Thank you.

Welcome to another episode of Cloud

Unplugged.

Today,

we are not going to do the news,

actually, a bit of a curveball episode.

Given the fact that the news is just

all AI anyway,

what we thought we'd do is talk about

whether AI is a bubble,

is it all hype,

and get some data in our conversations to

share about

what we believe maybe it is,

maybe it isn't a bubble.

It's a little bit hard to predict.

So we just thought we'd present some of

the facts and talk through things.

Because there's been different bits of

surveys coming out.

There was the Writer Enterprise,

Morgan Stanley have kind of come out and

said things.

McKinsey, there's reports there.

So yeah, Salman, is it a bubble?

Do you think it's all hype?

What's your opinion?

And then we'll get into the data.

Of course,

we will definitely get into the data,

John, but I'm going to call it out.

It's hype and bubble at the moment.

Of course,

it's the best piece of technology we've

seen in a long time, John.

Not denying that LLMs are great at being

a very good autocomplete,

but it seems to me like there's a

bubble.

And of course,

there's data that we'll mention about

quality,

about what the companies have been doing

recently.

You know, they're running out of tokens.

Budgets are being blown for Uber and

they're not really seeing any return on

investment.

I'm calling it out right now that it

is a bit of a hype and a

bubble.

It's a great tool and it should be

used like a tool.

And all this doom and gloom about AI.

this and that and all these IPO charts.

That's my opinion.

But opinion needs to be backed by numbers,

John.

And I know you're a numbers man,

so we'll talk about some numbers now.

But what is your opinion, John,

before we go into the details of what's

going on with this thing?

So what is my opinion?

And this is a bit anecdotal,

so I'll go for the anecdotal before we

kind of get into the numbers.

Anecdotally, I would say, well,

this is maybe less anecdotal,

but I think the monopolization of AI from

a market perspective, i.e.

Nvidia, Corweave, Microsoft, Google,

Anthropic, etc.,

The way the money keeps moving around,

the way the investments are offset against

other investments,

as in like this CapEx expenditure,

they're going to invest in loads of chips

in the video and the video is investing

in CoreWeave and guaranteeing then the

cost on the chips to them if they

don't kind of meet the targets, et cetera.

Google get an investment from equity,

kind of, I guess,

raising equity to kind of de risk

themselves financially for new data center

builds.

So it seems quite highly concentrated

financially,

which feels higher risk almost because

It won't take much for one of those

dimensions to shift,

which will then have ramifications for the

others.

On then the consumer side,

so outside of where all the money is

and where the investment is going.

Sorry, just to add to that as well.

the US market is much is really the

economically is predicated on AI at the

moment.

I think seventy five percent of the market

is all AI is all AI based.

And I think it's all CapEx based.

I think two point five percent.

We are getting some numbers,

but two point five percent

of US GDP is based on all the

CapEx expenditure,

which means they keep needing to spend

CapEx to keep growing the GDP,

which is obviously putting a lot of

pressure really on AI as a whole

economically, which is now raised.

People have come out and kind of raised

concerns.

Federal Reserve has kind of come out and

raised concerns, et cetera.

But anyway,

flipping that to the consumer side,

I think it's

because the technology is a little bit

early,

the value recognition isn't being seen

just yet.

with anything i see it at the moment

as still being quite a tool um you

know and you've got to master a tool

like anything else to get good at um

databases or to get good at data in

general maybe snowflake or databricks

whatever you've got to start to specialize

in that product to do it really well

to do it justice and to really understand

how to use it um you can do

anything badly um you know and that's easy

to do anything badly so i think

The adoption obviously is increasing,

but the skill over the adoption is still

immature.

And so you're kind of on that curve.

As the maturity of the skill changes and

people start to get more switched on about

how to work with it properly and

maybe there's more maturity on the

outcomes it can drive,

the pipelines that guarantee the

consistency of the models,

and that starts to mature,

then I can see that the value will

start to get recognized, i.e.

businesses will see the value in AI

properly, especially on the agentic side.

Whether, though,

the capabilities and the features that

people produce have high ROI on them in

the end, i.e.,

Can you really drive amazing incentives in

your product using AI?

Does it move the needle financially for

your business?

Because now you've got these

differentiated features and how

differentiated are they and how much value

can you command on those features from

others financially?

Are people going to pay for that?

Are people going to double the price or

is it just marginal gains?

um that's the bit that no one really

knows has it cost you more than you're

gonna get return on you know when you

start to use it and i think that's

the little bit of all the unpredictability

of it all um

But what do you kind of agree with

that?

Or do you think I agree with that?

Because at the moment,

like with all the tech companies,

there was a report out for every dollar

OpenAI earn,

they spend one dollar twenty two cents.

Right.

So, of course,

all companies in the beginning are running

at loss.

We know this, but

I think the real cost of these

subscriptions and the model is going to

come up quite soon.

So right now,

everyone's using subsidized costs because

we talked about how much they're spending.

OpenAI is going to spend at one point

four trillion dollars over the next eight

to ten years on this compute.

By the way,

for initial public offering that they did

confidentially at the moment,

the company is valued at roughly around

one trillion.

So where are they going to get this

one point four trillion from?

I don't know.

But to go back to your question,

are people going to get to a point

where they can drive value from LLMs?

And you're absolutely right about

You have to use it as another tool

that you have to master.

They have to understand how do I create

repeatable,

reproducible outputs from whatever I'm

doing, right?

It's very good for doing a lot of

things really quickly.

I can do some documentation.

I can spin up a proof of concept

for productionizing it and doing it over

time and having an output that is

maintainable and that you can add

features.

Because

at the moment john if i ask it

to create me a website for cloud unplugged

and i don't don't like something i tell

it yo can you change this the thing

has to take all the waste all the

tokens and restart again from the

beginning right so it will take all the

input like oh let me just give you

the whole website again

And everything again.

But if it was a developer,

developer would go, oh,

you just want me to change the size

of the text?

No problem.

I'll just open the CSS file and just

change the text file.

And this is what we're seeing.

Some of the companies are now saying,

well, hold on a second.

Maybe we'll go back to the junior

developers that we used to have before and

get them to do some of this work.

You know, like this is happening in meta.

And people were fixated about, oh,

who's spending the most tokens?

That's like saying,

Oh, I have a chef in the kitchen,

and I want to know how much gas

he used.

I don't care about what you produce.

I don't care about how good the food

is,

but how much gas and electricity did you

use to cook the food?

That's what it seems like.

And all these companies fixated over, oh,

I have a leaderboard, you know, like...

uber wasted all their budget and so did

microsoft and on ai spending and people

without understanding the use of the tool

were using this thing and getting too

excited now of course i got excited as

well and i still get excited there it's

uh john it's like a roulette machine you

know like oh yeah let me just do

one more prompt i think you'll get it

right let me just do one another prompt

i think you'll get it but

just predicting the next token and i think

the risk as well economically if if you've

inflated something on one side so say say

that the hype is a hype um yeah

to what extent it's a hype is the

question do you know like how realistic is

the value that's been promised

You know, and that's on a spectrum,

you know,

of like how much value can you really

get from it in the end?

And no one really knows.

So if you kind of see it as

upstretchings,

you've got that on one hand.

But if the hype is the hype and

then you've got technologies on the other

side,

so like robo taxis or self-driving cars,

et cetera,

that start to displace taxi drivers.

So you've got like, I don't know,

three hundred thousand or however many

jobs,

taxi driving jobs or whatever it is in

the UK,

I think something like that anyway.

And you are then inflating something on

the other side or incentivizing on the

other side.

And then the costs change,

the cost profiles change,

and the economics start to disbalance.

So it starts to distort the economics.

You've just displaced a stream,

economic stream.

So what you've done is you've just removed

in your country, really,

a whole financial structure.

It wasn't massive necessarily,

but it's a whole market in itself,

taxi driver market, cars, rentals,

car rentals, et cetera.

Obviously,

it's a chain of things that get disrupted

within it.

People have got higher purchase cars

because they're attached to it, et cetera.

All that starts to get disrupted,

and then you've got a robo-taxi owned by

one company.

And if it doesn't work out,

you've almost started to tilt the economy

in a direction that may be

isn't ready to be tilted into before

you've started to maybe risk the

displacement displacement of another

economic structure um i think that's where

it's more you know i guess if i

still look at it objectively there's more

apprehension on such high impact for

something that hasn't really been fully

understood it's almost like you're we're

all racing in a direction

And it's kind of carnage.

Do you know what I mean?

It's like everyone's trying to get out of

a football stadium all at the same time.

And everyone's like scrambling and running

out.

And there's no one really being like,

well, actually,

if we just slow down and we just

all steadily kind of go in an orderly

fashion,

it's less risk than us all stampeding,

you know, kind of outright.

So that's kind of how it feels a

bit.

Yeah, and for sure, John,

the economic impact that you talk about,

of course,

is going to impact some of the industries,

like you talked about robot taxis coming

in and impacting the taxi drivers.

The technology advancements happen all the

time, and they do have an impact.

Let's go from horses down to cars.

That was the whole industry that got

impacted.

And I think the weird thing about all

of this AI hype is

is how it's overhyped by the people who

are behind building these tools oh yeah

all the anthropics take your anthropics

takes your open ai every day they're

coming in and they're saying oh we're not

going to need any developers oh i came

into this podcast today one of these i'm

not going to say well okay some people

from anthropic i came to this podcast

today and i opened a fifty uh pull

request on the way to the taxi

Well, okay, fine.

Yeah, you did it, of course.

Now,

I don't know how much of that is

truth and how much of that is just,

oh, we need to do an IPO.

Let's just quickly get it out just so

we can get our names up.

Now, let me give you an example, right?

When the steam engine came out,

Because these people are saying, no,

you don't need software developers.

When the steam engine came out,

people were like, oh, right,

we won't need coal anymore because steam

engine is so much more efficient.

So we won't need that much coal.

But what people fail to realize,

if you have something that's super

efficient,

the usage of that thing goes up.

So everybody started using steam engines

in factories now and in ships and

everywhere else.

The coal usage actually went up.

It went up.

So, you know,

it's the same kind of thing.

So I don't really like the narrative,

of course,

speaking as a developer and a platform

person.

I don't really like this doom and gloom

narrative.

Okay, we've got this new technology.

Nobody will need it.

And now people are starting to realize

that actually, as you say, John,

maybe this technology is not all what we

hoped it was.

I know people keep talking about AGI is

around the corner.

This thing can't even predict something

properly.

You give it the same prompt twice,

it's going to give you something else and

start hallucinating, start making things.

So for me, John, it's overhyped.

And these companies now at the moment

seems like they're racing to do an IPO

just to cash in still on the hype.

Yeah, I kind of agree.

Open AI have just done a confidential

listing of IPO.

That means they're working with the

authorities to figure out what are the

steps they need to do.

And we can't see the details yet.

Nothing's been made public,

what their income is,

how much their debts are,

what the revenues are, what the risks are.

I'll be very interested.

to know when the report comes out,

how much they've spent.

There's estimations around the world at

the moment.

People estimate how much they've actually

spent.

That's when we're going to know the real

picture behind.

Well,

I don't know if we will ever know

the real picture.

That will know the real picture behind how

much is actually costing for these

services to be built and consumed.

And that is, John,

when people are going to have to pay

the real money

this right anthropic yesterday announced

this fable model which is based on mythos

which is like closed nobody can use it

and that thing is ten times more expensive

in terms of input tokens and fifty times

more expensive in terms of output tokens

right it's like a drug dealer saying i'm

gonna give you and then they reset it

everybody's uh limits yesterday just so

people can try this new drug for free

and then just get hooked in and they

were like okay guys

Now you actually have to start paying the

real money once you're hooked in.

Do you think with Fable,

do you reckon it's astronomically...

worth what what what was the rate in

token is saying it's like two point five

times more expensive was it or something

yeah it is something like that wasn't it

yeah i guess the question is do you

get two point five that's if that is

correct two point five times more value

comparatively from four point eight to

fable and it's like well i doubt i

doubt that i doubt it's like the

multiplier like a value is

is kind of true.

But yeah,

I kind of agree with all the people

that have the most to gain control of

the narrative in the media,

and they're the ones that are all on

the hype.

What is funny, though,

is like we were saying last episode,

that all the students are kind of booing

every time AI is mentioned.

So there's a civil...

you know, unrest.

Civil unrest, John.

Civil unrest on what's kind of going on.

Yeah, there is.

I don't think that isn't true.

But yeah,

if we get to some of the facts,

I don't know if you saw this,

but

The Federal Reserve has flagged AI as a

top systemic risk.

The Military Authority of Singapore,

they've said that the market stretched,

kind of another term for bubble.

ECB, they're calling it a bubble.

So it's all started to move.

I think everyone's kind of looking.

And being like, OK,

there's a lot of expenditure happening.

There's a lot of investment.

Those are capex spend.

That's great economically.

Seventy five percent of the economy in the

US is now based on AI.

OK, so I think everyone's a bit like,

OK, this feels a little bit unnerving now.

It's not going to take much.

I think there's something to happen on

April the twenty eighth.

There was a report that suggested OpenAI

was missing their internal targets,

kind of growth targets.

That dropped the share price of Oracle and

NVIDIA by three percent.

And then Broadcom and AMD,

they got affected.

So that goes to show how fragile the

market is on the earnings and the same

for

basically what could happen is that could

kind of continue.

So open AI could come out and give

their earnings.

They could be way,

way off financially than what everyone was

expecting.

And then the hyperscale is a bit like,

oh, maybe we need to cut our capex.

And if the growth looks like it's slowing

down or isn't there,

and then they start to pull back and

then the investment in the chip starts to

pull back and that affects the video and

then on it goes and it starts to

starts to trickle around and through the

economy that's i imagine what is probably

likely to happen if the growth forecasted

growth isn't being matched um in the hype

um but i guess we'll see and then

on this other side um and some of

this you can kind of take with a

bit of a pinch of salt but

because I think it's a bit questionable.

But MIT,

this is back in twenty twenty five.

So things have changed.

But they had ninety ninety five percent of

fifty two enterprise organizations said

they had no return on investment.

I think they spent thirty to forty billion

across something like three hundred Gen-AI

initiatives.

This is coming from like this is like

Yale Insights.

And they all said like,

ninety-five percent basically didn't get

anything off the back of the thirty to

forty billion.

And then PwC,

they've said that fifty-six percent of

CEOs also reported getting nothing from

their AI efforts.

And right as May this year,

twenty-twenty-six, I think this first,

I think the Q one,

seventy-five percent of executives admit

that AI strategy is more for show

uh than internal guidance um and forty

eight percent call adoption a massive

disappointment and thirty nine percent

have no formal plan to drive revenue from

ai um

are all the different statistics.

Thoughts on those little numbers?

Sounds like when you go to the

parents-teacher meeting and the teacher is

so telling you the report is really bad

for your child and you're a bit

disappointed because it showed so much

promise when it joined the school and

you're like, yeah,

my kid is going to fly through it,

ace all the exams.

And just to add to that, John,

a lot of work is being done by

There's a lot of paper.

I'll just pick one.

Code Rabbit did a study on some work.

They used these models differently.

It doesn't matter what model you pick,

John.

That doesn't make any difference, right?

Yes,

some models are going to be a little

bit better than others.

And the usage of the models absolutely

depends on how you use it.

is how you get the most out of

it right but that's taking you know people

like oh you need to have the right

skills you need to write this you have

to write guardrails blah blah fine yeah of

course yes you do what they found john

was with ai assisted they were finding

seventy percent more bugs so code rabbit

published out this report and they found

hundred and seventy four percent more

vulnerability

And this is not just them on their

own.

Cornell University have done a bunch of

these studies.

And they're published online.

You can find them.

And maybe we can put in the show

notes afterwards.

And they're finding that not just that the

CapEx numbers that you're saying, John,

are looking bad.

They're reflecting what is the actual

reality.

People are not seeing value.

That's one thing.

know not seeing value you can you need

to uber said they can't really attribute

the the budget that they blew for their

ai in in in the first half of

the year for the whole year they can

the ceo cannot point it to any feature

that they released using yeah well that's

fine right you can you can't see any

value but at the same time people got

these hard metrics where they're like yeah

it's good in places but we also have

other problems i've got more bugs to be

fair though if ai's if you are going

all out on ai

And all the code is now kind of

contributed by AI.

Then anything that's released just by

default has been AI generated and AI

contributed,

which then means technically there's then

some form of value,

even if there's fifty six billion more

bugs, even if it's kind of sloppy code,

even if it's got more security

vulnerabilities.

I guess, I mean, depending on the numbers,

you could look at, you know,

depending on the spin on the data,

you could be like, yeah,

it's all AI code generated.

We're shipping.

But it's like, yeah,

because you rolled out AI, like across,

you told everybody to use AI.

And then by proxy,

all your features obviously are now going

to be generated by AI.

It doesn't necessarily mean the outcome is

good.

exactly how are you going to compare that

to what you were doing last year without

ai and of course you can compare that

because like how much money did we make

right yes the global markets the impact

the economics impact but this stuff uh you

know like uh it's it's it's it seems

a bit absurd because uh there was uh

there was this report uh from

from this news come out for the last

couple.

It's a little bit old now.

There's some work done by Deloitte for the

Australian government where they spend a

lot of money.

I think quarter of a million Australian

dollars on doing this report and analysis

for the Australian government.

And then a person was reviewing it.

And then there was a reference in the

report of all because Deloitte will

produce a report.

You need to do this to get better

at whatever, whatever.

Right.

So the person found out that all the

references in the report were made up.

They were hallucinated.

They're all hallucinations from AI.

So the report was, we can extrapolate,

was basically generated by AI.

The person was like, what is this?

This is kind of useless, right?

But usually what people do when they get

a report that's generated by AI,

they feed it to AI to summarize it

for them, right?

That's what the people generally do.

Oh my God, I would never do that.

I would never do that.

You'd never catch me doing that.

Deloitte's response was, yeah,

fair enough.

The reference was probably made up.

But don't worry.

It still doesn't change the strategy

that's outlined in the document.

The strategy is still the same.

And this is another Canadian lecturer,

a researcher.

Again,

she got some work done and read it.

And she found out that it generated a

reference

that belonged to her that she hasn't even

published.

Right.

So it just seems to be that all

the all the hyperscalers because who wins

from this, right?

The people who've got the model,

people who train the model,

so the chips that you build on,

who runs the infrastructure,

and the circle that you talked about in

the beginning of friends,

just investing in each other,

they're the ones who are getting value

from it.

And

Yeah, I've used AI for various things.

Of course,

it makes an improvement everywhere,

but now we're starting to see that there's

so many issues related to it.

And this is across the board, John.

This is across the board.

So we, of course,

will come to a conclusion in a few

minutes about what it is,

but it doesn't seem...

it doesn't live up to the hype it's

not living up to the hype for me

for sure it's not living up to the

hype I kind of agree I think my

prediction will be that someone's going to

miss their earnings or multiple companies

will miss their annual earnings

I think then the run rate of those

companies obviously starts to reduce

because obviously it's forecasted against

something.

I think then the hyperscalers will

probably start to cut back on the

investments a little bit.

Market cap will drop and have massive

impact.

I think also maybe if any Fortune one

hundred company starts to say that they're

going to cancel AI agent rollout or maybe

they get hacked and then they're like,

actually,

we need to pause or that then obviously

also has high impact to the market and

adoption because obviously that's risk

based.

Then because of the impact, then the Fed

start to raise rates because of inflation,

the rates that make it more expensive for

AI companies to raise capital.

And obviously,

then it goes and then it means that

there's less investment going into all

these AI companies because the cost of the

capital is too high because the interest

rates have gone up.

And then all of a sudden it all

starts to turn in itself a little bit

and then it kind of starts to stay.

That is my...

If I was to predict,

I reckon that's probably what will happen.

That's nothing to do with people still not

having success with AI, though.

But I think it's just more on the

hype.

I kind of agree,

but at the same time,

what will also happen,

I a little bit disagree on this because

you might remember, John,

when Uber came out,

when they did the IPO and they were

funded in twenty twenty ten.

Right.

So they started in twenty ten.

They were unprofitable till like twenty

thirteen for thirteen, fourteen years.

They didn't make any money.

Yeah.

Right.

And

I think these companies are going to do

the same thing.

They'll get floated.

They'll be unprofitable for years and

years and years.

And then eventually, because initially,

the reason why Uber wasn't profitable,

but of course,

because they're investing in

infrastructure, the system itself,

and they're undercutting the taxi

services.

And it was way cheaper to catch.

They were doing offers for Uber drivers

and eventually got it.

And that's what these AI companies are

doing.

And when they get floated,

it will be the same.

I don't know about that, though.

I think the only difference is...

Uber was different because it was a single

company that only specialised in taxis and

everything was invested.

So their trajectory was all about

capturing the market,

taking the market demand.

These are established companies that have

been around a really long time that don't

just do AI,

who've got market share already IPO'd.

So there's a lot of share price in

there and they're reusing their own

capital.

So they're investing their own capital

into that direction, which is like,

so i think the impact could be higher

obviously to you know investment and share

price and everything else and i think

that's what could tank tank not

necessarily tank the economy i don't think

that would but i think it could at

least start to have a negative impact if

things aren't

as everybody would hope on the happy path,

I guess.

Fair enough.

There's shortages for transformers for

these data centers because all of this

stuff still requires data centers.

You need to still build all these data

centers based on the projection that

people are saying,

these companies are saying we need US

dollars worth of infrastructure to be able

to

capture to be able to meet the demands

for the customers but on the other side

you're seeing a bunch of these ai projects

are getting cancelled the consumption is

going down so i think you are right

maybe the impact will still be the same

like it might not be as hyped when

it when it goes ipo but i think

people will still buy into the hype

because you know these companies are

masterful at

coming,

turning up and making a high level

statement saying, oh,

I haven't opened one of the Claude creator

just turned up and said for six months

I haven't opened an IDE.

I've not looked at any piece of code

that's been created.

I don't understand that part.

I actually don't understand that part.

Maybe I'm not using the tool correctly,

John.

Yeah, I think you're a denier.

I'm a denier, John.

I'm a cynic.

Call me a cynic, John.

Call me a cynic, basically.

You just don't believe in these things.

I like John.

You keep me grounded like this, right?

I'm a flat earther.

But yeah, I think it's been good.

I think it's been good to kind of

like, we obviously don't,

no one really knows how things are going

to play out.

We can't see into the future,

but I think it's good just to kind

of pull on the numbers.

I think there's a lot,

still a lot more

possibilities and a lot could change

regulation as well.

We haven't even spoken about regulation.

That's a whole other topic.

What impact that would have when that

starts to happen, et cetera.

So I do think at the moment it's

too much on the happy path.

Something is going to give don't think

that means it's not necessarily valuable.

It's a synopsis, isn't it?

Yeah.

So, John,

before you end on the scale of one

to ten,

how high this technology is versus what

the reality is.

So the higher the number,

the better it is.

Oh, the higher the number,

the better it is.

I'd probably say it's on a,

I would say it's on maybe a seven.

I would say, because I think, well,

market, I'd say he's on a nine,

the high value to, to, uh,

to a consumer, I,

a company trying to use it,

or I'd say it was on a,

probably on a seven because I think, um,

Or maybe a little bit less,

because I think there's still a lot you

can do to make it good.

And I think it's still a lot of

improvements that are going to come out

that make it much more high value for

businesses.

What about you?

Yeah, I think I'll go with like eight,

a little bit higher than that.

But we're talking about technology as it

is today.

What comes out in six months time?

Don't know.

But AGI is too far away, man.

there's no AGI.

Exactly.

Richard Dawkins said,

animals are not intelligent either.

So how is a Python script going to

be more intelligent than an animal?

You just wait.

You just wait, Simon.

You just wait when we're all,

the next podcast episode is two AI avatars

having a chat and we're just out having

lunch.

Right.

Anyway, we better wrap this up.

It's been a bit different,

more fun to kind of get into the

detail rather than just the hyperficial

news.

But we shall speak to everybody next week.

Cheers.