very same dollars are

capitalizing those costs.

And that's the really important thing to

remember when you're looking at at at

the profit boom that we're seeing in the

high-tech area right now. We saw this uh

from '98 to '01.

Uh from the middle of '98 to the middle

of 2000

um the earnings peak in that cycle uh

S&P operating profits went up 30% over

those 2 years.

Um

basically going at that clip a little

higher maybe even right now.

And then does anybody know how fast it

dropped from mid-2000 to 2001?

>> Pretty damn fast, I'm sure.

>> Dropped 40%.

>> Yep.

>> Yeah. When order books dropped and and

uh costs continued

and particularly depreciation, but order

books collapsed and and

uh profitability of the S&P 500 dropped

as much in that year, which was a mild

recession, as it did during the global

financial crisis. S&P earnings were down

about 40% both both periods. So,

we really have to kind of watch that.

But what we're focused on uh Jack is

what we think are

inherently unprofitable business models

that are attached to this AI ecosystem

where

any way you kind of look at it on a

best-case basis uh the returns on

capital are going to be de minimis.

So, we would look at things like the

Bitcoin miners turn data center

developers, even the neo clouds. Where

if you make just heroic assumptions

on profitability

and you give them 10-year life on the

chips

uh uh

you still get four, five, six

returns on capital in the out years.

And I just think that that those are

going to be winnowed away over time.

I've joked with my clients that you want

to be long with the chips produce, not

where the chips reside.

I think that that's probably still a

valid investment thesis going forward.

>> So, I I want to talk about the Neoclads

in a moment, but first let's talk about

this depreciation question. We're in a

CapEx boom. Most of the chips that are

being bought are being capitalized, so

they don't uh you know, they don't go

out of operating expenses, they go out

of they're capitalized and then they are

depreciated away over many years,

whether it's four years, it's seven

years, whatever. So, are you saying Jim

that the earnings are inflated? You you

I mean, I think you did say that, by the

fact that the depreciation hasn't hit

yet. Just when is the depreciation going

to hit and how is that going to impact

profits? I know Val has a lot of

thoughts on this, too.

>> So, there's two problems. Um

one is that

an awful lot of this capital spending

for the people spending the big money

like Alphabet and and Microsoft and

Amazon and Oracle, um a lot of that is

going into what's called construction

progress right now.

And and that all of those costs, the the

cost of the chips, the cost of the

labor, the interest costs, all of that

is capitalized, it's not expensed until

the the data center comes online and is

producing revenue.

So, that's an important thing. So,

for setting aside the life of the chips

themselves, there's now, increasingly

because of lags, there might be 12 to 18

months where you've spent money on the

data center, but it isn't producing

revenue yet and you aren't depreciating

those assets.

Now,

as a cynic on this stuff, in order to be

conservative, what we are using in our

modeling is 10-year life on the GPUs.

There's misconception that the

AI bears are saying, "Oh, it's 2 years

or 3 years." We're using 10-year life,

which is basically you running these

things 365

24 hours a day.

You're not going to get physical life

much more than 10 or 12 years out of

them. So, so to be safe, I'm looking at

business models and assuming that you

write the GPUS GPUs off over 10 years. I

think that's a safe bet.

>> Well,

um 10 years is pretty aggressive,

probably. I'm not sure how many GPUs

will be there in 10 years from now.

Um

very safe bet, I'm sure.

So, I would agree with Jim on the the

real bet or the real investment is

really the chips or the servers or

whatever it is that goes inside the data

centers

uh as opposed to the landlords, as you

call it. I think it's actually very good

statement. So, I'm not going to argue

that the new clouds are fantastic

investments. I think that

uh

uh from a depreciation perspective,

yeah, we can focus on depreciation.

Maybe it's not 10, maybe it's six,

whatever it is. Definitely it's not two.

The reality though is that

um these chips are

so tight as we speak

>> Mhm.

>> that the prices rental prices for GPUs,

which are really old, like six, seven,

eight years old, are going up in price

as we speak.

>> Uh that wasn't the case until December.

Uh they started to increase. By the way,

into December, these prices were down

20, 30% year-on-year, which is very

normal. GPU rental prices should be

going down every year. You have to

expect that. You have to build that into

your models because new GPUs, new

architectures are coming, which are much

more efficient. And the dollar per token

is much lower for the new GPU. so

there's absolutely no reason for anybody

who has access to new GPUs to even hold

them. They should just throw them into

the ocean, put new GPUs at the end of

the day because the tokenomics is so

much more efficient.

The reality though is that it's so tight

since January that now prices are up

40-50% even more as we speak. That

definitely changes the economics of the

new cloud in the near term.

I have no idea that will continue or

not. All I'm trying to say is that this

is a very dynamic market.

>> It has changed the valuations of the new

clouds. I don't know that their

contracted prices have changed that

much.

>> Uh yeah.

>> Because hyperscalers aren't dumb

themselves. And remember

in this business model, these are

equipment leasing companies in effect.

Uh if you are if you are

taking buying chips from Nvidia and then

renting data center space from somebody

else and then renting the chips out to

Microsoft or Google or Meta,

you're an equipment leasing company.

You're you're not a high-tech company.

You're a finance company in effect. And

you're making a bet on the life of the

chips and what you can get over the

contract and and and

but but when some of these companies

that are truly and many of them are run

by former finance people, um the core

weave guys are the old Magnetar guys.

>> Yeah, yeah.

>> You remember them from the global

financial crisis.

Um and so

you know, these are if Blackstone is in

your business, Blackstone just got into

this business with a new rate, you know,

you're in the finance business.

And that's a really important point to

remember and always remember that the

hyperscalers can buy the chips

themselves.

They're choosing to rent them from the

new clouds. Why?

>> Well, I'm sure I'm not sure exactly what

the answer is on the why, but I'll tell

you why. The reality is that they were

not prepared for it and they don't have

the access for it and Nvidia wants to

create a balance between the

hyperscalers and the new clouds because

in video it doesn't want to get beholden

on four customers in perpetuity. So,

they choose to actually feed a

competition for the hyperscalers.

Uh hence they give more supply to the

sorry, to the new clouds.

>> So, they don't want to sell to Microsoft

directly, they'd rather sell to

CoreWeave who then

>> I'm saying they sell it to both.

>> Leases to Microsoft.

>> Jim, what do you think the answer is?

>> Uh the answer of why the hyperscalers

are spending money via CoreWeave rather

than just themselves on their own

balance sheet.

>> they're they're spending money on both

clearly. Yeah. Yeah, in fact they spend

money on both. Yeah. [laughter]

The the the amount of money that the

hyperscalers are spending directly is

massive. But again, it's it's a gold

rush mentality. So, they're they're

Whoever has capacity will probably sign

a deal.

The problem, Jack, is is that given that

dynamic right now, you should be getting

really good ROIs. If you have capacity

of a power data center right now and you

have the chips or someone will bring the

chips, you should get incredibly high

ROI returns on invested capital right

now. This is If not now, when, right?

And these deals where you get

granularity on the deals where they

actually give you quite a bit

are working out penciling out at 7%, 6%,

5%, 8%. They're all They're all single

digit ROICs pre-tax.

And so again, it gets back to my point.

If that's the best you're going to do

now, I'd much rather own other parts of

the of the chain than just the

middleman, if you will, the financial

middleman.

>> So, I totally agree with you there.

One thing that you talked about is that

you basically

classify them as

REITs effectively, right? The CoreWeaves

of the world or whatever.

>> Yeah, they're equipment leasing

companies.

>> Um

and it's I agree with it's much more

about technology at the end of the day.

Technology is the ultimate

differentiator in that game.

>> That's the value add.

>> That's the value add. Hence, the value

add is in the chips and the wrappers of

the chips that go inside the data

center. There is not a ton of technology

in somebody buying land or having access

to a grid capacity or putting some

Vertiv transformer or whatever it is

over over the connection.

>> in short it for a couple of years and

will.

>> They will say, but eventually the value

does move to the technology drivers.

>> I don't think we're going to have power

bottle next, for example, 3 years from

now. I don't think we're going to have

labor bottle next 3 years from now. We

may have them for the next 18 months,

but ultimately equities are long

duration assets, right?

>> Totally, yeah.

>> And you should be looking at the the

core business over the whole cycle or

over long

long periods of time. And and pricing an

equity off current spot prices in a

shortage

can be exciting. It has been.

But it can also be valuable.

>> All right.

So, I'm not going to defend the new the

the new cloud since I don't invest in

that at all anyways.

Uh but I wanted to make a comment that

they're not exactly the same as some of

the REITs like Equinix or Digital Realty

or whatever you call

>> Those are the legacy guys, yeah.

>> These are not only the legacy. These are

guys that have just shells. You use a

customer bring your own servers. You

stick them into a cage and you say, "Oh,

thank you. I'll just pay you for the

rental and I'll come back in 10 years to

change the servers."

Coreweave is such a Nimbix, especially,

they actually do have some technology

above and beyond what they buy from

Nvidia. They do have software layers.

They do have optimization layers.

Nimbix, for example, doesn't have 100%

of their revenue contracted out to

hyperscalers. It's about 50 to 60% and

the other 40-50 is actually used for

inferences we speak. And that's where

they can actually price a lot more in a

spot because inference adoption right

now is the one driving the spot

increases. And they can pass that

through and benefit from it as we speak.

So, it's not exactly the old dumb

shells. There's definitely some

technology, but it's not the technology

that's driven by the semi guys. The

technology coming from Nvidia or

Broadcom or whatever whoever it is the

sterile labs doesn't matter.

That is a light years above the

technology being provided by

a core before the world.

>> I understood. But but also keep in mind

this is the technology space and

technology can change and we could see

inference going to our phones or to our

desktops. I know that that there people

said no no it's not economic and never

will be economic.

But some of those same people are also

then telling me that we're going to put

them in space. So, you know

>> Let's talk about data centers in space.

We have there.

>> [laughter]

>> Jim do you want to go with the space?

I'm sure he has space arguments.

>> Okay, what do you want to know about

data centers in space?

>> Is it a good idea? Should should we be

investing in these in data centers in

space?

>> Well, you're going to get a big chance

next week.

There's

>> [laughter]

>> the only way that that thing works is if

there's lots of data centers in space

and and pile drivers on moon and

colonies on Mars. But anyway, um

Look,

so so

the costs of of of

putting a data center in space are

obviously considerable have a lot to do

with launch costs. Um but

couple of observations knowing the data

center space pretty well.

Power costs are actually despite the

bottlenecks power costs are very small

percent of data center costs. They're

about 5 to 7% of revenues.

So, if you're doing this because the sun

is a free source of power,

you you're starting on the wrong foot.

And and the power is not the problem and

in fact I think power will be as we

discussed less of a bottleneck going

forward. Um so the the cost the other

the big costs in space are radiation

because it's hard to radiate in a

vacuum. So, you the space station, for

example, has these enormous radiators.

So, that's number one.

Number two is radiation itself and

complex systems

exposed to space radiation over long

periods of time tend to break down.

But then you get into simpler things

like the the idea of redundancy and

insurance, right? Like if stuff breaks

in data data centers all the time. If

you look even at the simple old legacy

data centers, their capital their

maintenance cap exes through the roof.

You know, stuff breaks. The HVAC goes

down. This goes down. That goes down.

Things are always needing replacing. You

send a tech out with the right part, the

right equipment, they replace it, and

you're back up and running.

In space, you got to

send a launch, you know, hopefully with

a humanoid robotic uh to do it, but but

you have another launch.

And so,

you begin to get into issues of

redundancy,

insurance, and whatever. And then the

cost whatever cost savings you might be

getting begin to immediately erode

dramatically. And then, of course,

there's the problem that the the vehicle

that the prime

uh the prime uh

the

proponent of this

uh that's coming public next week, um

their Starship hasn't made Earth orbit

yet in 12 flights.

I I keep reminding people of that that

all these great promises are built on on

a rocket that has not yet achieved Earth

orbit um

and has blown up, I think, six or seven

out of the 12 flights.

Um so,

we'll have to see. I I you know,

obviously, it's a it's an amazing story.

As I said, you know, this the TAM of

space is infinite.

>> It is. There's all this space there.

>> So, then yesterday I pointed out, yes,

but the TAM of space is infinite versus

infinite entropy. If you you know,

randomness in space is infinite, too.

So, you know, it's going to be a

tug-of-war that I don't think anybody

has to worry about for the next five

five or six or seven or 10 years.

>> Jim, I just want to get your thoughts um

on the SpaceX IPO. I take it you won't

be a a buyer other than via the index.

Um

you know, shorting shorting new issues

is is you know, famously quite risky,

but you have the S-1 is out, so you

you've had a a chance to look at it. How

are you thinking about shorting that

that company both in terms of you know,

whether you're actually bearish or

versus actually putting a position on,

which is a completely different thing?

>> You know, so look, I mean,

the numbers don't work on the existing

business. Even with Starlink uh Starlink

is profitable. Starlink is is a decent

business. Um its growth has slowed

dramatically. They've had to cut price.

And the the prospectus says uh points

that out. Um to drive unit growth. Um

but it's it's a profitable business.

It's earning about $4 billion right now

annually uh operating. Uh and we think

about 25 to 30 billion of invested

capital.

So, it's it's it's a good It's not an

insanely good business. It's It's a good

business.

Um the problem is the launch business

still loses money.

>> I was surprised to learn that in the

S-1.

>> Yeah, the launch business is still

losing money after spending billions and

billions and billions.

And and part of it you know, trying to

get Starship to work. Um and also the

launch business subsidizes Starlink. So,

you have to be a little careful.

Starlink may not be as profitable as it

says it is because it's getting cheap

rates to launch from from its parent.

Um and then XAI is the wild card, right?

It's I mean, it's losing lots of money.

It's spending lots of money.

Um it cut a very short-term deal with

Anthropic for space for rental space.

Um but it's just a sinkhole right now in

terms of cash.

So you have to believe in in

Mars and the moon and data centers in

space to justify almost two trillion

dollars. I mean, it's like Tesla itself,

right? Tesla you can't justify on

selling automobiles. It's all the stuff

that's going to come.

Um like I said, bull markets um you

know, put a put a premium

on uh on forecasts and and bear markets

put a discount on reality.

I think that's that's the truth.

>> I I want to know um when you talked

about valuing cyclical businesses as if

they're secular businesses.

>> Can can we just finish up with the

SpaceX thing? So

putting the IPO aside,

um

I want

push back a little bit on one thing.

When

Elon Musk is talking about data centers

in space, he's not

he doesn't want to put him there because

it's cheaper energy. Of course it's

cheaper, but you're absolutely right.

Energy is 5% of the cost of uh CapEx,

and another 10% is the

the the shell and the land and the

equipment. 85% is really the stuff that

goes inside the data center, which is

basically what we invest in.

Uh so it's not about the cost. It's

about the

amount of electricity or amount of

compute heat sinks in his mind is

needed. Let me just dimensionalize it.

He was very specific that he believes

that the world over the next several

years will need 1 terawatt of compute

capacity. In his imagination.

Let me just dimensionalize that.

A terawatt is a thousand gigawatts.

The amount of CapEx being spent now by

the hyperscalers and Oracles or whatever

of the world is about 750 billion

dollars,

>> Mhm.

>> which is about 15 gigawatts at the most.

So, he's talking about a thousand

gigawatts versus what is being spent

this year, which is 15 gigawatts. So,

he's basically saying, "All this stuff

right now is kind of a waste of time.

We're just gibber-jabbering about small

amounts of money. It's much bigger than

what you'd think."

That's why he's going there. It's not

the cost, it's the amount that's needed.

And by the way, the full grid in the

United States is like 1.5 terawatts.

And you need to have a spare. So, he's

basically saying, "I need the full grid,

period. That's why I need to go to

space."

I have no idea how to discount his

timelines or ambitions. That's for other

people. I'm not sure it's a great idea

to be shorting him because it hasn't

worked out for many people over time.

What I do know for a fact is that the

reason he's saying this he believes

there's a thousand terawatt, sorry, a

terawatt of need for compute

is because he doesn't see a break

in the basic technological scaling laws

in AI that exist.

What that really means, laws in AI are

the bigger the cluster, the more compute

you use to train a code, the better the

output, the higher the IQ of the code.

And everybody's trying to get the higher

higher IQ all the time.

If he was seeing a break in that,

he would not have even remotely

mentioned that he needs a terawatt of

compute capacity. He'll be like,

"I already have it all.

Some of my stuff is empty anyways. I'm

renting it out to Anthropic as we speak

because my stuff doesn't quite really

work all that well."

That's the bottom line. It's a

technology argument.

>> Well, were you Were you investing in

'99?

>> Uh I was in McKinsey then. I was working

for these companies.

>> Okay.

>> And I felt the pain because I was next

to the CEOs of these companies when they

bookings went down the drain.

>> So

So

post-Netscape, but but really toward '98

and '99, one of the guiding

unending truths of the internet was that

traffic was doubling every quarter.

And MCI WorldCom went out of their way

to tell people that on quarterly calls.

And it it was one of those things that

just then became

embedded in the psyche that the internet

was growing so fast you could not

imagine because it was doubling with

power scale laws and even costs coming

down, if traffic was doubling every

quarter.

And so

um it was very interesting. There was a

a gentleman from Bell Labs at the time,

Anthony Aczilco, you can look him up.

Um he put a paper out in early 2000, I

believe it was, but but he circulated it

in late '99.

And he pointed out based on a lot of

rigorous data that he looked at,

internet was really growing fast. It was

doubling every year, not every quarter.

Um still fast, right?

And uh and traffic continued to do that

for a number of years into the 2000s.

The problem, of course, was was that

everybody was building their business

models and order books based on this

belief that

I can't go wrong. Whatever I spend my

money on, it is going to be taken up by

by internet traffic. So the networking

companies, the phone companies, the the

long distance everybody

the capex boom just accelerated. Um and

and then the realization hit in in early

2000 that

uh that someone had kind of made that up

at MCI. And everybody just run with it,

the media, whatever. And and it's my

view, having lived through it and seen

and we were short Lucent Nortel at that

time at MCI,

um was that order books

suddenly collapsed. As CFOs and CEOs

told everybody, "Okay, we don't need

20,000 routers

um you know, this year.

Just cut our order back to 4,000.

Uh

and and the biggest spenders back then

is a myth. By the way, the biggest

spenders back then were enterprises.

Were were big companies like AT&T, uh

Merrill Lynch, Bank of America,

uh Coca-Cola, who were networking

their their equipment

to talk to each other. And then on top

of that, you had Y2K.

I know we replaced all of our PCs cuz we

were terrified in uh in the second half

of 1999. Um so, I'm I really take a

jaundiced eye on these forecasts of just

immense need for compute at today's

prices. Um it might happen,

but history tells us that that these

kind of insane exponential growth rates

tend to get constrained by the real

world.

>> Yep, you're absolutely right. That

should be taken with a 10 grains of

salt, especially knowing from whose

mouth that's coming from.

Um

couple of thoughts on I think we should

finish the '99, '00 comparison because

that's a interesting comparison.

You're absolutely right.

There was like much slower growth than

what MSCI or whoever it is was talking

about.

>> But still still quite quite rapid.

>> Yeah, but exactly.

Um

right now the growth actually can track

directly yourself by just looking at the

for example open routed token counts.

And you can see the growth of tokens

being tracked, which is a small

percentage of the industry token usage.

But you can at least see the as opposed

to waiting for some corporate CFO to

show up once every 3 months to tell you

something that he made up in the back

room.

So, you can track that much more

directly.

Um and the reason the

uh GPU rental prices going up is because

the token usage is exploding and they

just don't have enough GPUs to run the

tokens for. This doesn't mean that will

continue in perpetuity. We can discuss

what can break that.

But at least for the time being, the

real facts are suggesting that

you don't need to to listen to CFOs. You

need don't need to listen to some

accountants to tell you what the growth

is. You can just see it for yourself.

And you can By the way, you can see it

in your own usage, in your own offices.

So, that's one thing.

The other one is when you talk about

'99, 2000, there's actually two

technology differences, very different

from right now.

Number one,

of course, all the spend was on fiber

rights, fiber into the ground, the fiber

horizons that you mentioned.

>> Yeah, that's not necessarily true.

>> And Cisco routers and switches that you

had to hook up to the to the stuff.

>> of it was PCs, but yeah.

>> But, PCs were fine.

But, let's put the PCs aside. Everybody

talks about the fiber glass, right? The

dark fiber. At that point in time.

>> The fiber company spent only $50 billion

in aggregate in 5 years from '98 to '02.

The CLECs spent another $50 billion. We

went back and looked at the numbers. So,

the the two flawed bankrupt business

models of that cycle, fiber companies

and CLECs, spent a total of $100 billion

over 5 years,

or or 20 billion a year. They were a a

small part of the overall TMT spend back

then.

Um

>> Well,

>> Most were profitable companies spending.

>> Well, by the way, they were profitable

companies spending. Uh I agree with

that. But, a lot of the spending was,

you know, it was coming from the

revenues of Ciena and Cisco.

>> Yeah, but a lot of people thought it was

all it's the dot coms and the fiber

companies.

>> Well, other people say that.

>> Yeah, that was that was a small amount

of money.

>> But, but the the the explosion in the

Cisco and Ciena revenue back then,

right? JDSU, which I'm sure you

remember. You were probably selling it.

Uh which is currently Lumentum, whatever

it is.

All of these companies, that's

fiber-related stuff. They were selling

something fiber-related. Even if it

wasn't just pure fiber, it was the

switches, the routers, the lights on the

end of the fiber.

>> Yeah, but that a lot of companies

ultimately realized they didn't need.

The the they they didn't talk about

that.

>> There were two reasons why they didn't

need it. Number one is when you install

fiber into the ground, 70% of the cost

was fixed cost of the blue collar

workers with the bulldozers. That they

have to come, dig a trench,

put the fiber, leave.

So, if you spend 70% on a bulldozer with

a blue collar worker, might as well put

as much fiber as you can humanly

possibly. Of course, you will overbuild

it. I would overbuild it, for sure.

Um so, that's one. Second, they will say

technology changed

around multiplexing.

And basically, the multiplexing allowed

time division multiplexing allowed to

increase the amount

>> Technology So, as I said earlier, as

technology changed, and and who's to say

that's not going to change with token

usage?

>> Well, we don't know. Um

that may very well change. Here It It By

the way, it's very possible to change.

The reason that token

usage is going up is the scaling laws I

was talking about some extent.

And these scaling laws could change.

Somebody could break them. They're not

physical. They're not physics, they're

not mathematics laws, they're empirical

laws. They're like Moore's law. Moore's

law was alive like 35 years. Scaling

laws for AI have been around 12 years.

That may change. So, if somebody comes

up with a new AI architecture, new

model, which is not a large language

model, which is not a transformer, just

a new model,

that somehow gets a lot for nothing in

terms of capacity investment, this whole

discussion changes.

>> And China hasn't done that, cuz some

people say China is is doing that.

>> Well, people people say that very

emphatically when Deep Sea came out in

January 2025,

and all of these AI companies sold off

by 30 to 60% over 3 weeks, because

everybody felt that, "Oh my god, this is

the multiplexing phenomenon, right? All

of a sudden, we broke the scaling laws.

We can get a ton of tokens for basically

very little cost." That obviously wasn't

true.

>> You should be very careful of things

China says.

>> Yeah, you should be very Well, but there

are many people in the Silicon Valley

that continue parrotting that whole

China paradigm also.

>> As my friend Jim Grant calls it, the

People's Republic of Madeoff. I mean,

you know,

>> [laughter]

>> um

uh

uh And that's some experience with

China. It's

>> So, DeepSee wasn't that, obviously.

>> DeepSee was wasn't it?

>> was not it. It did not make a scale up.

It was just the next step in reducing

costs of tokens.

Uh combining several different

algorithms that were already known to

everybody in the world.

Uh There may be something else though.

If that happens, picks and shovels, we

have to have a very different

discussion. That is the nightmare

situation. So, if somebody keeps me

awake at night about my lungs,

it's that.

>> That might be a segue to the first

slide.

>> Uh

Uh

>> I'm not sure how to segue, but

>> Yeah, I I I want to I want to talk talk

about memory. You know, the history of

chip making, memory is has been a

commodity business. You have

everyone's competing with each other,

producing as much memory as possible,

prices are going down, companies are

going out of business.

Why is it different this time? And I

actually, you know, should say as many

people know, the three big memory

producer producers, uh you know, one of

which is American, two of which are

Korean, their stocks have gone up so

much, their actual forward price

earnings ratios have gone down because

their pricing power expected has gone up

so much. But why is this different? You

know, everyone knows that like okay, you

um you know, the time to not buy an oil

company is when it looks cheap because

when the price is at 150, forward price

is is six, but that's not the good time

to buy it. Why is this different?

>> Well,

the the most dangerous word that this

time is different.

>> Yeah.

>> So, I'm not sure this time is different

in the sense that I'm not going to sit

down and argue that memory prices will

never ever ever come down.

I've lived that for long enough time.

I've been on the buy side with 26 years

or whatever it is. I've seen that movie

so many times.

Uh

What I do believe though is that

the

amplitude or the peak

in terms of need for memory is higher

than anytime before

over the last 25 years.

And I think the peak is shallow for a

while, and for a while that could be

like 2 3 years, whatever it is, 4 years,

before there is a rollover in pricing of

memory.

The market right now is discounting

believing that that rollover with a big

sharp price decline in memory prices is

like 6 to 9 months out.

That is

because of that belief, this memory

stocks are trading at 6 7 forward

multiples. I mean, not even not even

forward, 2026 multiples.

So, they're like the cheapest dirt in

the world, 6 to 7 multiples.

The only time you believe you have a

multiple like that is if you believe in

a imminent downturn, like 6 months out

downturn.

This is unlikely to happen. Let me

explain why it's very unlikely to

happen.

Um

it is very much supply constrained,

and it is very, very hard to add

capacity very quickly willy-nilly.

This is true in general for

semiconductors, and we can discuss why

semiconductors are actually the ones

putting the brakes right now the whole

AI boom, which by the way would have

been way bigger than what it currently

is if if it wasn't for the brakes on the

semi guys.

But the two reasons they why I cannot

add too much.

Um number one, even if you have infinite

amount of clean room, clean room is

these big facilities which are super

clean inside so there's no contamination

of the wafers.

Even if you had infinite amount of that

stuff,

you have to have equipment.

The equipment makers like ASML, Applied

Materials, whatever it is, they cannot

really grow their revenues or their

shipments by much more than 30% a year.

It is a supply chain complexity that

constraints they grow to about 30 to 35%

a year. That's kind of a max.

So, you just cannot cannot add more than

30 35% per year bits. Bits is the piece

of informa- piece of cell in a way for

that stores the information.

That is the ultimate determinant. You no

matter what you want, you just cannot

add more than that. Oh, by the way,

there's not enough clean space either

because

the memory makers were going for a

downturn

where their prices were going down and

margins were going down all the way into

through 2024. Even in the beginning of

2025, prices were pretty darn weak.

These are

these people in the memory world or in

the semiconductor world are dramatically

different from the Silicon Valley

people. These are like 60 70-year-olds

with a lot of experience. They have seen

that movie many times before. They don't

believe any 30-year-old that shows up

from the Silicon Valley telling them,

"Oh, I need like a hundred times more

memory than what you have ever made."

>> They're cautious. They're cautious.

>> Extremely cautious, right? So, they

never actually even prepare for this

additional uh clean space that they

needed.

>> Except the CEO of Taiwan Semi last night

actually pushed back on that very idea.

The CEO of Taiwan Semi is one of these

70-year-old guys who's seen seen it all

and has seen the cycles. And there is a

belief out there

that those people are keeping the brakes

on expansion because they don't want to

expand too fast and see the downside of

the cycle.

He actually said last night, he said,

"No, we're we're gearing up as fast as

we can.

Um you know, there's other bottlenecks

out there, but we are going to be

building chip plants as fast as we can."

And so, he did push back, interestingly,

on the belief that, "Oh, well, there's a

bunch of old guys in Taiwan and Korea

who aren't going to let this get out of

hand." And so, pricing will stay

>> But that's your operating at full as

fast as we can. It is true. They're

doing it as fast as they can.

Back to my main argument, the equipment

companies cannot grow more than 30% a

year. That's as fast as they can.

>> That's a different That's a different

constraint, but okay.

>> This is the physical constraint. The

physical constraint here. This is

This is not like, "Oh, I just want to

add It just can't. They're physically

constrained." By the way, this facility

is expensive and they take 5 years to

build.

>> So, that will increase then the cost for

everybody else.

>> For sure.

>> Yeah.

>> Me- Yeah, exactly.

>> work both ways.

>> Definitely, semiconductors are very

inflationary.

>> Yeah.

>> The minute Moore's Law slowed down 5, 6

years ago, deflation is same as stopped

and it became inflationary. And they've

been driving inflation across the board

for the last 6, 7 years.

>> And so, now you have been looking on the

skeptical side, uh to use a

Jim Chanos' word, at the users of

memory, right?

>> Oh, yeah. Um

So, memory has gone up in prices, many

people know. DRAM, flash me- So, DRAM is

the memory where you store your

operating systems, super fast, super

expensive. Flash is where you store your

pictures and videos, whatever it is, uh

much cheaper.

Um These prices have gone up 4 to 5 x.

It was 100% driven by the data centers

because

first of all, the models changed from

pure chatbots

to reasoning models that require a lot

more tokens that need to be stored, a

lot more of that. Then you had

increasing context windows, which is

context windows where you actually ask a

question or you throw a million-line

software code you want to

change or improve.

And then at the end of the day, over the

last 5 months, uh the agents came around

and the agents just suck a ton of uh um

They They need to store a lot of more

information than anything else before

that. So, the storage need of AI during

hyperscalers just exploded over the last

12 months because of these technology

changes. It wasn't a willing They just

ordering stuff because they feel like

they don't need it. They just needed it

right now.

Of course, the memory guys are not

prepared. Prices went up for the roof.

They are 4 to 5 x as we speak. They'll

go more.

They will go more. They'll definitely go

more. They're definitely going more.

They're going like 30% of quality easily

now.

Uh

the

the issue now is for PC makers,

smartphone makers, consumer electronics

makers, all the gadgets that we use,

right?

If you're a like an Apple or somebody

else,

your bill of material that you were

paying bill material is the cost that

you were paying as part of your cost

structure for memory used to be 20%.

For a PC, smartphone about 20%.

Now it's like 50%.

The only way

this memory this this manufacturers can

survive, by the many of them

working for like 5 6% margins, software

margins, the only way is for them to

pass the cost in increase to the

consumer.

That is why smartphones are going up in

price, not Apple, but everybody else.

PCs are going up in price. You can

actually go and see it in the store at

Best Buy right now. They easily up a lot

compared to before.

So they pass it through that. Consumers,

but that's a very elastic market. We as

consumers we see a PC being up 50% in

price, well, you know, we wait. We're

going to wait for another 12 months in

the hope that the price comes down. So

that pushes the units down. So the units

for PCs and smartphones this year right

now are probably down mid-teens.

Very rare to be seen by the way. Almost

never you see that situation. They just

flat. These are like ex-growth flat

markets for like decades over a decade

even for smartphones.

Uh down 15 is not fun. So this

I think there's shorting opportunities

on a bunch of component makers that

actually sell components to the PC

makers or the smartphone makers that

don't have the pricing power. They just

take it on the chin on a unit decline.

That's what I meant.

>> Okay, that that makes sense. Jim, what

what do you think about memory prices?

And I mean, certainly there is a price

or maybe there's not of of DRAM and NAND

and memory where

the supply chain is incredibly incent-

incentivized to build production as as

quickly as possible. I mean, do you do

you agree with what what Val is saying

that they already are moving as fast as

possible and there are just physical

constraints that cannot be surpassed.

Cuz you know, when lithium, cobalt, oil,

natural gas, anytime the price goes up,

all the CEOs and the mining people say

like, I mean, just to to build a mine

takes 7 years. We could never do it. And

then a year later the price has

collapsed, you know, cuz of supply has

gone come online. I know memory is

different, but just want to get your

thoughts.

>> Yeah, I mean, I in my 40 years I don't

think I've ever made a single dollar

being short

the DRAM companies. I just it's it's

it's a cyclic business. Um people go way

overboard on the way up. They get way

too pessimistic on the way down. Um you

know, it's it's

it's just a business we've never been

able to time correctly, so we've

generally not played in it. Um and

generally have not played in the pure

semiconductor um

area. I would however point out now that

um we're getting also in the CPU area.

>> Yeah.

>> Um but we're getting uh

some really interesting

uh deviations in valuations on

businesses that that, you know, are

going to be around and profitable and

growing for the next 5 to 10 years um

versus uh companies that are now trading

these trade at two times revenues and

now trading at 10 times revenues or 12

times revenues.

>> What do you mean, Jim? What do you mean?

>> Well, I I mean so so you look at at some

of the the CPU companies that have just

taken off.

>> Um You talking about like Dell, HP kind

of stuff?

>> I'm talking about Intel.

>> Oh, Intel and AMD.

>> Yeah, yeah, yeah, AMD. Uh and and and

then you look at at the companies

Taiwan semi GPUs, Nvidia,

um and others, uh Broadcom. Um

you're beginning to see

you know, some pretty amazing valuations

on companies that are still going to be

in pretty competitive markets versus

companies that are going to be in

oligopolistic

markets.

>> Yeah. What What you're saying is that

Intel has gone up so much and it's in a

competitive industry, whereas Nvidia is

dominant and Nvidia is still cheaper

compared to these

>> Much cheaper.

>> Yeah. Yeah. Yeah.

>> Much cheaper.

>> Do you agree?

>> Oh, Nvidia is definitely much cheaper

than Intel.

Um

Let's put that in perspective, though.

Intel hasn't made money in a couple of

years.

Uh Intel used to be a monopoly forever.

They lost everything They lost a lot to

AMD, fell behind.

>> Yeah, I'm not looking at them on

earnings, but what I'm looking at them

on is revenue.

>> Yeah. Yeah, I understand. You're looking

at revenues.

Uh

so I'm not going to argue with you on

Intel one bit. What I do want to make a

point of valuations, right? Because sort

of that ties up to this whole thing.

Stuff has gone up. Some things are

expensive. Some things are not

expensive. My point is that

this hasn't been even though a lot of

these semi companies have gone up a lot

over the last 2 months.

Um

this is not a situation where all the

valuations of every single one of them

is through the roof, which by the way

was the case in '99, 2000, which I do

remember. Even a thing like Cisco back

then was trading at literally 160 times

P multiples.

>> Have you seen Tesla lately?

>> That's between you and Elon.

>> [laughter]

>> Uh but

uh

>> There are There are companies trading at

150 times earnings in the market.

>> I'm sure by the way, but

>> But not in the Not in the SMH. Not in

the semi conductor side.

>> Not in the SMH side. Not in the semi Not

in the semi side.

>> Not in the semi side. So, on the semi

side or even the hard tech side in

general,

you can have probably the most

exaggerated valuations right now are

more on the networking side, which are

like 50, 60 forward multiples.

On one end, that's the most extreme

versus Tesla, whatever it is.

Uh memory I've already mentioned five,

six, whatever it is on the extreme other

end, Nvidia's at like 15 times on 2027

EPS.

Broadcom, after the decline this

morning, is at 12 times 2028 EPS. So,

this is not a space that is like '99

where everything was frothy and has gone

nuts in this evaluation.

>> Mhm. Uh what do you What do you make of

>> I don't even know why Costco is more

expensive than just about any similar

company or why

>> Walmart.

>> Or Kroger, I don't know what these

things are. They Yeah.

>> Yeah. What Okay, so the the the Costco

of the semiconductor index, I think is

the um the equipment manufacturers that

supply TSMC and the the memory

providers, so that's, you know, Lam

Research, ASML, as you mentioned. And I

know you're you're you're kind of not in

love with that sector, not talking about

individual companies. Why do you, you

know, it seems like even though those

Do you think that those don't merit

those high valuations cuz isn't it very

likely that, you know, if they're

selling to memory companies, memory

companies are going to need their their

products very high? And they have the

what's it called, the razor blade model

as well.

>> Yeah. So, look, these are very sound

companies. They have fantastic business

models. They're effectively

semi-monopolists in their niches.

Uh some of them are monopolies in their

niches. Um the the issue is back to what

I was talking like 5 minutes ago, 10

minutes ago, that

their growth is capped at about 30%.

And they don't increase prices to

increase the growth above 30%. And

they're already trading at like, I don't

know, 35 forward multiples.

Uh 35, by the way, is not horrible for a

30% grower. It just It's much more

expensive than Nvidia or Broadcom or

some of the others that have even higher

margins. So, it's more of a relative

situation supposed to, oh my god, these

are bad companies. Nothing like that.

>> I see. Um well, we're running out of

time, Jim. I'll give you the the final

word.

>> Um well, I think I mean, we probably

agreed on more than than we've

disagreed. I I

think that's fair to say. Um

I think again, it's it's a market in

which there's going to be opportunities

on both the long and the short side. And

in AI specifically as well,

um and just I would tell tell um

the attendees today to just be careful

that you're not putting

magical valuations on mundane

businesses. Um because one of the things

we do know is capital is flowing

immensely into this space and that tends

to reduce returns.

Um and it will flow to to everybody in

in this part of the cycle.

Um it will stop flowing to those

companies that have the mundane business

models going forward as that becomes

clear. And I suspect we'll know that

within the next 18 to 24 months.

>> We'll leave it there. Thank you very

much.

I hope you enjoyed that as much as I

did. Val talked about potential

opportunities in the memory and optical

photonic space, whereas he seemed a

little bit more skeptical about the

semiconductor equipment companies that

supply the semiconductor fabs.

Interestingly, Jim Chanos is not short

at all any of these semiconductor

companies, and instead he's looking at

being skeptical about data center

players such as Coreweave, so-called

neo-clouds, as well as legacy data

centers that may be made obsolete by AI.

I want to thank MacroMinds again. I will

include in the description a link to

where you can donate to the MacroMinds

Foundation as well as more information

about the three nonprofits that it

supported this year, NYC First,

Opportunity Music Project, and 100 Women

in Finance. Over the past year on

Monetary Matters, I've shared my view

that semiconductor earnings would surge

on AI CapEx, and I've been helped

enormously by guests such as Citrini and

Angus Shillington from VanEck, as well

as others. I still think that there is

value to be had on the long side. I

still like broad semiconductor exposure,

and though I've had some success in

owning call options on Marvell and

Teradyne, if you had to ask me the name

that I'm most excited about right now, I

would say Nvidia. At the same time, I'm

attentive to the risks that this is a

giant bubble and that the return on

investments for these vast sums will not

materialize. I also think that regarding

the status quo where semiconductors earn

tremendous profits while the model

companies report huge losses, while

ultimately that status quo is

unsustainable, I do think that it could

continue for a year or perhaps multiple

years. That is one thing I have noticed

about technological booms is that they

frequently last longer than many people

think.

But if you wanted to know what I think

under my head are semis long or a short

right here, you have my answer.

I think that even if this is a bubble

and that the bears are right, being

outright short semiconductors right now

might not be a risk worth taking. Also,

I will share just on the bearish side

that Meta's AI strategy makes no sense

to me whatsoever and while I'm not short

Meta currently, definitely consider me a

bear on the stock. I'm also aware that

this view is slowly becoming consensus,

so it could be wrong. Speaking of bears,

I just interviewed the most outspoken

skeptic about AI and the data center

capex buildout, Ed Zitron. That

interview will go live on Sunday, June

21st, so stay tuned for that. He has

data on the operating losses of at least

one model company, which in 2025 was

absolutely staggering. You're not going

to want to miss out on that. More

generally on monetary matters, Max and I

plan on having bulls and bears to talk

about AI, not just on the large language

model companies, but also the

hyperscalers, the neo clouds and of

course the semiconductors. I hope that

whatever your view is, the viewer, you

can find value and information that is

accurate. While I have a lower degree of

confidence in how long this boom

continues, what I have high confidence

is is that the US economy and in

particular the US stock market is

increasingly becoming a concentrated bet

on whether AI is going to work.

Tremendous will be the rewards if the

bet pays off, as will be the losses if

it doesn't pay off. Please subscribe to

the Monetary Matters YouTube channel,

leave a rating and review for Monetary

Matters on Apple Podcasts and Spotify.

Check out Max's podcast Other People's

Money and also don't forget to check out

Monitoring the Situation, a live stream

that happens every day where Max and I

host from 4:00 p.m. to 5:00 p.m. Eastern

with tremendous guests.

Until next time.

>> Thank you. Just close the door.