0:03

This is Business Breakdowns. Business

0:05

Breakdowns is a series of is

0:07

a series of conversations with

0:10

investors and operators diving deep into

0:12

a single business. For each For

0:14

each business, we explore its history,

0:16

its business model, its competitive

0:18

advantages, and what makes it... makes

0:20

it tick. We believe every believe

0:23

every business has lessons and secrets that

0:25

investors and operators can learn from. from,

0:27

we are here to bring them to

0:29

you. to you. To To find more

0:31

episodes of check out check out

0:33

joincolossus.com. All opinions expressed by hosts

0:35

and podcast guests are solely their own

0:38

opinions. Hosts, Hosts, podcast guests, their employers,

0:40

or or affiliates may maintain positions in

0:42

the securities discussed in this podcast. this

0:44

This podcast is for informational purposes

0:46

only and should not be relied upon

0:48

as a basis for relied upon as

0:50

a basis for investment decisions. I'm Zach

0:53

Fuss and today we're breaking down arm

0:55

holdings. Armed designs the architecture powering

0:57

billions of devices from smartphones and

0:59

centers to to devices and

1:01

automotive systems. In this In

1:04

this episode, we'll explore value unique

1:06

value proposition. it thrives as a how it thrives

1:08

as a licensing giant in a

1:10

market dominated by leading -edge manufacturers. this business

1:12

discuss the business model, that the partnerships that

1:14

drive its growth, in and its role

1:16

in enabling companies like Apple, the and

1:18

Qualcomm. We We will also

1:20

unpack ARM's business history, including its

1:22

acquisition by by Softbank. its failed

1:25

takeover by Nevada, and its IPO

1:27

earlier this year. currently Orm currently

1:29

$150 billion market market cap,

1:31

but sales $5 $5 billion.

1:33

robust a rather robust 30

1:35

times revenue multiple. is, of course, This

1:37

valuation is of course predicated

1:39

on its 90 % plus gross

1:41

margins, by arm's by ARM's dominant market

1:43

share and its and its share share

1:45

gains in markets like automotive,

1:47

cloud computing, and IoT, IOT, with with

1:49

its technology being licensed in

1:51

nearly 30 billion chips annually. annually. To

1:53

break down I am I

1:56

am joined by Goldberg,

1:58

who is the CEO

2:00

and lead at D2D Advise.

2:02

a technology and strategy consultancy, and who

2:05

has spent various roles in the semiconductor

2:07

ecosystem. We hope you enjoy this conversation.

2:09

All right, Jay, it is great to

2:11

have you back. The world continues to

2:13

change and evolve, particularly as it comes

2:16

to your area of expertise, semiconductors. Today

2:18

we're talking about arm. So to start

2:20

things off, I know this is a

2:22

loaded question, and it has a fair

2:24

degree of complexity. may require a bit

2:27

of a semiconductor 101 education for our

2:29

audience, but let's start with a brief

2:31

overview of arms business model, what exactly

2:33

they do, and how they fit into

2:35

the broader semiconductor ecosystem. So I'm going

2:38

to give you an analogy on arm

2:40

to give an intuitive sense of what

2:42

they do. This is not a hundred

2:44

percent perfect analogy, but it's what we

2:47

work with. Arm licenses, it's intellectual property.

2:49

It's IP, two companies that design chips.

2:51

So they don't make chips themselves and

2:53

they license their IP to companies like

2:55

Qualcomm or Invidia or Broadcom, these big

2:58

chip companies, who then design that IP

3:00

into their own chips, which they get

3:02

manufactured at CSMC or one of the

3:04

foundries. So arm is fairly early in

3:06

the process of this whole semiconductor flow.

3:09

The way to think about arm is

3:11

that the IP that they provide is

3:13

almost like a blueprint, but a special

3:15

type of blueprint. So imagine you're an

3:17

architect and you're designing a house. As

3:20

an architect you differentiate yourself by how

3:22

the house looks. What's the curb appeal?

3:24

How does the light flow? These big

3:26

architectural features. Typically as an architect you

3:29

don't get rewarded much because you have

3:31

the best designed plumbent. You don't have

3:33

the best designed bathrooms. Those are important

3:35

in a house, but that's not really

3:37

where you as the architect differentiate yourself.

3:40

And especially in lots of places, bathrooms

3:42

are generally very standard. And what you

3:44

could do is just take the generic

3:46

blueprints for a standard bathroom and you're

3:48

going to copy and paste those around

3:51

your design. That frees you up to

3:53

design the parts of the house that

3:55

really matter. that are really going

3:57

to drive your

4:00

business as an architect

4:02

as an architect I think it's a

4:04

good analogy for how arm interacts with

4:06

its customers. with There's a certain amount of

4:08

low of work that needs to be

4:10

in lots of types of chips. Certain Certain

4:12

types of math, certain types of functions

4:14

need to be in all chips, but

4:16

there's no way that a no way that a

4:18

Qualcomm or a NVIDIA really differentiates around those.

4:20

It's basic math things that need to

4:22

get done. that need to that IP from

4:24

from arm. and then they work that into

4:26

their chip, and then they can differentiate

4:28

on things on things It's how their It's work,

4:30

how their communication systems work, or for

4:32

NVIDIA, how their graphics and AI processing works.

4:35

graphics and AI processing vital function, providing a

4:37

really important piece of the overall

4:39

functioning of a chip. of the with

4:41

still allowing the flexibility for their customers,

4:43

their licensees to design their their they

4:45

see fit. what they do is they

4:47

license that IP, So what their business model

4:49

works essentially IP they charge business payment and

4:51

then they take a royalty, their amount of

4:53

money per chip per chip licensee actually ships.

4:56

actually a really helpful analogy

4:58

to frame it. My research it. In

5:00

my research, clearly, GPUs have the day, at

5:02

least in the last call it 12

5:04

to 18 months. My understanding

5:06

is that this is a CPU

5:08

oriented business. So help So us us to

5:10

better understand the importance of their

5:12

CPU versus the GPU architecture that

5:14

people are working with today and

5:16

the interplay between those two in

5:18

a future world state. The The CPU is

5:20

very much a general -purpose chip. run It

5:23

can run any type of workload. the It

5:25

can run the low -level functions of the

5:27

keyboard the mouse and the hard drive

5:29

for your computer. It can handle the operating

5:31

system and the applications. It's

5:33

a general -purpose purpose jack-of-all chip. chip. But of

5:35

the geometry and physics of of

5:37

you can always design a chip

5:39

that does a single one of

5:41

those tasks better than the CPU. CPU.

5:44

the case of GPUs, they were

5:46

first developed to run graphics really,

5:48

really well. A CPU could run basic

5:50

graphics, but a GPU can do much better

5:52

better than a CPU. GPU can't run GPU can't

5:54

run the operating system well. It can't do

5:56

all the low -level functions as well. can It

5:58

can do graphics really well. So an engineer... you're off,

6:00

what do you need the chip to

6:02

do? need You make those trade You you

6:04

design different tasks to different chips. And

6:06

with GPUs now becoming important tools for

6:09

AI, with GPUs these becoming chips are even

6:11

more finely tuned towards chips are even more finely

6:13

And that's similar to GPU math. don't

6:15

need to get into that's video story.

6:17

In regards to need to what's common to

6:19

all this is you have certain general

6:21

purpose functions In ARM sits at the heart

6:23

of all those. to So you don't

6:25

need certain general purpose IP, that ARM sits

6:27

at of a GPU. of all those. because

6:30

not what the GPU is there to do, but you're gonna

6:32

want it inside the CPU. You're gonna

6:34

want that functionality. inside the most compute systems,

6:36

you're always gonna need some functionality. So all

6:38

the multiple tasks. So even in these

6:40

big NVIDIA systems that have to 72 GPU

6:42

cards all linked together, there's gonna be

6:44

some number of CPUs in those as

6:46

well. So the way to think about

6:49

arm in relation to the growth of

6:51

AI, 72 GPU the hot topic now is

6:53

going to are exploding. We're seeing so many

6:55

more GPUs being sold eight or doing really

6:57

well selling all around the world. there

6:59

is some degree of CPU attached to

7:01

that. depending on where they source the

7:03

CPU from, that can often be

7:06

an ARM -based CPU. And in fact,

7:08

NVIDIA probably prefers to sell its own

7:10

CPU, to sell its own It often attaches

7:12

those to its GPU systems to its that

7:14

CPU is ARM CPU is arm-based. we

7:16

we their unique business model model the

7:19

CPU versus GPU GPU dynamics, I I

7:21

guess I I to really try to

7:23

wrap it all around is that

7:25

core technology. What is it

7:27

about IP IP and their

7:29

business model, how it strips, that differ

7:31

differ from their competition? armyly faces

7:33

two elements of two elements of

7:35

competition. they have been seen as they

7:38

have been seen with X86. with to take a step

7:40

back and take a step back and

7:42

say, called sometimes called set architecture, an

7:44

ISA. And that's just and that's just a

7:46

framework for how ships are supposed

7:48

to handle these types of math,

7:50

these sorts of problems. 40 years

7:52

ago, ago, there were dozens of

7:54

different of different ISAs. And over time, down to

7:56

two, two. the arm There's the arm

7:58

architecture, and then there's the... X-86 architecture. And

8:00

today there's two companies that provide X-86 architecture,

8:02

AMD and Intel. Intel started it and then

8:04

AMD joined along. And arm is the other

8:07

instruction set architecture, which historically wasn't used for

8:09

computers, wasn't used for PCs, really got its

8:11

big boost from smartphones and mobile. That's changing.

8:13

We can get into that. Historically, you'd use

8:15

X-86 in a PC and you'd use arm

8:17

in a smartphone. Most recently we've seen another

8:19

ISA enter the fold called risk five. R.I.S.C.

8:21

v. Risk 5 and that scene is an

8:23

open source, not exactly open source, but it's

8:25

open. It's not owned by one company. It's

8:27

designed by a consortium that anybody can contribute

8:29

to and use freely. So Risk 5 is

8:31

out there as an alternative to arm. It's

8:34

still very early days for Risk 5. So

8:36

it's on the, I'd call it an emergent

8:38

potential competitor as opposed to certainly not a

8:40

dire threat to arm anytime soon. So if

8:42

we think about how we got here and

8:44

work from the start. the general familiarity with

8:46

arm came through the early days success they

8:48

had with apple but how did they go

8:50

from this what i would call niche player

8:52

to someone that's so dominant in the architecture

8:54

of the future today what is the business

8:56

story arm came out of England out of

8:59

the UK there was a company called acorn

9:01

computers and this is back in the seventies

9:03

when everybody was making computers and acorn was

9:05

selling computers through other people's brands typically like

9:07

the BBC and they had a couple hit

9:09

products and they realized that to be competitive

9:11

they wanted to design their own ship. So

9:13

they tasked the team with designing a chip

9:15

as a photo CPU of the day and

9:17

that was the kernel of what would become

9:19

arm. Eventually they split that off the chip

9:21

design team and the hardware team went their

9:23

separate team and the hardware team went their

9:26

separate ways and the genesis of arm the

9:28

company today was originally called Acorn Risk Machines

9:30

and later advanced risk machines. We can talk

9:32

about what risk means later but the point

9:34

is it was part of Acorn it got

9:36

spun off into a three-way joint venture joint

9:38

venture between Acorn. VLSI, chip company of its

9:40

day, and Apple. Because Apple originally wanted to

9:42

use this arm IP. for the

9:44

Newton products. The Newton

9:46

didn't do particularly well,

9:48

but well, but Arm did

9:51

pretty well. made a

9:53

lot of money from

9:55

Newton, which it then

9:57

used which then grow as

9:59

part of portfolio. part-up And

10:01

from the get -go, the

10:03

Arm was an IP

10:05

licensor. They stopped making

10:07

their own ships when

10:09

they spun off they spun off,

10:11

their IP first to

10:13

Apple and then to

10:16

many others. to many others. slowly

10:18

but surely that grew. grew. then they

10:20

started adding customers. I think after

10:22

Newton, I think first few big customers were

10:24

fax machines, were designing designing fax machines

10:26

of all things. things. And then I think

10:28

what really tricked them into the beginning

10:30

of this big growth trajectory was

10:32

Nokia. was Nokia Nokia their their chip Texas Instruments,

10:34

started using ARM in the late

10:36

80s or early 90s. And the first

10:38

ARM first mobile phone was actually the

10:40

phone was which is the 8110, which is the That

10:42

was the first That was mobile phone. And

10:44

very quickly, other mobile phone companies started

10:46

to realize the appeal of using ARM

10:49

realize the ARM of using arm mobile phones. And

10:51

the reputation And very early on was that

10:53

it was much more power power than

10:55

alternatives, especially something like like X86, which at

10:57

that point really evolved into something that

10:59

was meant for a device that was

11:02

powered, a laptop or a PC laptop or a

11:04

to power. Mobile phones were so power

11:06

were so power conscious ARM had a lot of

11:08

appeal. it So in was in a lot

11:10

of feature phones. time, time, their capabilities increased,

11:12

and what really really kicked it in

11:14

the high gear was was 2007, the launch of

11:17

the iPhone and this explosion in in smartphones. suddenly

11:19

you needed very very complicated very advanced

11:21

processors for your phone the the best way

11:23

to do that was to start with

11:25

an the core the core of the phone

11:27

to to answer your question really the the

11:29

explosion came with the growth of mobile of

11:32

mobile. So history dictates here, here, arm

11:34

went on to dominate mobile

11:36

devices. Presumably they have a growing presence

11:38

in other markets, automotive, internet

11:41

of things. My guess My guess is

11:43

that if I walk through my living

11:45

room, there's probably eight or nine

11:47

devices that have some arm architecture licensed

11:49

to them. to But I guess I guess

11:51

most is this increasing demand for AI

11:53

AI. and machine learning. So I So I

11:55

guess in contemplating why a business today

11:57

that is rating close to to billion

11:59

of revenue. has a hundred and billion market

12:01

cap, it kind of begs the question, the

12:03

what the heck is going on here?

12:05

on here? So it's a few a few things. is, the

12:07

arm of today is very of today is very, very

12:09

different than the arm of just even a

12:11

few years ago. long For a long time, they

12:13

had talked about expanding into new markets. By

12:16

the mid much a they were very

12:18

much a companies. Ninety-something percent of the was

12:20

derived from derived from In the last

12:22

decade, but especially in the last

12:24

three, four years, they have made

12:26

huge inroads into other markets. The

12:28

most important of which is probably

12:30

the data center is they're helping all

12:32

the center, where Amazon, Microsoft, the Google,

12:34

Facebook, all in all, bye, bye, bye. at

12:36

this big Super Facebook, Ali now

12:38

designed their own CPUs to run

12:40

have workloads, and those are all based

12:42

on ARM chips. And this is an

12:44

alternative to using on arm Intel and AMD data center

12:46

chips. These companies went out and designed their own

12:48

chips. to using place for any of

12:50

those companies to start with would be ARM, because

12:52

there's really no other alternative. went So and

12:54

say their own and foremost, their biggest

12:57

growth opportunity. IOT is is

12:59

out there. IoT is a messy market. there,

13:01

They're doing okay there, but nobody does too

13:03

IoT. It's too complicated and too messy. is

13:05

certainly really, really, really interesting. I think I think they've

13:07

made good inroads there with a lot of their partners,

13:09

their a lot of their A are doing well there. are

13:12

takes a long time to mature. These are long

13:14

product cycles, five to a years to develop a chip

13:16

into production product as opposed to to or two years

13:19

for everybody else. a So we'll see how that

13:21

goes, but it's looking pretty good right now. as opposed

13:23

to one or is doing well everybody pretty much

13:25

everywhere. else. And I think you're right, AI is such

13:27

such the story of the day. it's

13:29

not it's not quite as clean

13:31

a story there, because I said, they're

13:33

not necessarily going to have an an

13:36

in every GPU, or every AI accelerator that's out

13:38

there, but there is some rate. You'll

13:40

need you'll you'll need other control functions.

13:42

There's cores in networking chips that

13:44

sit next to all of this to

13:47

connect all these all servers together. So

13:49

there's a lot of content going

13:51

into stuff that's to the core invidia, AI

13:53

AI accelerators. And ARM is absolutely benefiting

13:55

from that that hugely. But I But I

13:57

personally think arm's valuation is justified by more than

14:00

than just AI, it's this broad

14:02

expansion into so many other things. to

14:04

back up a little up a little

14:06

bit, I know there's

14:08

kind of this classic debate

14:10

amongst semi around risk risk and

14:12

these two and these two

14:14

fundamental architectural approaches in processor design.

14:16

It feels It feels important here. So

14:18

maybe we should just spend a

14:20

little bit of time explaining those

14:23

two and why two and why adopted

14:25

a risk approach a how that contributed

14:27

to its success. success. It's It's funny,

14:29

this risk versus CISC debate is something that's

14:31

been in computer science topics for a

14:33

long time. Back in the early

14:35

days of the this debate was incredibly

14:37

important. was I remember I took some

14:40

computer science classes in the I took some

14:42

and that was the topic of

14:44

the day. was 90s, and was is better, of

14:46

of famous moments around it. But

14:48

at heart, it's two ways to think

14:50

about how you architect a chip, it.

14:52

and at stands for to instruction set,

14:55

and chip, and risk stands for complex instruction set.

14:57

And the basic idea is you you have

14:59

a chip the the chip runs on

15:01

zeros and ones, a but a level

15:03

above that, are there are assembly language.

15:05

There's human readable code that the chip translates

15:08

into. zeros and into And the idea behind the

15:10

idea is that each of the critical mathematical

15:12

functions you want to implement in the you

15:14

you have a separate instruction for those. the

15:16

And as opposed to have a there's a

15:18

much smaller number of instructions you can give

15:20

the chip. And if you want to

15:22

do more complex a you have to string

15:24

a few of the simpler instructions together. So

15:26

the simplest way to think about this,

15:28

trigger the example, it's not quite perfect, but

15:30

quite perfect, but SISC architecture, you might have something

15:32

like a square root function, where it

15:34

calculates the square root. It knows how to

15:36

do that math and it moves to It

15:38

once knows to do the square root of a number. and it moves

15:40

you would have to do a

15:42

series of the subtraction, addition steps in

15:44

order to get that same command implemented.

15:47

and ones around to do at the time, used to

15:49

of religious about which one was better. And the

15:51

truth is, I don't think you could say that

15:53

one is better than the other. other. There are just

15:55

different situations when one works better than the other.

15:57

There's always trade is This is engineering. There's always trade

15:59

-offs between what is good. what's right for the application in

16:01

front of you. front of ended up happening

16:03

was risk, initially it was very, very

16:05

memory memory You needed a lot of

16:07

memory because you a to remember all those

16:09

steps. you had to the time, all more appealing

16:11

because it didn't need as much memory

16:13

and memory was very expensive. Over time,

16:15

need as -chip memory got very cheap very,

16:17

very quickly and that made risk much

16:19

more attractive. got very because it was

16:21

this reduced instruction set. more you

16:23

can say it ended up requiring less

16:25

power. That's oversimplifying it a bit, but

16:27

at the heart of it, you needed

16:30

less power to do these calculations because

16:32

the steps involved in the complex instruction

16:34

sets were complicated by design and those

16:36

ended up chewing up more power. up chewing up

16:38

more how that's how initial reputation for being

16:40

much more power efficient than power What

16:42

ended up happening in history is

16:44

x86 took the CISC path, the

16:47

complex path path, took the the risk-reduced

16:49

set set path. that's how they how they

16:51

back in the late in the early early

16:53

80s. And it's humorous to me that it's

16:55

still a topic that keeps coming up.

16:57

We have have the project, which is

16:59

very explicitly risk based. Coming on stream

17:01

today saying, hey, we we have all the

17:03

benefits of risk. And I think as

17:05

much as this is about low chip

17:08

interactions, what what it really, really what what

17:10

this difference between sys can risk

17:12

meant. risk meant, was risk-allowed users a degree of degree

17:14

of flexibility their in designing their chips. power

17:16

It wasn't just the power savings

17:18

that made risk appealing, it was much

17:20

more about the flexibility. I I need

17:22

risk to do something slightly different

17:24

than what what done. done. I'm gonna take

17:26

this other approach and I'm gonna

17:28

optimize for power savings, I'm gonna optimize

17:30

for this function and that function

17:32

and that flexibility was really important. Cause

17:34

one of the critical differences between

17:36

X86 and Because that X86 is owned

17:38

by two companies arm is that X86 AMD and

17:40

Intel. companies ARM is available to be licensed

17:42

by anybody. It's very, very hard to

17:44

license anybody to Anybody with enough money can

17:46

get an ARM license. an So then So

17:48

a step further. a step further, of the

17:51

interesting things about things about this five, this project, is

17:53

that it is free for anyone to

17:55

use, use. or at least to take the

17:57

code, code, so it's even cheaper and more

17:59

available. to ARM, and it and it is

18:01

by design even more flexible. You can

18:03

really shape this around and use RISC -V

18:05

in ways that you can't even do with

18:07

ARM, which itself is already pretty flexible. are

18:10

drawbacks drawbacks to that, don't have to get into

18:12

them. are software complexities down the stream

18:14

from that. down the stream heart, really,

18:16

the debate is around the debate of

18:18

instruction sets to customize to for

18:21

each individual chip design each licensee to

18:23

do as they see fit. see fit. It's It's

18:25

not every day that you

18:27

approach a business that trades at

18:29

30 time sales this has this

18:31

classic where there's this open this open free

18:33

to hypothetically free to use, nothing

18:36

free. is free. but I think before

18:38

we but I think before we

18:40

go into the competitive landscape and

18:42

the dynamics around the X86 and risk 5, I just

18:44

want to better I just want to better

18:46

understand what that partnership and licensing model

18:48

is really like. How How does that model

18:50

work going forward in a world where there's

18:52

so much competition? And what

18:54

are the key advantages to licensing

18:56

Apple and Apple and Qualcomm and Samsung

18:58

and others versus vertically integrating the

19:00

business in a way that could

19:03

help to maybe make their competitive

19:05

advantage more durable? This is a little is

19:07

a little complex answer. Let Let me talk about historically,

19:09

speculate on where things are going. where Historically,

19:12

it was essential in early days that they didn't

19:14

manufacture chips. chips. Part of the problem when

19:16

you manufacture a chip is a chip someone has

19:18

to do the design. That's expensive. You design. have

19:20

to hire designers to do that. that's what

19:22

to and NVIDIA do. to They that's do their chips.

19:24

And then once you've designed your chip, you

19:26

have to pay someone else to manufacture it,

19:28

typically. then So Qualcomm or Broadcom will pay

19:30

TSMC to do the manufacturing for it. So

19:32

that's a lot of or expense. will pay There's a

19:34

lot of capital involved. You have to pay

19:36

for a sets. You have to build inventory.

19:39

You have to take ownership of all those

19:41

chips. That's a lot of expense, inventory,

19:43

and and you have to take in the early

19:45

days, it just wasn't financially viable. viable.

19:47

It It was much simpler to just just their

19:49

IP. It was It was also something the market

19:51

really needed. needed. So I think that was driven

19:53

originally by necessity and the earliest CEO was dead set

19:55

on on this model. He He saw the opportunity

19:57

and grabbed it and really ran with it.

19:59

it. And then time emerged, you got this

20:01

big ecosystem of lots of other

20:04

customers, lots of licensees designing their

20:06

chips that really helped grow the

20:08

ecosystem. grow the there was a period

20:10

in the early in the there was a

20:12

big debate between big and and arm, And

20:14

this was around around mobile mostly, but other as

20:16

well, where where ARM was able to

20:18

succeed it it certainly did its own

20:21

set of R &D and innovation, but

20:23

but then they had hundreds of of licensees

20:25

are also innovating and exploring the market

20:27

and finding every segment and every

20:29

niche. and And that drove a huge

20:31

amount of volume of chip designs chip

20:33

X86, which was Intel and AMD. Big

20:36

big, capable companies, but but limits to

20:38

how much they could explore. So arm was

20:40

was able to, beyond the technical merits

20:42

and the power savings available and spones. They

20:44

had they had this giant ecosystem, which

20:46

was very, very powerful in expanding the

20:48

university dressable market. And then over

20:50

time, because of the rise of rises of

20:52

it also drove volume to the

20:54

foundries. It It really the the growth of

20:56

TSMC into the it is is today. you

20:58

have So you have this big ecosystem

21:00

that that worked really well in its

21:03

favor. now we're at a point we're at a

21:05

point today, the where the market has consolidated

21:07

a lot. There's always lots of competition

21:09

in semis. in The question is, is, does

21:11

want to build its own chips? it's certainly,

21:13

I certainly, I think, in a realm of

21:16

the possible where they're going to move up

21:18

a step and design their chips. I I

21:20

don't think they're going to go so far

21:22

as to necessarily put their own label

21:24

on a chip and design it and sell

21:26

it into the market like a merchant solution.

21:28

But I do think it's very possible

21:30

that they will take a lot of steps

21:32

to help their customers, their licensees, bring arm

21:34

chips to reality. a step a step in

21:36

here that's important. In between having ARM IP

21:38

and actually then sending it to the to there's

21:41

a lot of steps that have to

21:43

take place. take place. We the IP, taking

21:45

the IP from the digital files and good

21:47

ideas. good ideas, and translating that into

21:49

the design for a chip. That's a fairly

21:51

cumbersome process, and I think ARM is

21:53

now taking a lot of steps to make

21:55

that process easier, it, and I it. that I

21:57

think what that will probably do is open

21:59

the... door to new customers who may not

22:01

design chips today, but with ARM's help,

22:03

they can bring those chips to market

22:06

faster, better in ways that probably weren't

22:08

possible before just relying on merchant partners.

22:10

And that's sort of what we're seeing

22:12

with the hyperscalers. ARM has done a

22:14

lot to help those companies design their

22:16

chips, and I think they're going to

22:18

take a few more steps and get

22:20

very, very close to doing a full

22:22

design of their own chips. And when

22:24

you consider those steps to designing chips

22:26

and partnering with other technology companies that

22:28

participate in other parts of the ecosystem,

22:30

who else are you bringing together for

22:32

this project, then ultimately spec yourself into

22:35

something which I assume has extremely high

22:37

switching costs on a go forward basis?

22:39

So I think that the chief complexity

22:41

here is in the software realm, because

22:43

chips don't exist in isolation. You build

22:45

a chip so that it can run

22:47

some form of software. There are certainly

22:49

incremental steps in the design process, but

22:51

I think what's more critical than any

22:53

of that is the software that's going

22:55

to run on these chips. And a

22:57

big part of the ARM story of

22:59

the last decade has been the amount

23:01

of work that software companies are doing

23:04

to make their software run better on

23:06

ARM architecture. So in theory, you have

23:08

all these different ARM chips out there

23:10

in the data center. Like I said,

23:12

Amazon has one, Microsoft has one, Facebook

23:14

has one. There's a fair degree of

23:16

compatibility that if you have software that

23:18

runs on Microsoft's CPU, you can fairly

23:20

easily port that over to run on

23:22

Amazon. CPU. And that's the real critical

23:24

part of that. And there is a

23:26

whole history of how ARM got there.

23:28

It was a lot of work and

23:30

a lot of blood, sweat, and tears

23:33

that went into it. But it's at

23:35

a fairly advanced stage now. And I

23:37

think that's really critical. That compatibility layer

23:39

is very, very important and appreciates how

23:41

people think about what their choice of

23:43

design and ISA. So this business has

23:45

a very interesting history in the way

23:47

it's developed, but recent history has been

23:49

full of all types of dramatics. There

23:51

was the soft bank take private. There

23:53

was the Nvidia transaction. There was more

23:55

recent IPO. What was everyone thinking as

23:57

those things were going on from an

23:59

industry perspective? and also what is what is the

24:01

strategy behind the scenes why to why

24:04

these different parties were so interested in an

24:06

asset that at the time people thought

24:08

everyone was overpaying for but in retrospect,

24:10

seemingly they got a good deal. and

24:12

they got a good deal. So ARM about

24:14

arm years ago. years ago at the

24:16

time, there was a lot of head

24:18

scratching. They paid that paid 32 billion for it. it

24:20

I think many people, myself included, didn't

24:22

quite see it. it And then for a long

24:25

time, for most of a decade actually ARM

24:27

went to sleep. I think in part they had

24:29

done so well in mobile, they they were

24:31

the only in in town for mobile. got They

24:33

got acquired by at a a point in

24:35

which it was clear that ARM was gonna

24:37

be the only thing in mobile. thing in

24:39

wasn't gonna be there. All the last the last

24:41

tiny that had left over from the 80s

24:43

and 90s, they were all done. 90s, they were

24:46

all gone. had a monopoly on on chip

24:48

for mobile phones, and that was a

24:50

massive market. market. acquired them,

24:52

and they didn't have any pressure after that. They

24:54

just had to do what SoftBank asked, but they

24:56

didn't have the pressure of the public market asking

24:58

them quarter after quarter what's next, what's next. quarter

25:01

then at some point, what's next, what's investment fund. They

25:03

needed an exit. They exit. needed liquidity. it's NVIDIA came

25:05

along and offered to buy them for an billion,

25:07

needed which if billion, and a great deal. You paid

25:09

came you sell it and offered to especially because there

25:11

wasn't a lot of growth left if You didn't

25:13

really know what to do You it. didn't bank,

25:15

It's an interesting question paid to

25:18

why for $54, wanted to buy It's an

25:20

Certainly part of it was as to were in

25:22

the process then of designing their own CPU.

25:24

Certainly. they They saw this growth wave coming

25:26

for AI, knew that they and they knew that

25:28

they would need more powerful, more capable CPUs.

25:31

and they probably weren't particularly happy with

25:33

the pace of that arm, the sleepy the sleepy

25:35

company to delivering to them and they wanted

25:37

to accelerate work work ARM was doing

25:39

for data center workloads. Beyond that,

25:41

it's a little tough to see the that

25:43

invidia had in ARM. Nvidia didn't

25:45

play in mobile at that point. I

25:47

think it comes down to down to Huang,

25:49

the CEO of of invidia. His in my my mind

25:52

willing willing to make big bets, take big

25:54

chances. He's not afraid of failure. And as

25:56

And as much as he's this huge success

25:58

today, he's made a lot. mistakes along

26:00

the way. his real ability is his

26:02

ability to not be afraid of those

26:04

mistakes. to keep moving and to keep moving Most other companies

26:06

Most other companies do some of the things that

26:08

have over the years. They over the years. up, would give up.

26:10

give CEO would give up. The board would kick

26:12

him out. him That didn't happen kick him out, that didn't he

26:14

was able to keep pushing things through and making

26:17

these big bets. And I think ARM was in

26:19

that camp. It wasn't necessarily fully thought out. It

26:21

was a need, and they tried to acquire it.

26:23

And then when it didn't work out, they moved

26:25

on to something else. And then AI happened. we The

26:27

ARM we didn't, deal broke three years ago. we'd, And

26:29

the one good thing that came out of that

26:31

was, up and said, oh, wait a up and said, oh, wait

26:33

a second. need We've been sleeping on this. in need And

26:35

they company back in gear. so they brought in a

26:37

they knew they had to take it public. in

26:39

a new brought they new management team, brought in a

26:41

new CEO, brought in a new CEO, who has done a

26:43

tremendous job of reinvigorating the company. in a new waking

26:46

them up they starting to get them really

26:48

active they excited again, in, they pushing them. brought in,

26:50

He fixed all these things. He things. fixed

26:52

pricing, fixed over, and he fixed marketing, and said,

26:54

just go down the list. All these things

26:56

that hadn't been done said, they've tripped up

26:58

or overlooked, said, and he got them moving

27:00

again, headed in really good directions. And I

27:02

think at heart, what Arm is really trying

27:04

to do today is the the value that they

27:07

that they can get from the industry. if you

27:09

look at you look at it by

27:11

numbers, in 2023 on they went public,

27:13

on average, they earned about So they licensed the

27:15

IP, they get a So they payment, that they

27:17

get a royalty payment. cents That was

27:19

about on .07 per chip on average. age,

27:21

Today, it's over to nine close to rough math,

27:23

it rough math, probably gets to 12

27:25

cents over the next three, four years. years. And you think

27:27

about it, R makes, arm in the case of case

27:29

of invidia, a system

27:31

for a ,000. $100,000, arm probably makes a

27:33

buck or two on that. or two you look at it in

27:35

that context, you're like, well, maybe there's a little room

27:37

for a little bit more going well, maybe there's a

27:39

that across all the chips and bit more these

27:41

end markets. And you start to realize

27:43

there is this potential to greatly increase their

27:45

value in all these end And just to illustrate

27:48

how big that addressable market is and

27:50

how often we interact with their licensed

27:52

technology, can you just give an example

27:54

of where their chips are? like the In some

27:56

ways, it's like or inverse of whether

27:58

or not, is there seemingly everywhere. But I

28:00

just want to illustrate that point to better

28:02

drive home the market opportunity they have in

28:04

the event that they can push pricing further

28:06

they can push pricing much every electronic

28:08

device in your house in your house

28:10

content in it. in it.

28:12

that run on that will have

28:14

will have some arm cores somewhere in there,

28:16

maybe running Bluetooth or Wi-Fi or something.

28:19

Everything has ARM everything electronic. PC's smart phones, smart

28:21

home devices, your -Fi router, your

28:23

thermostat, your your lock. There are some

28:25

advanced Apple plugs. into plug it

28:28

into the wall. Some of those

28:30

have very low -cost ARM cores

28:32

in them to do something

28:34

who knows. your Your AirPods, all of

28:36

your speakers, in all of that

28:38

has ARM content in it. probably

28:41

refrigerator probably has some in it. arm Your

28:43

TV has lots of arm content in it.

28:45

cars. They have a little cars, they have a little

28:47

bit today. They're gonna have a lot

28:49

more in the future. Anything with

28:51

any digital smarts is gonna have some

28:53

kind of arm content in it. it. It's

28:55

It's really incredible. Part of this

28:57

conversation you obviously you the importance of

29:00

Jensen the his culture at and his culture

29:02

at semiconductor industry has become

29:04

has with all these

29:06

mercurial founders and executives.

29:08

have Lisa and Dee, Pat Gelsinger, who

29:11

who recently aside aside How Intel. is How

29:13

important is management to this business

29:15

they have they had as much

29:17

influence directly as some of these

29:19

other luminaries have in their particular

29:21

businesses? There's three three people I

29:23

think who deserve credit for arms success in

29:25

world. The first is Sophie Wilson, who

29:27

one of the founders of the arm

29:29

design team back when they were back

29:31

of they were still And I think she, more

29:33

than almost anyone else, really led

29:35

to the technical success that they had

29:38

initially. that they had them to

29:40

be appealing to Apple. them to And for

29:42

reasons, she didn't actually go to arm

29:44

when it got spun off. actually but she's the

29:46

often unsung hero of all this. the

29:48

The next important person would be Simon The

29:50

who was the founding CEO would separated

29:52

from Saxby, He had come in from, I

29:54

think, Motorola. He had this

29:56

vision of arm as an IP licensor and

29:58

he IP the... licensor, the business model and really

30:00

drove it to success. He's the one

30:03

who got them into their first deals, into their

30:05

first really them them to become where they

30:07

are today. them then the third one

30:09

is the current CEO, then the third I touched

30:11

on this a moment ago, where he

30:13

has reinvigorated the company and really positioned it

30:15

for its next big growth wave. and It's

30:17

a good case study in how management

30:19

can matter. I don't want to case any

30:21

of the past CEOs, even when they're I

30:23

of want they had a decent management team the

30:25

but it was task for different purposes. But

30:27

I think those three more than anyone

30:29

else have really been the ones who are

30:31

positioned, positioned to where it is. armed to where it

30:33

Given how pervasive arms technology is

30:36

throughout the semi the semi ecosystem combined

30:38

with their licensing a royalty model, I

30:40

I imagine it manifests itself in

30:42

an incredible economic engine. I I think

30:44

I read prior to this margin profile

30:46

is more more emblematic that software with

30:48

90 % gross margins, operating margins exceeding

30:50

40%. Can you just take us

30:52

through the business's financial profile and

30:54

maybe highlight the most important financial

30:56

KPIs? KPIs. We've been talking

30:59

a lot about abstract high -level things, about

31:01

technology. things about and their history, and their

31:03

but deep down they have this really

31:05

powerful economic model as well. as

31:07

On paper it looks a lot like a software company.

31:09

you have You margins in excess of

31:11

90%. of 90% that that flows through to

31:14

the bottom line and operating margins that

31:16

are 40, sometimes that are 40, That's very powerful

31:18

very powerful, because what is is happening here is

31:20

they have a big upfront cost. They

31:22

have to do the R the work. work.

31:24

to to keep pushing the technology forward. Gotta

31:26

pay to pay a lot of very smart computer

31:28

scientists and electrical engineers to keep coming up with

31:30

new advances, new products, new features. new

31:33

products, that's done, then you've spent all

31:35

that on R you've the cost of marginal

31:37

sales is essentially of while the sale is

31:39

are collecting a few pennies per they're And

31:41

it doesn't matter to them whether that's of

31:43

or a sales is cost them no extra to

31:45

sell that. And that all flows through the

31:47

bottom that It's a very, very powerful model. sell that,

31:49

to If you think about the leverage that provides,

31:52

to the royalty rate is gonna continue to increase.

31:54

to increase. over time over the next few years. years.

31:56

And even though though it's only a

31:58

few pennies, nominal increase in royalty. to all of

32:00

that flows through the bottom line. It's

32:02

just such a powerful amount of leverage

32:04

in this model. I do want to

32:06

caveat that a little bit. I say

32:09

it looks like a software model, but

32:11

it's important to understand that this is

32:13

not software. Arm is licensing IP. It's

32:15

not software. There's some pretty important differences.

32:17

If you think about things like you

32:19

can't patch this software website, website goes

32:22

down, you can patch it, you can

32:24

patch it, you can patch it, you

32:26

can patch it, you can patch, you

32:28

can't, Design process and so I don't

32:30

think you should think about it as

32:32

software It just has that sort of

32:34

economic model that looks very very similar

32:37

to software and over time I think

32:39

this model will evolve a little bit

32:41

Adopted what looks like a premium or

32:43

a premium model where they're using different

32:45

pricing schemes to get users in the

32:47

door again, it's not software, but they're

32:50

using a lot of those growth tools

32:52

that software companies use I think that

32:54

will slash out their sales profile their

32:56

revenue growth as well, but they're also

32:58

looking at new products talking about moving

33:00

up to stack, maybe getting very close

33:02

to building their own ships, that will

33:05

come probably at the price of the

33:07

cost of lower gross margin percentage with

33:09

the benefit of added gross margin dollars.

33:11

So that tradeoff's worth it, even if

33:13

percentage points come down to you, absolute

33:15

dollar pool gross. But for the most

33:17

part, I think they'll keep doing this

33:20

model and will continue to deliver these

33:22

kinds of results. And then I guess

33:24

the next five to 10 years are

33:26

going to be incredibly interesting for how

33:28

everything evolves. both in the electrification of

33:30

everything and the digitization of our entire

33:33

consumer economy. But there are risks to

33:35

this story. Growth won't necessarily run unabated,

33:37

although they've gone from one to two

33:39

to five billion dollars in revenue quite

33:41

rapidly. What are the risks to this

33:43

story? How real is that open source

33:45

competition? What is the decision tree from

33:48

here and where things could go? And

33:50

the best case scenario, the base case,

33:52

and then if things were to deteriorate,

33:54

what would have to happen? So I

33:56

think it's important to understand it, arms

33:58

product, instructions that architect... these ISAs are

34:01

are deep into chip deep,

34:03

deep and functionality. very hard

34:05

it's very, very hard to replace

34:07

it. We saw this a few

34:09

years ago, actually, when Apple moved

34:11

from Intel its own M series CPUs its own

34:13

Apple spent years preparing all kinds Apple spent years

34:15

preparing all kinds of software support

34:17

for that, preparing developers for that,

34:19

really put a huge effort into

34:21

getting people ready for that transition

34:23

because that transition essentially broke software

34:26

compatibility. So things written for written OS to

34:28

run X86 Intel Silicon wouldn't necessarily work on arm. And so

34:30

so Apple had to spend a

34:32

lot of money to make sure

34:34

that they did, at least at

34:36

some basic level, still function. still

34:38

today, if you're a programmer and

34:40

you're dealing in some newer stuff,

34:42

you download some new software in some

34:44

some new language, dealing still a

34:46

distinction. Are you running this on

34:48

Intel or are you running this

34:50

on Apple some newer stuff. point being that

34:52

it's very, very hard to replace

34:54

it. And that is an immense,

34:56

immense barrier to entry. entry. And we

34:58

look at there's this situation now

35:00

where ARM is suing Qualcomm, and

35:03

they've had and for years. Qualcomm's

35:05

probably one of ARM's biggest customers, one of

35:07

its biggest one of They don't like ARM, they don't

35:09

like being sued by ARM, but what choice

35:11

do they have? They They can't just go out

35:13

tomorrow and say, all right, we're gonna stop using

35:15

ARM, we're gonna switch to have? They It would

35:17

take them a decade to really work that through

35:19

their whole portfolio. using arm, we're

35:21

competitive threat that It would posed was to

35:24

a period through took its eye

35:26

off the ball, risk five was

35:28

able to capture a lot of

35:30

new growth opportunities. So for

35:32

instance, we see this big wave of

35:34

new chip companies coming up in China. That

35:36

That explosion took place in the late

35:38

when arm ARM wasn't quite ready for it

35:40

wasn't totally paying attention to their companies in

35:43

China today that are risk five five centered.

35:45

They're mostly doing embedded IOT low-value value devices.

35:47

. . that was the threat was the threat

35:49

is that risk five would get its foothold

35:51

and slowly work way up into other things. things. One

35:53

of the big changes that took place when

35:55

Haas took over as CEO is he fixed

35:57

pricing in such a way to make to make risk

35:59

five lesser. attractive in comparison to ARM. to Just having

36:01

having to to isn't the thing that's having famous

36:03

still a lot of work that has to

36:05

take place to design that that a real chip,

36:08

to harden that. And ARM has a big

36:10

advantage there. that. And fixing the

36:12

pricing and fixing some of the

36:14

licensing terms, the he greatly reduced of the

36:16

the appeal he greatly for all kinds

36:18

of use cases that otherwise of risk

36:20

5 for all kinds of use cases don't want

36:22

to say risk 5 is not a threat,

36:24

but I do think the degree of

36:26

threat is not huge right now. and if

36:28

it were, it would take years to develop.

36:31

I'm going gonna have my my friends get

36:33

angry at me, but I'm gonna say

36:35

that risk five is not ready for

36:37

data center workloads today. ready maybe a workloads

36:39

metrics, our how fast it goes, that

36:41

very raw it goes, that very raw it's comparable, maybe.

36:43

But in terms of everything else that

36:46

goes into it, the design it, the design

36:48

the hardening, the software, we got years

36:50

and years before risk five is really

36:52

ready for for let alone let center. data

36:54

Now, I could be wrong. It could

36:56

come sooner than we than There could be

36:58

some there could be some but I don't see risk

37:00

five today as five today as an to arm the way

37:02

you could have arm the way it to be a few

37:04

years ago. to be a back to

37:06

your question, So the base case is

37:08

they just continue to grow. They add value,

37:10

they capture more value, they get into

37:12

more markets. they I would say the upside

37:15

case is the they turbo is that. where

37:17

they really start to flex their

37:19

muscle. their in some they have a

37:21

monopoly on on ISAs. especially now that

37:23

they've blunted risk fodge In many senses, they're

37:25

not quite a monopoly a monopoly legally,

37:27

but functionally, they're pretty close. And

37:29

if they use that to increase

37:31

their value big degree and a big degree,

37:34

and they really flex their muscles

37:36

to you start to get significant

37:38

increases in royalty rates. that's certainly I

37:40

think that's certainly possible. just keep without that,

37:42

if they just keep going to the steady

37:44

incremental, slowly pace, I more value still I think

37:46

there's still a position to do really

37:48

well. But there's definitely a scenario where they

37:50

really turbo -charge that that. and it gets much

37:52

bigger than even these numbers would suggest. suggest.

37:55

The bare is, we just see a radical

37:57

shift in how we do compute. talked

37:59

talked a little. bit the tap between AI accelerator and

38:01

ARM based CPUs. Depending on who who you ask, it's

38:03

something like two to one to eight to

38:05

one today. If If it goes to like a

38:07

hundred to one, you where you really just

38:09

don't need that many CPUs. the arm market, the TAM,

38:11

the TAM, the adjustable market, is nowhere

38:14

near as big as we would have expected.

38:16

So growth growth plateaus at some point. how I

38:18

see I see the addressable market question is really

38:20

be focused on exploring the bear the bear

38:22

case. Just to go back to the economics,

38:24

we're talking about pennies per device and

38:26

billions of devices, some of which have some

38:29

of which the tens of thousands

38:31

of dollars of it relates

38:33

to automobiles, but also to automobiles, but

38:35

also of dollars. thousands of does the

38:37

contract work such that they

38:39

do, or or Steve a a de minimis revenue

38:41

per chip? chip? Is it crazy

38:43

to think that that can go

38:45

materially higher? I think the I think

38:48

the chief criticism you have leveled at arm or

38:50

15 years ago was that just kept prices

38:52

too low. too If you really want to

38:54

get into the mechanics of the model,

38:56

like I said, there's two components. There's

38:58

the I said, payment then there's the ongoing

39:00

royalty payment. And I think for a

39:02

long time, they moved towards favoring big

39:05

the payments at the expense of lower

39:07

royalty rates, which made sense as the

39:09

industry big and you started to have these

39:11

massive customers the you needed to fund

39:13

the R &D that those which made to advance

39:15

the ARM ecosystem, ARM architecture. and what what

39:17

ended up happening was was companies couldn't afford

39:19

arm licenses. arm So one of the things that's

39:21

been fixed in recent years is they've reset

39:23

all of that. recent in a

39:26

lot of ways, they have what are of

39:28

software like pricing tiers they where you pay

39:30

a little bit upfront like a premium model,

39:32

where you get limited access for a small

39:34

upfront payment. But then as you expand where

39:36

you you pay more and more. Essentially, it's

39:38

a lower upfront payment in exchange for

39:40

higher for a payment. I think this really worked

39:42

wonders. It's got a lot of people

39:44

in the fold. The thing is, if

39:46

you get people in the door and you have them

39:48

designing on payment, once you have that, they're almost locked

39:50

in. locked Like I said, it's very expensive to switch

39:52

to risk to switch to risk five. So lower the the upfront

39:54

cost significantly, get get more people in the door,

39:56

and then you make that up by having

39:58

higher royalty rates down the stream. I think

40:00

that's pretty much what they're moving towards. There's other

40:02

things they have going on as well, where they're moving

40:04

up the stack and doing more of the physical

40:06

design work. That's another good way for

40:08

them to increase their royalty rates. So

40:10

there's a lot of levers they can pull. And

40:12

then they've pretty successfully diversified

40:14

their revenue streams from one

40:17

that was primarily mobile to

40:19

now networking, auto, mobile, internet

40:21

of things, consumer electronics. But

40:23

are there any other areas or

40:25

market opportunities where they're under penetrated and

40:28

you feel like that story is

40:30

just taking off? Yeah, touched

40:32

on it a minute ago as automotive.

40:34

Automotive is the big opportunity. think that

40:36

most semiconductor companies are looking at today

40:38

where we can quibble over the exact

40:40

numbers, but there's a few hundred dollars

40:42

of semiconductor content in cars today, and

40:44

that's growing double digits every year because

40:46

just more and more things in the

40:48

car are becoming electronic, especially with electric

40:50

vehicles, but certainly all vehicles in general

40:52

you have more advanced digital cockpits, infotainment

40:54

systems, driving assist, and I'm not even

40:57

talking about autonomy. Autonomy is huge, huge

40:59

semiconductor content if and when it arrives.

41:01

But even before we get there there's

41:03

just layer after layer of more semiconductor

41:05

content in cars. and it's more compute,

41:07

and more compute means more arm. And

41:10

our typical question and conclusion is,

41:12

what are the lessons that you've

41:14

learned through your studies for ARM

41:16

that can be applied in evaluating

41:19

other businesses in the ecosystem? and

41:21

at the same time for operators, investors

41:23

in the space. or can be

41:26

applied to their businesses in order to

41:28

take some of these lessons from what

41:30

ARM has done, and a business model

41:32

that on paper sounds incredible and that

41:34

they've licensed pervasive technology that you would

41:36

apply elsewhere. The first

41:38

one is you need to invest heavily

41:40

in R &D. They spend 30

41:42

% of revenue goes into R &D. It's

41:44

expensive to do what they do. This is

41:46

a company with very high gross margins,

41:48

but also high operating expenses, because they need

41:50

to keep pushing the capabilities of their

41:52

architecture further and further afield. So one

41:54

of the areas where they under -invested, I

41:56

think, was AI, tensor cores, GraphicsCores, they

41:59

have... those products today, just not many people

42:01

use them because they just under -invested in

42:03

them in the last decade. And there's

42:05

definitely an alternate universe in which they hadn't

42:07

done that. They actually invested heavily in

42:09

AI and they'd be even more important today.

42:11

And we'll see. That's certainly another one

42:13

of those areas where I wouldn't count them

42:15

out yet. They still have some big

42:17

ambitions to increase their content in AI. But

42:19

investing in R &D, it's expensive. You got

42:21

to do it. It's lots of very

42:23

expensive talent required. But do that and do

42:26

that well. I think it's really important. The

42:28

other really important lesson for

42:30

ARM is having an ecosystem. as

42:33

a competitive advantage. And I don't think

42:35

they exactly set out to do

42:37

this, but in hindsight, it's just so

42:39

powerful, the fact that they have

42:41

hundreds or thousands of licensees all contributing

42:43

in some way, participating in building

42:45

up that ecosystem, building up the software

42:48

compatibility, making it more appealing, exploring

42:50

every market segment, every market niche. That's

42:52

been a huge, huge force multiplier

42:54

for them. So you need the R

42:56

&D to attract that ecosystem and keep

42:58

them engaged. Once you have that

43:00

in place, it's just immensely valuable. Well,

43:03

this has been a case study in

43:05

a business that clearly is on the

43:07

right trajectory. And in order to grow

43:09

into this valuation, we'll ultimately need to

43:11

sustain it for quite some time. Seems

43:13

like the pieces are all in place.

43:15

We'll see how the future unfolds here. Yeah,

43:18

I have to admit, I think they're well positioned.

43:20

It's fun to see I've followed this company for a

43:22

long time, and it's nice to see them really

43:25

hustling and moving again. Thanks so much. Thank you. To

43:27

find more episodes of breakdowns

43:30

ranging from Costco to Visa to

43:32

Moderna, or to sign up

43:34

for our weekly summary, check out

43:36

joincolossus.com. That's J -O -I -N -C -O

43:38

-L -O -S -S -U -s.com.