HomeVideos

Altman says the world should accept 'some bad things happening' with AI, plus China and earnings

Now Playing

Altman says the world should accept 'some bad things happening' with AI, plus China and earnings

Transcript

1702 segments

0:01

Hey, welcome to Market Hang. I'm Dan D.

0:03

Franchesco. I'm joined with Lauren,

0:05

Luke, Shai. We're here to chat about a

0:08

bunch of different things, but um happy

0:09

Monday, first of all, if if there's such

0:11

a thing as a happy Monday. I don't know

0:12

about you guys. I'm a little bit like

0:13

Garfield. Not a big fan of them. Uh but

0:16

let's jump right into uh AI safety,

0:18

right? We have to wait a couple months

0:20

for Avengers Doomsday, but the Avengers

0:23

taking on the AI Doomsday is already

0:25

here. Uh the White House announced that

0:26

there's an AISAR. um a lot on their

0:29

plate from you know concerns about AI

0:33

safety to potential doomsday but I want

0:35

to start with an even broader question

0:37

happy to open it up to the group is this

0:39

even a solvable problem is this a

0:41

possible test is this a fool's errand

0:43

because this tech move so quickly the

0:46

people who are building it don't seem to

0:47

know how to unpack how to you know build

0:49

safety guard rails how is the US

0:51

government going to figure it out

0:53

>> I mean I think if it was a solvable

0:54

problem the kind of current people who

0:56

are being put in charge of it might not

0:58

be the best place to do so. Like uh when

1:00

I when I think of people who would be

1:01

very well placed to solve this,

1:03

unfortunately, they're probably the same

1:04

ones who are either developing the

1:06

technology or screaming about the fact

1:08

that you know there's a 10% chance it's

1:09

going to kill us all. So I I think that

1:12

uh you know is a is a bit of a a hangup

1:14

here in terms of getting to a solution.

1:15

I remember last time I was here at Yahoo

1:18

was Caleb Silver who said you know we

1:19

didn't really put seat belts in cars

1:21

until there were deaths. We didn't get

1:22

you know drunk driving laws until there

1:24

were deaths. So it probably means the

1:26

best deterrent after a certain point of

1:28

time is, hey, here's how badly we're

1:30

going to punish you if things go wrong.

1:31

But I think, you know, that would need

1:33

the things going wrong part to happen

1:35

first,

1:35

>> right? Lauren, what do you think?

1:37

>> I think technology always moves way

1:38

faster than government. And I think Jay

1:40

coming in and making this binary rule of

1:44

120 days. What happens after the 120

1:47

days? The 120 days, that doesn't mean

1:49

that anything will be enforced. That

1:50

doesn't mean that regulation will be an

1:53

oversight. So really what what is the

1:56

120 days and do we need regulation over

1:59

AI to a capacity but we also know that

2:01

AI always just moves a lot faster than

2:04

what government could ever do.

2:05

>> As a journalist key to setting deadlines

2:07

is make them a little bit loose and

2:09

ambiguous so that if you don't hit them

2:10

you know it's fine. Sh I want to bring

2:12

you in. What's your what's your take on

2:13

the AISR and and the task ahead?

2:16

>> Listen the AI principle stepped in.

2:18

Nvidia, Nvidia is a perfect example of a

2:21

company trying to turn the problem into

2:23

uh an actual infrastructure opportunity

2:24

and you saw that last week. OpenShell is

2:27

going to create this kind of software

2:28

boundary around the agent while Sentry

2:30

is going to run on Bluefield 4. But

2:32

regardless, you're seeing that all

2:34

right, Jensen knows this is going to

2:36

become an issue and rather than just

2:38

waiting for someone else to create these

2:39

guard rails, we're going to create the

2:41

opportunity for everyone else. And

2:42

that's why you're seeing cyber security

2:43

companies just continue to go up to the

2:45

right. But yeah, like you guys

2:47

mentioned, I mean, we have to make sure

2:49

the narrative doesn't stick on slowing

2:51

the AI buildout, just making sure it

2:53

becomes more secure and letting the

2:55

progress continue. I think that Jetson

2:57

stepped in at the right time.

2:59

>> Oh, yeah.

3:00

>> Right. Like there's so far from from

3:02

what I've seen, this is like the first

3:04

uh real steps or things we're taking

3:05

towards in terms of trying to regulate

3:07

the the outputs. So far, like it seems

3:09

everything's been controlling the

3:11

inputs, people not liking data centers

3:13

in their community and so forth. So this

3:15

seems like it's a really kind of nent

3:16

tiptoeing into an area that's so far

3:19

kind of been wild wild west which means

3:20

you know we're probably going to get

3:22

things wrong but also means I think uh

3:25

mostly what we've seen from the

3:26

president is there's absolutely no

3:27

desire to fall behind. If you had to

3:29

kind of pick one single message from the

3:32

president and not I think would be you

3:33

know leadership and anything that

3:35

threatens that is bad. Well, also I

3:37

think that right now for AI like all the

3:41

AI issues have crossed the boundary of

3:43

like that shouldn't have happened.

3:44

They've been oopsies. They haven't been

3:46

intentional. So I think that's where

3:48

like you have to make sure you have some

3:50

guard rails before it becomes real

3:53

external threats where someone's

3:54

actually trying to attack these systems

3:56

and these autonomous systems are getting

3:58

more advanced. Like I think it's really

4:00

important right now to do it before uh

4:03

whatever financial health care system

4:05

all of a sudden gets violated and now we

4:07

have to go back five or 10 years and who

4:09

knows who was behind that attack aka

4:12

>> some other country or

4:13

>> I love AI oopsies. I love it. I can just

4:16

picture the open AI blog. We had an

4:18

oopsie. Sorry. But I think that gets to

4:20

a bigger question right about this like

4:22

what level of comfort do we have

4:25

sacrificing some safety in for the sake

4:27

of of innovation? Luke, you brought up

4:29

like the the the um seats, you know,

4:32

seat belts with cars. Um you know, and

4:34

I'll bring in Sam Alman here. We could

4:35

talk a lot about him. He had a big

4:36

Vanity Fair piece, um today, but kind of

4:39

one of the many things that he said

4:40

recently, um you know, is you need to

4:43

accept some bad things happening to reap

4:46

the benefits. So, it's the old like you

4:48

got to crack a couple eggs to make an

4:50

omelette. I I I do agree with that, but

4:52

at the same time, if you know, way back

4:55

when when they were building the Model

4:56

T, if Henry Ford was like, "Listen, some

4:58

of our cars are just going to blow up.

4:59

It's, you know, it's gonna happen." I

5:00

think a lot less people will be in cars.

5:02

So, I guess, you know, Lauren, where

5:04

where do you see the balance as far as

5:06

an acceptance of taking some amount of

5:08

risk versus not kind of throwing the

5:10

baby out with the bath water, so to

5:12

speak?

5:12

>> I mean, I think the question is what is

5:14

considered bad,

5:16

>> you know, and and how do we go from

5:18

there? Because is it bad?

5:21

Is it bad for the consumer or is it bad

5:24

or or is it good for the investor

5:27

>> but but the consumer is the one that

5:29

reaps the the terrible not even terrible

5:32

benefits but gets the the bad things

5:35

happening to them,

5:36

>> right? Well, is it bad if if is it bad

5:38

if my like my address my data gets

5:40

exposed from someone who's doing a

5:42

search on me or something like that?

5:43

Like I I would consider that bad. If an

5:45

agent did that, I would say, "Hey, we

5:47

should probably punish that agent or I

5:49

should get some kind of remuneration

5:51

from that agent." So, like, yeah, like I

5:53

I see the scope already for things that

5:55

have been, you know, bad and yet still

5:58

the underlying technology might be, you

6:00

know, very useful from a consumption

6:01

point of view. You know, Muse is the one

6:03

who, uh, you know, let me know when my

6:06

emails comes in of the of the topics

6:07

even for for what we're discussing

6:09

today. So, you know, kind of always

6:10

useful having the the personal

6:11

assistant,

6:12

>> right? Yeah. SH, I don't know if you

6:14

have thoughts on like where what's your

6:15

comfort level as far as taking a risk

6:17

taker.

6:19

>> Listen, I think AI AI has such a PR

6:22

issue and the unfortunate part is that

6:23

the two speakers are children. Like Sam

6:27

and Dario are just children. They should

6:28

not be like the figures of this once in

6:31

a generation movement. But

6:32

unfortunately, it is. So I think that

6:34

for Sam's comment for example

6:36

specifically like he was just talking

6:38

about zerorisk AI versus manage risk AI

6:40

but the way he presented it was just so

6:42

poor and the standard is just like not

6:45

he was like people are going to use AI

6:46

for scams. Nobody's going to ever hack

6:48

something with it. Of course they're

6:49

going to. So I think that there is going

6:52

to be case studies but it's going to be

6:54

the minority stake. Like it's not going

6:56

to be the bulk of them like fire. like

6:58

are we not going to create fire just

6:59

because someone might uh launch

7:01

something that's going to cause a

7:02

catastrophic boom? No. You just have to

7:04

contain it and make sure there's things

7:06

in place in order to make sure it

7:08

doesn't become catastrophic. So, I want

7:11

to um to the regulatory aspect of it and

7:14

and how we kind of manage the risk. I'm

7:16

going to do a little uh jump in the in

7:18

the time machine. Go back to 2010 and

7:21

I'll make this story quick, but there's

7:22

this thing called the flash crash. Some

7:23

of you might be familiar with it, might

7:25

have remembered it. market kind of drops

7:26

and then comes back in a couple seconds.

7:28

To make a long story short, people

7:30

freaked out. They were like, "Oh, what's

7:31

going on with this algorithmic training

7:33

and a lot of regulation, a lot of things

7:34

happen. Michael Lewis wrote a book about

7:35

it." But specifically, this thing came

7:37

called regulation at came out by the

7:39

CFTC. And part of it was we want all the

7:42

source code for all the algorithmic

7:43

traders and we want to be able to hold

7:45

it so we can understand who are these

7:46

people playing in these markets. Now,

7:48

the reason I bring this up is because

7:49

people had big issue with it at the

7:51

time, right? This is proprietary stuff

7:53

that these firms were making a lot of

7:55

money on. How this relates to AI is if

7:57

there's some type of regulation and the

7:59

regulation becomes well we need to

8:01

really look inside your models. We need

8:02

to understand these black boxes. I

8:04

wonder how comfortable especially some

8:06

of these people that have been screaming

8:08

to say we need more regulation are going

8:09

to be to hand over the keys to what are

8:11

according to their investors trillion

8:13

dollar valuation type companies. I I

8:15

that's where I think you know what I

8:16

brought up in the beginning. I think

8:17

there's a real issue as far as being

8:19

able to actually regulate this stuff

8:21

because ultimately you're going to have

8:23

to give the keys to the castle to people

8:25

that are the castle's worth trillions

8:27

and trillions of dollars. But I don't

8:28

maybe I'm not looking at it right. What

8:29

do you think?

8:30

>> Isn't isn't this kind of what they're

8:31

asking for though? Like isn't this what

8:32

isn't this the point of Dario's pacing

8:34

the frontier like please please invite

8:36

someone in make it someone you know

8:37

independent but probably third length

8:39

kind of at least government sponsored in

8:42

some way or who's looking for government

8:43

backing of this to to come in and

8:45

basically say like give us give us the

8:47

all clear serve as the red team

8:48

effectively in terms of you know testing

8:50

any new models we might want to come up

8:52

with poke holes in what we've done be

8:54

embedded it seems like you know to a

8:56

large extent this is what they're asking

8:58

for whether or not they'll they'll get

9:00

it from Jay Clayton another remains to

9:02

be seen,

9:02

>> right? Yeah. I don't know, Shai or

9:04

Lauren, you have your thoughts on on how

9:06

to kind of manage this risk or how to

9:08

regulate this.

9:10

>> Well, I I think uh Jensen's doing it for

9:12

us, acquiring hugging face, pushing on

9:15

open source. Like again, Sam and Dario,

9:18

I don't want to pick on them, but like

9:19

they're doing this out of defense. Like

9:21

if the open source narrative gains

9:22

momentum, like they have a duopoly right

9:24

now in a frontier model that has no

9:27

ceiling. So I think that right now it's

9:30

really important to put more focus on

9:31

technical standards that make these

9:33

companies actually implement these type

9:35

of um AI issues because if you can

9:38

create rules around agent permissions,

9:40

cyber security testing, external

9:41

enforcement layers for example like

9:43

doesn't matter about giving up the juice

9:45

that anthropic open AI has all that

9:48

matters is we cannot slow down American

9:50

companies from moving quickly on an AI

9:52

race. That's the number one northstar

9:54

across the board. But Dario and Sam,

9:56

they're they're doing a road show right

9:57

now for their IPO. So they are saying

9:59

some more self-fulfilling uh prophecies,

10:02

but Jensen stepping in again uh and

10:04

making sure you can see the smoke

10:06

between the fire.

10:07

>> So is your your pitch is that open

10:09

weights the future, open weights the way

10:10

to kind of avoid a lot of these these

10:12

issues. Basically,

10:13

>> it will be because we're going to see

10:15

the weight and parameters of these

10:16

models. There's going to essentially you

10:17

can customize it per enterprise use

10:19

case. You're going to see all the

10:21

nitty-gritty of it. And that's how every

10:23

revolution has gone by anyway like

10:25

software specifically cloud like

10:27

everyone starts with open source then

10:28

creates these managed source

10:29

applications that's benefit for your own

10:31

company. But right now there's the

10:33

exponential rate of AI anthropic and

10:36

open AI have gone from 0 billion of

10:39

revenue to hundred billion in a couple

10:41

years. That's never happened in

10:42

existence before in any revolution. So I

10:44

think that right now things are

10:46

happening too quickly and there is

10:48

somewhat of a prisoners dilemma that's

10:50

happening that they aren't caring as

10:52

much as they should as the

10:55

simplisticity of what should be good for

10:57

AI or not and I do believe that the open

11:00

source is a solution and you're seeing

11:02

the progress already this year. I do

11:04

wonder though with on the on the openw

11:06

weightight side the business model is a

11:07

little bit harder especially with the

11:09

valuations that a lot of these companies

11:10

are at when you're essentially opening

11:12

up the distribution right and you're not

11:14

necessarily gaining that revenue capture

11:16

that you would from a proprietary model

11:17

how does that work at the same time when

11:19

these companies are saying yeah we want

11:20

to go to market with a $2 trillion

11:22

valuation

11:23

token usage it's all tokens usage you're

11:26

going to start seeing you these comp

11:28

you're going to start seeing software

11:29

companies for example start being less

11:31

SAS and more utility the more they use

11:33

them, more that you see agents running

11:35

continuously, these the inference

11:38

consumptions are going to become outcome

11:40

based and then you're going to see all

11:41

these companies start charging based on

11:43

outcomes. And guess what? If you're a

11:45

client, you're going to love that

11:46

because you're only paying for what

11:48

you're actually using at the end of the

11:50

day. And it's going to just be a cut of

11:51

your whatever money you're saving on top

11:54

of whatever you're doing on your capex

11:56

and opex. So, I think that

11:57

>> yeah, it's going to be it's a different

11:59

lens. We all know yet things are

12:00

happening quickly. Luke, you I know you

12:02

mentioned talking about Muse and I don't

12:03

if you want to talk about token usage or

12:05

how you're kind of seeing value from the

12:07

AI usage you're having.

12:08

>> Well, I I think more to to Shay's point

12:10

like a lot of the things we've seen in

12:12

Agentic Finance at the enterprise level

12:14

like that's been at the sandbox or

12:16

testing point up until this time and I

12:18

think Q3 in the back half of this year

12:20

is is really when you're going to see

12:22

more of that actually accelerate in

12:24

practice. So I think kind of like the

12:26

next nine months are the kind of the

12:28

rubber hits the road moment. So far in

12:30

the AI boom, the the profitability

12:32

question has mainly been driven by

12:34

effectively capex beneficiaries and and

12:36

not having to realize much of the

12:37

depreciation yet. Uh whereas now I I

12:40

think we're in a bit of the phase where

12:42

you should see to the extent that agents

12:44

do improve ROI, you should start to see

12:45

that more and more and more in the

12:47

numbers. And you know, in terms of my

12:49

personal AIUS, it's a great um it's a

12:51

great fight me bro. It's a great, you

12:52

know, you you put in whatever you think

12:54

just um sometimes in the workplace it's

12:57

it's difficult to even engineer

12:59

constructive criticism back and forth.

13:01

Uh you can you can make this thing fight

13:02

with you and tell you why the thing

13:03

you've just written is absolute nonsense

13:05

and and pick apart and pick holes in

13:07

your argument. So I uh I I don't know,

13:08

maybe I'm a little off that way, but I

13:10

love having someone do that for me.

13:11

>> Yeah, Lauren, I want to um utter a very

13:14

scary phrase, but I promise it's not

13:16

that scary, which is I'm going to pull

13:17

up an old tweet of yours. Um, but it's

13:19

from a few months ago or a couple month,

13:21

I think last month where you said, "I

13:22

still believe we need to utilize AI

13:24

more." Um, you know, since then we've

13:27

had a lot of these agents come out, you

13:29

know, Muse and and Instinct and whatnot.

13:31

I'm just curious how maybe even that

13:33

short time, have you found utilizing AI

13:35

more? Do you still think there's room to

13:37

run? What's your what's your

13:38

perspective?

13:38

>> I mean, AI isn't going anywhere. I I do

13:41

think we should absolutely lean into AI.

13:44

But what I think the distinction is is

13:46

understanding that AI doesn't replace

13:48

humans. It doesn't replace risk. We

13:50

humans are the ones that put risk and

13:52

judgment into place. We should not just

13:55

be relying on a third party anything uh

13:58

to to make those judgment calls. So I

14:01

think as long as we understand that we

14:03

are the brains, we are the controllers,

14:05

we get to set the data points, we get

14:07

to, you know, strategically make the

14:09

next moves and not rely strictly on AI,

14:12

I think we'll be fine. I think the

14:14

problem is is that there are so many

14:16

people that think AI can make human

14:19

decisions. And I I don't know why people

14:23

feel so safe and comfortable doing that.

14:25

>> Yeah. And I also worry too about just

14:26

the deterioration of skills that come

14:28

from that, right? Like if you're going

14:30

to continue to outsource things to AI,

14:32

sure there's certain things maybe you

14:33

don't want to deal with, but like

14:34

ultimately you're going to lose the

14:36

ability to do that or it's going to

14:37

deteriorate. And that to me is is a big

14:41

issue, right? This like AI atrophy

14:43

that's kind of happening across the

14:44

board. Um I don't know does anyone here

14:46

like are there certain things that

14:48

you're like absolutely not that I'm I'm

14:49

keeping AI away from or or you guys all

14:52

in like I'll I'll throw it on whatever I

14:54

can. I mean writing I have to like I I

14:56

have to rate I have to rate myself right

14:58

or else uh you know what are what are

14:59

they paying me for. So that's a that's a

15:01

big thing I got to keep to me but uh you

15:03

know that's one place where there'll be

15:05

a hard hard fast wall. I I don't know

15:07

about you any place.

15:09

>> Um I mean not for me person I guess what

15:14

immediately comes to mind for me is

15:16

education. I think that there is a lot

15:18

of utilization in education and we're

15:20

seeing the impacts of it when it comes

15:22

to this younger generation and how they

15:24

are not as uh financially or um

15:30

as smart as the older generation when it

15:32

comes to their reading and and writing

15:34

skill set levels. I think there has to

15:37

be a balance and I'm education probably

15:40

was not the answer that you probably

15:41

thought I was going to come up with but

15:42

I just I'm hearing it more and more and

15:44

I think people are just relying too much

15:46

on the technology. Yeah, I think to your

15:48

like underlying point though, right?

15:49

Like most of our jobs if you had to like

15:51

distill them into basic functions, it's

15:53

either sales or quality assurance or

15:55

some kind of mix thereof. But like

15:56

there's a base of skills upon which

15:58

those rest and are built. And that's

16:00

that's probably where we are like

16:01

potentially seeing some more atrophy cuz

16:03

like hey, you know, at one point we were

16:04

all, you know, we were first we were you

16:06

welding parts, then the machine is doing

16:08

it, then we're just watching the widgets

16:09

come on the assembly line. And I feel

16:11

like we're at kind of the watching the

16:13

the knowledge widgets come off the

16:14

assembly line phase of AI. And it's, you

16:16

know, up to us to kind of hang on to to

16:18

what we have,

16:19

>> right? Yeah. So, I think um and and Shai

16:22

maybe interested to hear your

16:23

perspective on this. There was some

16:25

there was a report recently about deep

16:26

sea kind of closing the gap, right? And

16:27

and you talked a lot about the

16:29

importance for us to kind of maintain

16:31

pole position. Um h what's what's your

16:34

sense about how quickly that is closing

16:36

and whether you know us in the US need

16:38

to be a little bit nervous about them

16:39

catching up?

16:42

I actually fully believe they're closing

16:44

in on um the model part of the AI race,

16:47

but again I think models are probably

16:50

the easiest parts of the stack for

16:51

competitors to compress as research

16:53

spreads globally. Like I don't think

16:54

that's kind of really a differentiation

16:56

on the US dominance for us like we have

16:59

a much better durable advantage

17:02

underneath those models. When I say that

17:03

I mean compute, networking, memory,

17:05

cooling, power like all everything

17:06

that's required to deploy intelligence

17:08

at massive scale is a dramatically

17:10

harder problem than closing a benchmark

17:13

gap on some model on the latest and

17:15

greatest. So I think from that point

17:17

point of view, we are so far along and

17:19

China will not close the model that gap.

17:21

But on the model front, yeah,

17:23

absolutely. But go for it. It's a commod

17:25

that's the most that's the plumbing of

17:27

the entire stack is the model front. And

17:29

if they're uh closing the gap on that,

17:31

great. we have everything else.

17:34

>> I Well, I guess I I definitely

17:35

understand what you're saying about how

17:36

it's completely been commoditized now

17:38

and and the the value isn't necessarily

17:40

in there, but I imagine if we we don't

17:42

want to completely punt on it, right?

17:45

>> No, not at all. But I also think

17:47

Qualcomm like the Qualcomm example is a

17:48

perfect one for like you it's a great

17:50

reminder that tech leadership is isn't

17:52

permanent. So I think that we we kind of

17:55

we used to think about China as the

17:57

country trying to work around American

17:58

restrictions while here you have

18:00

Qualcomm essentially licensing IP tied

18:04

to a Chinese architecture that was

18:05

itself developed partly as a response to

18:08

manufacturing constraints. Like I think

18:10

there is a world where things are

18:11

happening so quickly that you just can't

18:13

like sit back and like know that this is

18:15

going to be a permanent thing. But we

18:17

have I sound like the biggest Jensen

18:20

fanboy, but like Nvidia just is helping

18:22

us have that gap with China. We also

18:25

have Broadcom like Brocom's huge for us

18:27

on AI networking. Micron's going to be a

18:30

much better bigger beneficiary on our

18:32

memory front, but the there's SK Hunx

18:33

obviously on that part of the world, but

18:35

and also we're not even talking about

18:36

power. Uh we have Vertive G, Verova,

18:38

like Constellation, like we have so many

18:40

US companies are going to help us

18:42

maintain this lead, but it's something a

18:44

mantra for sure.

18:45

>> I love it. All roads for you lead back

18:47

to Jensen and Nvidia. No matter what

18:49

it's Well, let me All right, let me I'll

18:51

give a little bit of the counter take.

18:52

The one thing that I found interesting

18:54

about Nvidia the past, you know, few

18:56

months is that it's definitely gone

18:57

from, okay, we're just we're not pock

18:59

committed in this, you know, gold mine

19:01

race. We're just selling the picks and

19:02

shovels to now it's okay, we're going to

19:05

own a hotel by the mine and okay, we're

19:07

going to transport the miners to the

19:08

hotel and okay, we'll maybe manage the

19:10

outside of the mine. Like, it's it's

19:12

getting more and more to just we're just

19:13

selling picks and shovels now. like

19:15

we're really banking on this whole thing

19:16

whether it's through the financing or

19:18

the funding or the investments. Does

19:20

that make you more you you I guess

19:22

ultimately view that as that's still

19:24

more bullish because they're becoming a

19:25

bigger part of the ecosystem. But I

19:26

don't know why why is that a good thing

19:28

that they've evolved from the pure play

19:29

of picks and shovels to now kind of a

19:31

little bit of everything.

19:33

>> They had to I think there's so many

19:35

competitors now all of a sudden like a

19:37

couple years ago was probably a handful.

19:39

Now there's dozens. There's going to be

19:40

hundreds. Like uh I think that for them

19:43

they had to kind of start selling the

19:44

narrative of full stack systems like the

19:47

most efficient way of getting the most

19:48

out of what you need for AI. And in

19:50

order to do that do that they had to

19:52

sell everything around just the actual

19:54

chip. Unfortunately the market is seeing

19:57

through it a bit and now they're like

19:59

we're not going to give you more than a

20:00

20 times earnings multiple which is the

20:02

market average until we see the bare

20:06

argument is wrong. And unfortunately

20:07

that takes time and time and doesn't

20:10

matter if aentic AI is taking off this

20:11

year, physical AI is taking off next

20:13

year and Nvidia is essentially going to

20:14

be picks and shovels of all these

20:16

graduating themes of AI.

20:18

>> Sure.

20:18

>> But it's but I think it's a little bit

20:20

more of just building around. It's also

20:21

like keeping it up, right? I mean this

20:24

point has been made obviously time and

20:25

again, but the circular financing I mean

20:27

it got to make you a little bit uneasy,

20:28

right?

20:30

>> Well, well the question is as a

20:32

shareholder, what do you want them to do

20:33

with all this cash then? Because they

20:34

can't acquire. They're going to get

20:36

blocked on any acquisitions they make.

20:37

So, in order for like them to have this

20:39

much cash, which they're producing a

20:41

stupid amount of cash every single

20:42

>> fair. No, they got a lot on the balance

20:43

sheet. Yeah. I mean, Luke, what do you

20:44

think?

20:45

>> Well, they they just did boost the

20:46

buyback authorization by what, you know,

20:49

135 billion. So, you know, jack that up

20:52

to 225 probably nobody's uh complaining

20:54

and you take the uh effectively the

20:56

circular financing concerns out of the

20:58

window. But no, I I think to SH's point,

21:00

one thing that's not only changed is

21:01

just the competitive environment for

21:02

Nvidia, but also just the financial

21:04

environment in general. If you're a a

21:07

borrower right now, uh given how both

21:09

just interest rates risk-f free have

21:11

gone up and also spreads have widened,

21:13

especially hyperscaler spreads, it's,

21:15

you know, it's getting a little tougher

21:16

to kind of clear hurdles. I think, you

21:19

know, nothing, you know, nothing crazy

21:20

when we're talking about some of the

21:22

revenue growth that a lot of these

21:23

companies are seeing. But uh to your

21:25

point and Dan, I I think you know for

21:27

the speed to continue in topline growth

21:30

across the industry, it does almost

21:32

require Nvidia to open up its its loving

21:35

arms, embrace everyone in this credit

21:37

wrapping and say, "Hey, actually these

21:40

risky companies, they are no longer as

21:42

risky as you think they are because

21:44

they're using our products and those

21:45

products have residual value." So, you

21:47

know, again, where do profits come from

21:50

in this? They're coming from, you know,

21:51

very large government deficits and very

21:53

large uh borrowing binges and capex from

21:56

the hyperscalers. Those are two very

21:57

very like safe blocks upon which to

22:00

build. But then you get into that kind

22:02

of Minskian realm of the safer something

22:04

looks. Every crash, every financial

22:06

crisis we have ever had, it's been based

22:08

on a safe product that looks safe for a

22:10

long time and then suddenly doesn't.

22:12

>> Mhm. Sure. I don't know, Lauren, do you

22:13

want to weigh in at all on Nvidia?

22:15

>> No, I think they they they've said it

22:16

all.

22:17

>> Perfect. Uh let's uh we can pivot here

22:20

for a little bit. Um we are now less

22:22

than a month away from midterms. Um as

22:24

crazy as that sounds. Uh so you know

22:28

affordability is a big thing on the uh

22:30

on the agenda. We wrote a big story

22:31

about how in all the swing states a big

22:34

focus has been affordability. From the

22:36

investors perspective, right? What what

22:38

do we think as far as trying to game

22:41

plan this? It seems like a lot of the

22:43

polling I'm seeing maybe split, maybe a

22:46

complete blue wave. It seems like a, you

22:48

know, the the red wall, so to speak, is

22:49

is a little bit of a um a pipe dream at

22:52

this point. Um, you know, maybe, uh,

22:54

Shai, we'll start with you on the

22:56

investing side. Is this something that

22:58

has already been priced into the

22:59

markets? Is this something that you need

23:01

to wait? We have so much more data than

23:02

we did maybe a decade ago as far as the

23:04

polling and the prediction markets. So,

23:05

it seems like we have a better sense of

23:06

where these races are going. What's how

23:09

is the market already or not yet pricing

23:11

out what could come in uh in November?

23:15

I think there's so much chaos that's

23:17

happening right now on the macro front

23:18

that it's not playing as much as I

23:20

thought it would. I think in back in

23:22

March uh future we dropped a note that

23:24

said that enjoy the summer because the

23:26

fall is going to be brutal with the

23:28

midterms essentially because what's the

23:29

lever that every politician is going to

23:31

pull uh heading into the midterms. AI is

23:34

too progressing too quickly. We need to

23:36

slow it down. Data centers, we need to

23:38

scrutinize it. That's the easiest lever

23:40

being pulled and that's what happened.

23:41

But the market's not caring as much as I

23:44

thought it would. Maybe because we're in

23:45

war mode with Iran still. Either way, I

23:48

do think that because there's so many

23:50

different aspects of the macro fund,

23:51

like I don't think investors should try

23:52

to predict one party winning, then

23:54

reposition your entire portfolio around

23:56

that result because the best takeaway is

23:58

like you mentioned, affordability is

24:00

going to become a constraint on policy

24:02

regardless of who wins. AI is a PR

24:04

problem, especially with electricity.

24:06

That's going to be one of the sleeper

24:07

issues because AI data centers, they

24:09

need enormous amounts of incremental

24:10

power while households are reading the

24:13

headlines. They see this. They don't

24:14

want their electricity bills going up

24:16

because the hyperscaler decided to build

24:17

a 2 gawatt campus 20 m away that's like

24:20

near their high school, middle school,

24:21

groceries, whatever it might be. So, I

24:23

think that there's a lot of political

24:24

pressure that's not going to go going

24:26

away from just the midterms. So, I think

24:28

just take take that bump along the way.

24:31

Know that's going to happen. But guess

24:32

what? power scarce and there's a way of

24:35

playing this that if data centers are

24:37

going to become harder and harder to

24:38

build the power theme the these data

24:41

centers whoever can make them the most

24:42

efficient will be the winners.

24:43

>> Sure. Yeah. Lauren, we were just talking

24:45

about uh electricity before but I know

24:46

as a you know trader, you know, youngest

24:49

on the stock exchange floor, what's your

24:51

what's your take? I mean, you've kind of

24:52

seen these these things before. How do

24:55

you kind of see it all playing out a

24:56

month out?

24:57

>> Um I think it's still a little bit

24:59

unpredictable. I don't think anything's

25:01

a shoe in. But what I am seeing or not

25:04

seeing enough of or what I'm looking for

25:07

is the private markets. What is going on

25:09

in the private markets? Because I think

25:11

a lot of what the information that is

25:14

being baked into uh where the market is

25:18

sitting has to do, you know, with

25:20

quarterly earnings and future quarterly

25:22

earnings. And you know, they're

25:23

projected to be doing well, but really

25:25

the private markets really tell a

25:27

different story as to what is going on.

25:29

And I don't think we are seeing enough

25:31

about you know the fallout of you know

25:33

energy oil

25:37

consumer spending you know what what is

25:39

actually going to happen and while there

25:41

are a lot of headlines of oh you know

25:44

consumer spending is is great and you

25:46

know people are spending you know more

25:48

money than ever in September um all of

25:51

that is is is noise because we know that

25:54

they're using debt to do a lot of their

25:56

spending.

25:57

>> Credit card bills are going up. Yeah.

25:58

So, what what is the what is it going to

26:01

look like when we actually are facing

26:03

reality, not when people are getting

26:05

supported by by

26:06

>> debt? What specifically stands out to

26:08

you in private markets that feels like a

26:09

real like, oh man, this is this is a

26:11

red.

26:12

>> It hasn't stood out at all. It hasn't

26:13

the story hasn't unfolded. And so,

26:15

that's what I'm trying to see because

26:17

we're just hearing only about the stock

26:18

market. We're only hearing positive

26:20

headlines and really the private markets

26:22

I think always kind of share more of a

26:24

story and that hasn't been told yet.

26:26

>> Right. Luke, what's your take on this?

26:27

more difficult story to tell, right?

26:28

That's the the liquidity does does some

26:30

damage both ways or helps you sometimes.

26:33

So, um something I'm watching for in the

26:35

the run to midterms is, you know, I

26:37

think we can agree no matter what side

26:39

of the political spectrum you you're on,

26:41

uh the past, you know, 10 12 years,

26:43

let's call it, they've uh coincided with

26:46

some uh more severe partisan rifts in

26:50

the United States. Uh, one thing that,

26:52

you know, this tells me is that the odds

26:55

of, you know, passing meaningfully

26:57

meaningful legislation, uh, after the

27:00

midterms, they generally go down,

27:02

particularly when there's been a change

27:03

of power. And, you know, the only thing

27:05

that might stop that is, you know, a

27:07

crisis in the event of, you know, a CO

27:09

2.0, which, you know, God willing, uh,

27:11

not not going to happen. So, one thing

27:13

that I've noticed after the past couple

27:15

midterms is they've been pretty pretty

27:17

big uh bonds over stocks events, which

27:20

is kind of not something that I think is

27:22

in the cards uh for, you know, for

27:25

everyone right now, especially given how

27:27

battered bonds have been. And, you know,

27:29

looking at two samples, a sample size of

27:31

two there. So, you know, this time is

27:32

always different and this time it just

27:34

happens to coincide with one of the

27:36

largest capex impulses of all time

27:38

starting to rely on external financing

27:40

about 6 months uh 6 months before the

27:42

tea date for the midterms. So, but you

27:44

know that's kind of something I keep in

27:46

mind as you know fiscal stimulus

27:48

probably just the impacts of you know

27:50

bills that have been passed in the past

27:51

are are waning. So, you're going to have

27:53

less of an impulse there and kind of

27:54

less scope to see growth there going

27:56

forward. So, you know, in terms of this

27:58

rally being built on a couple legs,

28:00

that's one of them. And you know that

28:02

one probably a little less supportive

28:03

for stocks, maybe a little more

28:05

supportive bonds going forward.

28:06

>> Yeah, I mean we got the Treasury yields

28:07

right there behind you. Um it's been

28:09

such a wild ride, right? I mean I think

28:11

someone compared it it's trading like a

28:12

meme stock, which is crazy to say, but

28:15

it it kind of feels that way. Um yeah, I

28:18

mean there there's so much to unpack

28:19

between now and November. I mean the the

28:21

private markets aspect is so

28:22

interesting, too, because the other kind

28:24

of big um elephant in the room is this

28:26

anthropic IPO, right? That's reportedly

28:28

coming, but we don't know. It's maybe,

28:31

you know, I I think the last report I

28:32

said is they're eyeing for

28:34

pre-Thanksgiving.

28:36

Um, that just feels like such a big

28:38

thing because one, obviously, it's it's

28:40

a massive AI player coming to market.

28:42

So, does the IPO open? How do investors,

28:45

you know, uh, digest that? Also, is

28:47

there a little bit of a is it almost

28:49

serving as a dam? Like we saw the other

28:51

day, um, one of the IPOs kind of delayed

28:54

like, you know, you don't necessarily

28:55

want to move around such a big whale.

28:57

You want to kind of give them their

28:58

space. So, you know, to your point about

29:00

the private markets, I'm really

29:01

interested to see like when that does

29:03

come to market, what's the kind of

29:04

knock-on effect across a bunch of

29:05

different um lanes.

29:08

>> Yeah. I mean, I think um I mean, it

29:11

it'll be an interesting story to see

29:14

what unfolds. I mean, I the one of the

29:15

things that I love about when companies

29:19

go public, we really get to see their

29:22

financials. We really get to see the

29:24

heart of the company. And I think

29:26

Anthropic will tell a really big story

29:29

of what potentially is going on in other

29:31

AI companies and especially ones that

29:34

are not public and what that

29:37

conversation looks like outside of just

29:39

having conversations about circular

29:40

money and and how that is, you know, I

29:44

feel like the topic of conversation. So,

29:46

I don't know. I'm curious to see will

29:48

they push that deadline? I don't know.

29:50

We'll see. But, you know, we'll see.

29:52

>> One thing Yeah. One one thing we can

29:54

kind of peek at there is like what

29:55

happened in the runup to SpaceX, right?

29:57

That's been the the biggest IPO so far,

29:59

you know, ever. And and this year, of

30:01

course, so what the data I've seen was

30:04

ahead of that investors sold

30:07

SpaceX likes like like companies. I

30:09

don't know how many companies there are

30:10

like SpaceX, but you know, they sold

30:12

mega cap tech in the runup to that and

30:14

to make room effectively for that. And

30:15

they also sold the biggest losers year

30:18

to date. You know, that we weren't too

30:19

deep in the year by that point. So I

30:21

what I find it very interesting about

30:23

the timing is kind of how it coincides

30:25

or doesn't coincide with tax law tax law

30:28

selling season which we are now for

30:29

mutual funds in the midst of this is you

30:31

know October should be the heart of it.

30:33

So, you know, I'm kind of looking to see

30:35

kind is there any pressure on, you know,

30:38

anthropic adjacent names in public

30:40

markets as you're kind of selling to

30:41

make room for this if you anticipate

30:43

getting an allocation in uh early to mid

30:45

November. And do the like the likes of

30:47

Nike continue to get pummeled, you know,

30:49

a stock that's been down every quarter

30:51

of of this year and, you know, continues

30:52

to set I think lowest since 2013 now.

30:55

So, I think those are some interesting

30:56

things, but so far this year, I'd say

30:58

the main thing the market has a problem

31:00

digesting is not equity supply, it's

31:03

bond supply. And that's cuz year to

31:04

date, if you look at hyperscaler

31:06

issuance and IG markets, that's 150

31:08

billion give or take in the US. And if

31:11

you tally up the two biggest IPOs in the

31:13

AI space, Sarah Bruss and SpaceX, you're

31:16

running at a little over 90 billion. So,

31:18

we we really had to digest a lot more

31:20

debt. We actually have that equity.

31:22

>> A lot of debt. Shai, what's your what's

31:24

your take on kind of anthropic? Will

31:26

they, won't they, and the impact it's

31:27

having on everybody else?

31:29

>> Yeah, I mean, I think the SpaceX IPO is

31:31

a great case study. I will say it's a

31:33

little different because a lot of the

31:35

space proxy names, the Rock Lollabs, AS,

31:37

Space Mobile, Plan Labs, like you can

31:39

see their charts. They're all down 50%

31:41

since the uh SpaceX IPO Halo effect. But

31:44

the difference in that uh ecosystem than

31:48

the AI ecosystem is a lot of people sell

31:50

those companies because they want the

31:52

category leader in SpaceX. They want

31:54

same theme, same exposure, but SpaceX

31:56

does connectivity category leader,

31:58

launch category leader. They do

31:59

everything. Anthropic, they don't really

32:02

do everything in the AI ecosystem. They

32:04

just do the model components. So I do

32:06

think that there is going to be somewhat

32:08

of a heartburn, I guess, in the AI theme

32:11

around the IPO just because capital's

32:13

finite. If you want to participate in

32:14

Enthropic, which I think a lot of the

32:17

big books out there, institutions are

32:19

going to want to, it's gonna have to

32:21

come from somewhere else. However, I

32:22

don't believe it's going to be as

32:24

disruptive as a SpaceX IPO was for the

32:27

whole space theme. Also, uh I do think

32:30

that like we get see what's under the

32:32

hood now finally on their frontier

32:34

models. Like what's the real margins?

32:36

Are they going to create something from

32:37

thin air of like, oh, this is a adjusted

32:39

frontier blah blah blah KPI that we're

32:42

going to use or is it going to be

32:44

somewhat real and we can actually judge

32:45

if there's a real ROI now on this AI

32:48

spend. And for me, I'm going to geek out

32:49

on that. I'm sure a lot of people will

32:51

do that as well.

32:51

>> Are they going to use the wei workbook

32:52

of accounting where it's like, well, if

32:54

you take out this and this and this and

32:55

this cost, then actually we're super

32:57

profitable. Uh Lauren, you you what

32:59

what's your you you kind of seemed to

33:01

make a sound there when he was talking

33:02

about anthropic not being as as hyper

33:05

involved. Do you see it as a little bit

33:07

different than that or

33:08

>> I mean no I mean AI is it's part of the

33:10

infrastructure. So I think there are

33:12

other companies out there that are that

33:13

are absolutely in the same realm of

33:15

space as well. So I don't know I for me

33:17

I just really want to collectively like

33:20

see what's under the hood, see what

33:21

their financials are looking like and

33:23

kind of assess from there what is

33:25

actually going on within the AI space.

33:27

It's going to be some fun. I I imag

33:30

>> I also think like the what's not being

33:32

discussed, which is the elephant in the

33:34

room, is Anthropic's going to have a

33:35

white check uh like a blank check they

33:37

can use now once they go public. They're

33:38

going to dilute like no one else. And

33:41

what's that going to do for the whole

33:43

environment, it's going to it's going to

33:45

improve it. They're going to spend so

33:46

much money across the board, across the

33:47

whole ecosystem. So I think that

33:50

semiconductors are a really interesting

33:51

angle on capitalizing and anthropic IPO

33:54

because they're going to spend a ton on

33:56

the whole ecosystem. Same with the cloud

33:58

computing companies. So I just think

34:00

that there is going to be you can make

34:01

the circular financing argument which I

34:03

I I agree it's kind of like

34:05

uncomfortable because we're in the first

34:06

stage of this big buildout. But I do

34:09

think it's going to lift the market a

34:10

bit more than people think. I one thing

34:12

that really sticks with me and and I use

34:14

the SpaceX comparison is so you look at

34:16

SpaceX right from where it opened to now

34:18

it's basically up about 4% give or take

34:22

right not from its IPO price from where

34:24

it opened to the the public right and

34:26

that's a business obviously that you

34:27

said does a ton of stuff I mean there's

34:28

I think the majority of satellites in

34:30

space right now are are are from espec

34:33

essentially SpaceX um and then you look

34:35

at anthropic and sh to your point like

34:37

just those miles pretty straightforward

34:39

now they're looking I the reports I saw

34:42

was a $2 trillion market cap. I like you

34:47

know now maybe I can't fathom how big of

34:50

a company is going to be and maybe my p

34:51

human brain can't understand the impact

34:53

AI is going to have but to think as an

34:55

investor when SpaceX again has like a

34:58

bunch of thriving businesses or or much

35:00

more mature businesses and open at a

35:02

much lower valuation is still kind of

35:03

holding steady who I guess I'm not

35:06

asking you guys to say buy or sell but I

35:08

guess do we think there's going to be a

35:09

massive appetite to buy in at such a

35:11

high price point it just feels like you

35:13

know to your point about the private

35:14

markets the VCs and the the private

35:16

investors have sucked all the value out.

35:18

There's nothing there's there's not much

35:19

left on the bone. I don't know but maybe

35:20

I'm missing something.

35:21

>> I I can tell you this stylist fact and

35:23

shy pointed out that you know might be a

35:24

lot of institutional demand for

35:26

anthropic. I can tell you the single

35:28

biggest day on Robin Hood of net

35:30

purchases of stocks was the SpaceX IPO.

35:33

The single biggest day. So will there

35:35

will there be appetite from place? I can

35:37

you know based on recent history I'm

35:38

counting there there will be one in

35:40

particular. Yes.

35:41

>> All right. That's fair. Also, also like

35:43

don't don't discount the doomer exposure

35:45

you need in your portfolio. Like if if

35:48

the world does turn uh oopsie, anthropic

35:51

is probably going to be the biggest

35:52

beneficiary off that. Uh but I also I

35:56

don't know. I I really do believe that

35:58

if you invest in Anthropic at $2

35:59

trillion, you fully believe that they

36:01

are going to be the ones that conquer

36:02

super intelligence or RSI for example.

36:05

And I think that if you do invest, don't

36:08

believe the 2026 numbers cuz like like

36:10

Robin Hood for example, like they were

36:12

kind of in the experimental phase with

36:13

AI oric AI. I think a lot of enterprise

36:16

usage on anthropic was experimental.

36:18

They heard the FOMO. They read it. They

36:20

want to implement it, see what they

36:21

could do with it. Nobody knows the

36:23

stickiness yet to it. It's not like

36:24

agentic AI where the more you put into

36:27

it, the more it remembers you and then

36:28

you're going to stay there forever. Like

36:30

not really for for claude. can just jump

36:33

between the models by copying and

36:36

pasting the memory. So, I think that I'm

36:38

curious on what the retention is going

36:39

to be in those metrics going forward

36:41

more than the actual revenue.

36:43

>> Lauren, do you have any thoughts on uh

36:45

>> I think for me as an investor or I guess

36:48

for novice investors, I personally

36:51

never, you know, when it comes to IPOs

36:54

immediately and looking to buy the stock

36:56

cuz the reality is most of the time it's

36:59

going to go down anyway. So, let the

37:01

numbers come out. let there be a

37:02

celebration and then maybe in a month or

37:04

two months, you know, before the new

37:07

year, decide if you want to if you want

37:08

to pick that up and put that part of

37:10

your portfolio.

37:11

>> Yeah, there is definitely a a risk

37:13

though always in not owning particularly

37:16

when things get added to indexes like at

37:18

a certain degree. It's not just

37:20

believing in the story. It's just not

37:22

wanting to be super super underweight

37:24

anthropic that can actually force a

37:26

little bit of buying. So, you know, and

37:27

I you can buy a lot of anthropic and

37:29

still be a complete unbeliever and

37:31

underweight it from from a lot of

37:33

perspectives. So, you know, I I think

37:34

that's kind of another underdisussed

37:36

part of the story of where IPO demand

37:38

kind of has to come from at the end of

37:40

the day. I look like you

37:41

>> also Yeah. Also, how funny would it be

37:44

that if Dario is the reason we have no

37:46

Santa Rally this year because he went

37:47

public end of November, like he would

37:49

just be the permanent Grinch forever.

37:54

>> The the timing is really interesting. I

37:55

mean the other obviously is is the

37:57

counter right is open AI has said we're

37:59

going to wait till 2027. I I do wonder

38:01

like on the one hand you get the

38:03

hindsight of seeing and you know how the

38:05

public receives anthropic and where it

38:08

makes missteps and what its S1 says and

38:10

all those different things. The downside

38:12

though is you don't get like you're not

38:15

the first mover advantage. So if you

38:16

know people are invested and it starts

38:18

performing really well are they really

38:19

going to want to pull out their money

38:20

from anthropic and go into your IPO? I I

38:22

don't know if there's a a right answer

38:24

there as far as being the first or

38:25

letting someone else blaze the path and

38:27

then kind of coming in behind them.

38:29

>> Yeah. I I don't know how to at all deal

38:32

with that horse race. Just that idea

38:33

because I think there's there's a lot of

38:35

confounding variables. You could be, you

38:36

know, very happy you've stayed on the

38:38

sidelines. However, I think, you know,

38:40

revealed preference is a thing. And I

38:43

don't think Open AI, if you, you know,

38:45

gave everyone truth serum in Q1 of this

38:47

year, did they want IPO this year? Did

38:50

they want IPO before Anthropic? Yes,

38:52

completely. So, the fact that they're

38:53

not is, you know, quote unquote bad.

38:55

That's kind of as far as I can go in

38:57

shaping up. It certainly doesn't seem to

38:59

be according to plan. Obviously, there's

39:01

been uh a lot of good reporting on

39:03

potential seuite battles over the

39:05

appropriateness of a potential OpenAI

39:09

IPO this year. And, you know, obviously

39:10

the kind of more financially uh savvy in

39:13

the group have have come out ahead, it

39:15

seems.

39:16

>> Yeah, sure. Um, I want to talk a little

39:19

bit too about just AI budgets in general

39:21

because that's like the big the big

39:23

overarching theme of all this is they

39:24

can build all these models, they can

39:26

build all these tools, but if people

39:27

aren't comfortable with the budgets that

39:29

they have or aren't feeling that they're

39:30

getting the ROI on it. Um, interesting

39:32

story in the journal a couple days ago

39:34

about um the fact that cheaper models

39:37

are actually more expensive because they

39:40

take more time to figure things out. Uh,

39:43

which I think is an interesting dynamic.

39:45

kind of reminded me of the idea of like

39:46

you buy a cheap pair of shoes versus an

39:48

expensive pair of shoes and you end up

39:49

going through three pairs of shoes

39:50

before your expensive pair of shoes

39:52

wears down. But um feels like a CFO

39:55

nightmare. I don't know. I'm I'm glad

39:56

I'm not in in that role. Uh Lauren, I

39:58

don't know if you had thoughts on the

39:59

the kind of continuations trying to find

40:02

the ROI on on AI and managing these

40:04

budgets.

40:05

>> Yeah. Um I do, but I'm going to let them

40:08

speak first and then I'm going to I'm

40:10

going to jump in. Luke,

40:11

>> yeah,

40:11

>> I think uh the the CEOs that have done

40:14

the best job of, you know, detailing

40:16

this and I I think J might agree are the

40:18

the ones who have focused on both

40:20

outcome based pricing and a lot of it

40:22

comes in the enterprise space where I

40:24

think even at Salesforce like Benning

40:26

off had a bunch of compete like a bunch

40:28

of customers you trotted them out on

40:30

their on their earnings call be like

40:31

this is what we're using it for these

40:32

are the results we've seen so on and so

40:34

forth. So, you know, on on the one hand,

40:35

I think we've had some kind of bad

40:37

examples like Google Sundar Pikai kind

40:39

of stumbled through an answer on like

40:41

what is it that you say that you do here

40:43

type of stuff on the AI ROI. the the

40:45

ones a little a little more downstream

40:47

seem to have no problems detailing,

40:50

okay, here's at least what our best

40:52

customers, our biggest adopters are are

40:54

getting from this kind of one thing that

40:56

I think though is is interesting is

40:57

that, you know, I'm reminded of a piece

41:00

that Matt Zaitlin wrote in 2022 about

41:03

why US productivity was bad because it

41:05

was everyone's first day on the job

41:07

effectively. That was, you know, hiring

41:08

was very high. that was the, you know,

41:10

the ratio of unemployed to job openings

41:13

was, you know, at a at a record low,

41:15

that kind of thing. And it seems like we

41:17

have the same thing in corporate America

41:19

of, you know, one reason productivity,

41:21

measured productivity might not be going

41:22

gang busters is because it's everyone's

41:24

first day triing an AI program and we're

41:26

trying to figure out, you know, what the

41:27

hell to do with it and make us a little

41:28

better,

41:29

>> right? We do forget like, you know,

41:31

really I for me, I don't know, and Chai,

41:33

I'm interested to hear your thoughts. I

41:34

feel like things really turned end of

41:36

last year, beginning of this year where

41:37

it was like, okay, this is I mean, it

41:39

was always interesting and powerful, but

41:40

this is really, you know, certainly from

41:42

I'm not an engineer, but on the coding

41:44

side, it seems like a flip really

41:45

switched. Um, but what's your

41:47

perspective on like kind of the ROI

41:48

you're seeing companies get on some of

41:50

their AI spend?

41:52

>> Yeah, so at Future, like we talked to

41:54

the tons of CFOs and CEOs of the biggest

41:56

companies in the world, and the

41:57

sentiment has changed drastically over

41:59

the year. I think a lot of CFOs are

42:01

starting to care less about what a

42:03

million tokens costs like and much more

42:05

about like what it costs to complete the

42:07

actual work outcome based like it's

42:09

going outcomes is probably the biggest

42:13

wording that we've heard uh in the past

42:15

couple weeks and I think it's going to

42:16

continue in 2027. If one agent costs

42:18

twice as much but resolves 10 times more

42:20

tickets, they're going to be much more

42:22

willing to spend. And like you mentioned

42:23

earlier, like the trial and error aspect

42:26

of AI, it's becoming a lot more costly

42:27

than a lot of the CFOs thought it would.

42:29

And that's going to start being more

42:31

intentional now go in 2027. And that's

42:33

why you're going to see like the Service

42:34

Now, for example, that they're going to

42:36

be one of the big biggest beneficiaries

42:38

of the next layer because they're going

42:39

to start selling businesses around the

42:40

outcomes rather than software seats.

42:42

You're seeing Palanteer literally

42:44

they're getting a cut of how much money

42:46

they're actually the companies are

42:47

actually saving in their in their actual

42:50

contract. their CRO at Palanteer. He

42:52

used to be a lawyer. That's not on

42:54

that's not on accident. That's on

42:55

purpose. These contracts are meant to

42:59

create outcomebased results. And I think

43:01

you're seeing Palanteer just kill it in

43:02

the application front. You're going to

43:03

see everyone else doing that too because

43:05

they're the only ones that really

43:06

produce an ROI and enterprise AI spend.

43:09

And you're going to see everyone else

43:10

follow suit.

43:11

>> Really interesting development for uh

43:13

the consultants of the world, right? to

43:15

switch from you know billable hours and

43:16

all that to now you got to put your

43:18

money in your mouth is uh Lauren I want

43:19

to come back to you on you any what's

43:21

your high level thoughts there

43:22

>> um I I think the question or what I

43:26

would like to be answered is okay we

43:29

we've seen enough of of the spending and

43:31

the AI where where is it actually

43:35

where is their productivity you know

43:38

where is the ROI and I think we have to

43:40

stop asking the question of okay yes

43:42

spending is going to help and what is

43:45

the actual return? And I think we are

43:47

just not seeing it enough. And so I

43:49

think when it comes to corporations or

43:50

these next earning calls,

43:53

>> where where is the money going? Like I

43:55

actually want to see the flow of it and

43:57

does it actually make sense? Because if

43:59

these cheaper models are supposed to be

44:01

cheaper and they're not,

44:03

>> then what what questions are actually

44:05

being answered and and I guess we'll

44:07

have to find out what what they say from

44:09

there. Shai, in the conversations you

44:11

have with the the CEOs and CFOs, what

44:14

about the idea of like when to upgrade

44:16

or when to make a switch because so many

44:19

of these developments are happening so

44:20

quickly and these new models are being

44:22

released. You have to kind of be nimble

44:24

as far as like, okay, we're bought in

44:25

now with this and this is working, but a

44:27

month later it could be someone else,

44:28

right?

44:30

>> I can tell you this, this past summer,

44:32

everyone was overspending on AI. I think

44:34

that you saw a lot of budgets budgets

44:36

get kind of ballooned up a bit. Uh end

44:38

of the year obviously everyone that's

44:40

when you can really tighten down the

44:41

upcoming year's budget so you realize

44:43

what do you actually want to spend

44:44

that's fixed. I think that right now

44:46

you're you're going to see AI behave

44:48

much more like a utility. I mentioned

44:49

that earlier on in our show where every

44:51

time an agent re reasons through a task

44:53

and decides what's wants to do do it's

44:55

going to have to do it 10 times 100

44:56

times over again. It's going to consume

44:58

so many more tokens. So, ironically, the

45:00

better the product becomes, the more

45:02

employees actually use it, then the

45:04

harder the bill can become to forecast.

45:06

And I think right now there is somewhat

45:09

of um it used to be like twice a year we

45:12

put together a budget for capex or opex.

45:15

It's becoming a monthly conversation now

45:17

that we've we've started noticing. I

45:18

think that you're seeing a lot of open

45:21

AI momentum the past month and anthropic

45:24

momentum. I think there is a left brain

45:26

right brain which is like a game of hot

45:27

potato between its duopolies of models

45:29

but also people are counting out Gemini.

45:32

Gemini 4 is actually was actually pretty

45:34

darn good and I think that a lot of

45:36

people already are in the workspace of

45:38

Google that they're going to care a lot

45:40

more about spending because inference is

45:42

just beginning. They're going to care a

45:43

lot more about the economics of it. So I

45:45

I do believe that it's the spending is

45:49

going to be tamed. I think right now

45:50

just everyone's overspending so it looks

45:52

way worse. Yeah, Google's an interesting

45:54

one because it was kind of they they

45:56

definitely had a moment there for a

45:57

while where everyone was kind of talking

45:59

them up and very very excited about it

46:00

and then it kind of drifted back which

46:02

kind of gets to u you know our quarterly

46:05

earnings right that are coming up. Uh

46:07

they're starting to trickle in now and

46:08

then big banks report next week and and

46:10

then the big ones the big tech. Um

46:13

anything standing out to you? It feels

46:14

like to me my perspective is every

46:17

earnings there's kind of one scapegoat

46:19

that gets the brunt of like all the

46:22

anger or or of you know uh questions or

46:25

pessimism about AI and that one stock

46:27

gets dinged and then everybody else does

46:29

better. You know for a while it was meta

46:31

there because of what they were trying

46:32

to push and they didn't have the cloud

46:34

infrastructure then you know sometimes

46:35

it's Microsoft. I do you see a similar

46:37

thing? do you think because it it feels

46:39

like always one is the one that gets hit

46:40

and then everybody else kind of is able

46:42

to escape by and this kind of also

46:43

speaks to how topheavy the market is but

46:45

I don't know is there anything Luke

46:46

we'll start with you anything top of

46:47

mind that you're thinking you know

46:48

looking at coming out

46:49

>> well it it almost speaks to the

46:50

dispersion that's been so important to

46:52

keeping the market at a high level like

46:54

if you had told me 3 years ago that hey

46:56

like some of the biggest companies in

46:58

the world are all going to be kind of

46:59

pursuing the same goal spending on the

47:01

same things to continue their dominant

47:03

positions in their markets and they'll

47:06

trade with extremely low correlations to

47:08

one another. I'd go like, you know,

47:09

that's crazy. No, there's no way that

47:10

happens. This is what has happened for

47:12

the entirety of the AI boom. So, you

47:14

know, it's almost like the same way in

47:17

which when uh, you know, Claude was

47:19

putting out a press or Anthropic was

47:21

putting out a press release about a new

47:22

cloud capability every day in Q1 and

47:24

software stocks, we get killed or more

47:26

recently how Muse has invented the

47:28

category of consumer inertia stocks,

47:30

which I don't think actually really

47:32

exists at all. We have to find victims.

47:34

like when one company does well to your

47:36

point, we almost have to find a loser in

47:38

the AI theme because that's almost just

47:40

how money moves around on a short-term

47:42

basis. But like when it comes to the the

47:44

broader story of earning season, I'm

47:46

more looking at like we have, you know,

47:49

pretty good nominal growth economy right

47:50

now. You a lot of it, you know, is

47:52

juiced by AI capex, but still even, you

47:55

know, median S&P 500 revenue growth has

47:57

been pretty pretty great. And right now

48:00

the equal weight S&P is further from its

48:02

record high than equal weight Europe

48:04

with you know diesel prices skyhigh,

48:06

food prices very high, yields high. I

48:08

this is a very unique environment for me

48:10

as an investor to see like the US

48:12

consumer stocks really be the outlet of

48:15

investor angst rather than you know

48:18

Europe in this kind of backdrop. So I'm

48:19

just looking to see if if it's enough

48:21

for companies to say hey like things are

48:23

pretty much the same as they were last

48:24

quarter and those stocks to to kind of

48:27

recover and bounce back. That's kind of

48:28

what I'm more looking for to see what

48:30

investor appetite is to embrace the many

48:33

given that the economy is still holding

48:34

up pretty darn well.

48:35

>> Sure. Lauren, what's what's your take

48:36

heading into earning season?

48:38

>> Um, you know, I there there's just so

48:41

much. I mean, I I think to say that the

48:44

economy is doing decent is interesting

48:47

take. Um, only because I I just I don't

48:51

know if the numbers are really

48:54

supported, right? like a lot of these

48:55

earnings are baking in the tariff

48:57

refunds that happened earlier in the

48:59

spring and so I don't to me I don't

49:05

I don't really have forecast for the

49:06

rest of the year. I'm looking to see

49:08

what 2027 holds. I mean a little bit

49:10

about you know with the midterm

49:12

elections and things like that but I

49:13

just don't think these numbers in my

49:15

opinion are realistic.

49:17

>> What does any sector in specific stand

49:18

out to you as a a better bellweather to

49:20

look at because obviously big tech is

49:21

kind of its own little animal.

49:22

>> Yeah. um consumer discretionaries that

49:25

that's where I look healthare uh seeing

49:28

>> where that goes. I just I I just don't

49:32

>> Yeah, tech is in its own space and I I

49:34

really don't know what I'm looking at

49:37

2027. Maybe we'll have a refresh and

49:39

we'll get way more clean numbers and

49:41

we'll go from there.

49:41

>> Sure. Shai, what's your take?

49:44

>> Yeah, I just think that everyone knows

49:47

AI demand's there. Infrastructure is

49:49

booming. I think that the cost required

49:51

to satisfy that AI demand is just moving

49:53

so much higher. I think a lot of people

49:55

are just doesn't matter what revenue

49:57

growth you have like you have Google

49:59

cloud AWS just producing like video game

50:01

like numbers on this scale is just like

50:03

it's unheard of like if you told me this

50:05

two years ago be like these are $5

50:06

trillion companies that's not the case.

50:08

It's disappointing investors cuz why is

50:10

that? free cash flow is not following

50:12

because they're just investing so

50:13

aggressively to build these AI cities

50:15

like they know it's going to uh be the

50:19

center of all the economy for decades to

50:22

come. They want to build as much real

50:23

estate as possible right now to benefit.

50:25

I do think that you're seeing names like

50:26

Micron being the beneficiary side of

50:28

that equation. Same with like uh

50:30

networking. But I think that like for

50:34

Metam Muse for example like everyone's

50:35

going to be hyping up that stock heading

50:38

up to that earnings. Why is that? first

50:40

Chad GBT moment in three years.

50:42

>> Guess what? Their capex, you think their

50:45

capex was high before they had an

50:46

application that would satisfy the

50:48

market. It's gonna balloon. Like Zuck is

50:51

the one of the few CEOs that could care

50:54

less about what the public thinks cuz

50:56

guess what? You can't fire him. Like

50:58

he's going to do whatever he feels like

50:59

he needs to do. Capex will balloon.

51:01

We'll see if the the street is satisfied

51:04

enough with the metrics they say on Muse

51:06

to justify the the bump. But again, like

51:09

I just think that big tech is kind of

51:11

dead money a little and I think that's

51:13

why semiconductors are kind of the

51:14

better place to be. They've gone

51:16

penalized due to some macro environments

51:17

like the tenure yield for example is

51:19

kind of really hitting the semiconductor

51:21

space due to funding these data centers

51:23

becoming more expensive, but that's

51:24

still the beneficiary.

51:27

>> Yeah. Uh so Mark's always going to do

51:29

what Mark's going to do. You know,

51:30

that's that's how he rolls. Uh we'll end

51:32

here on a fun one which is we had an

51:34

interesting story about this new concept

51:36

called loud working. Uh the idea that

51:39

job market as the most recent jobs port

51:42

still a little shaky still a little

51:43

uncertain. People are very uneasy. So

51:45

when you're at work you got to be loud.

51:47

I don't mean volume. I'm I talk with my

51:48

hands. I'm Italian. I always talk loud.

51:50

I mean more so just kind of letting your

51:51

achievements show through. Um and really

51:54

standing out. So uh you know we are all

51:57

in the media space. We're all loud

51:59

working constantly. This has been a, you

52:00

know, a thing. But, uh, what's your

52:02

take, Luke, on, you know, this idea of

52:05

people now trying to tout their

52:07

achievements and talk very loudly. Is

52:09

this something that you think has always

52:10

existed or or is really popping up now

52:12

more?

52:12

>> I mean, I'm happy it's spreading. Like,

52:14

I came in I came my first ever job, we

52:16

had a scoreboard with how many page

52:18

views you had up on up on a wall and how

52:20

many posts you had written that day. So,

52:22

like that's the environment in which

52:23

I've always worked. And so, the idea of

52:26

of loud working of saying like, "Hey,

52:28

this is what I've done. you better catch

52:30

me. It's like I I think that kind of

52:32

like healthy competition and as long as

52:34

it's kind of, you know, in the service

52:35

of teamwork and not putting down other

52:37

people's achievements, but constantly

52:38

racing to do more, do better. I think,

52:40

you know, that's how great organizations

52:42

operate. It's when people feel pride in

52:44

their work, pride in their

52:45

accomplishments, they're putting them

52:46

out there and others are following suit.

52:47

>> Mhm. Lauren, you cut your teeth on the

52:49

trading floor, right? Obviously, then

52:51

kind of continuing to go in in the

52:53

content game. What's What's your take?

52:54

>> I I love it. I mean, especially starting

52:57

in an environment, an allmile

52:58

environment such as the New York Stock

53:00

Exchange, um, it it is really healthy

53:03

competition and I and I think it only

53:05

motivates you to want to do more. So, I

53:07

I'm all for it.

53:08

>> Shai, what do you think?

53:10

>> It's the new age of Hunger Games. It's

53:12

going to be just intense and intense and

53:15

I think it honestly it it it's justified

53:17

because the we're in the stage of AI

53:19

where it's replacing the jobs of the old

53:21

economy in order to make room for the

53:23

new jobs of the new economy. So what I

53:25

would tell people is just to become more

53:27

leveraged. Like if you're in fear, like

53:30

don't scream it. Just over the next

53:32

couple years, like find ways to automate

53:35

repetitive workflow in your department

53:36

and make yourself irreplaceable because

53:39

guess what? You show the initiative and

53:42

you ended up winning the Hunger Games of

53:44

your department.

53:44

>> Yeah, I think it's it's definitely true

53:46

that the idea of, oh, we'll just put

53:47

your head down, work really hard, and

53:49

things will work out. Like those those

53:51

days are gone. It's it's it's long gone.

53:53

You have to be a self-starter. You have

53:54

to be willing to kind of jump in the

53:56

fray. And and I think Shay to your

53:58

point, you have to be real or sorry, you

54:00

have to be really uh specific about

54:01

finding sweet spots in your

54:03

organization. Like what's a problem that

54:04

I can fix that my boss or my boss's boss

54:07

will really notice? And then that kind

54:09

of puts you on the fast track as far as

54:10

like getting noticed and and getting

54:12

really thought out. So yeah. No, I mean

54:14

a lot to uh to chew on. What's uh you

54:16

know, anything you guys want to tout

54:18

>> shout? Has there has there ever been a

54:21

time where putting your head down and

54:22

just doing good work paid off? Like is

54:23

that is that a far gone time or does

54:25

that just never exist?

54:26

>> I don't think so. It seems like from the

54:27

movies.

54:28

>> Yeah, maybe that's me being too

54:29

nostalgic. Well, on that note, um

54:32

Lauren, Luke, Shay, thanks so much for

54:34

joining us. This was an awesome market

54:35

hang. And uh yeah, until next time.

54:37

We'll see you tomorrow.

54:38

>> All right, that was fun.

55:37

Heat.

56:01

Hey, heat. Hey, heat.

56:04

Heat. Heat. N.

57:05

Down.

57:25

Down.

57:30

Down.

58:14

Down.

58:37

Ah.

58:54

Heat. Heat.

59:12

Hey,

59:22

hey, hey.

Interactive Summary

The discussion on Market Hang covers several key topics related to technology and markets. The panel first delves into AI safety and regulation, questioning if it's a solvable problem given the rapid pace of technological advancement versus government oversight. Nvidia's proactive approach in creating AI guard rails and the potential for open-source models as a solution are highlighted. The conversation then shifts to the upcoming midterm elections, discussing market reactions, the affordability agenda, and the potential impact of AI's electricity consumption as a sleeper issue. The anticipated Anthropic IPO is analyzed, comparing its potential market influence to that of SpaceX and raising questions about its high valuation and impact on capital allocation. The panel also examines AI budgets and the return on investment (ROI), noting a shift towards outcome-based pricing and the increasing cost of trial-and-error. Finally, the concept of "loud working" is introduced, emphasizing the need for self-promotion and continuous skill development in a dynamic, AI-driven job market.

Suggested questions

10 ready-made prompts