0:00

When I say we're going to move something

0:02

to cloud, everybody's like, yeah, I'm in cloud,

0:04

right? It's been there for so long, pretty

0:06

much everything at this point is in cloud.

0:08

So what's so special about silicon design? Now,

0:11

actually, this is one of those areas

0:13

which has not been moved to cloud. Welcome

0:15

to Technovation. I'm your host Peter

0:17

High. My guest today is Nishit Sahe. Nishit

0:19

is the chief information officer of Maraval Technology,

0:21

a data infrastructure technology company that earns roughly

0:24

five and a $5 and a half billion

0:26

dollars annually. His sheet has been in his

0:28

role for a bit less than two years

0:30

out of the more than seven and a

0:32

half that he's been with the company.

0:34

Across his tenure, he's laid out a

0:37

digital foundation for the organization, leverage data

0:39

as a catalyst for growth, help the

0:41

company digest a series of acquisitions, also

0:43

help the company develop a nimble orientation

0:45

that has it set to absorb continued

0:47

to grow the head, while developing a

0:49

compelling value proposition associated with

0:51

artificial intelligence. I look forward to hearing

0:54

more about these topics and others through this

0:56

conversation. Nesheed, welcome to Technovation. It's great

0:58

to speak with you today. Thanks for your

1:00

interest, looking forward to the conservation. I

1:03

am as well. Well, why don't we begin

1:05

with Marvell Technologies Business, Nesheed? Would you mind

1:07

taking a moment and providing a bit more

1:09

depth into the description of the business itself?

1:11

Sure. So, Marvell is a leading semiconductor company, right? I

1:13

mean, lately we knew the lot. So now, the question

1:15

is why are we knew so much. Now, we took

1:17

a position of being a data infrastructure

1:19

of being a data

1:21

infrastructure infrastructure infrastructure infrastructure

1:24

infrastructure infrastructure infrastructure. And

1:26

what happened, I mean, for so

1:28

anything about data, you think about

1:30

data, you think about data centers,

1:32

you think about automotive, carriers,

1:35

all of that requires data. But now

1:37

lately with AI, you know, we have

1:39

really come to being the top layers

1:41

in the AI infrastructure. So, you know,

1:43

playing alongside the big names that

1:45

you've heard in the industry, model

1:47

is supposed to be the top

1:50

five semi-unit company, which is actually

1:52

fueling this whole AI infrastructure.

1:54

Very exciting and remarkable accounts for some

1:57

of the growth that I described before

1:59

as well. you talk a

2:01

bit about your role as Chief

2:03

Information Officer, if you would? What's within

2:05

your purview? Yeah, so as a

2:07

CIO, I actually have three or

2:10

maybe four pillars that I have

2:12

in the organization. What is your

2:14

typical CIO role, right? You, you

2:16

know, support infrastructure, network,

2:18

business applications, all

2:20

sorts of business

2:22

applications, engineering applications

2:24

and whatnot. So that is

2:26

one, the typical CIO role. Then

2:29

the other one is the

2:31

data office. So Marvel has

2:33

a big data driven culture.

2:35

So our decisions are pretty

2:37

much data driven. So we

2:39

have a data office, you know,

2:41

which consists of all the elements

2:43

of data office, data science,

2:46

data analytics, data

2:48

engineering, data governance. I'm

2:50

responsible for the enterprise

2:52

AI, you know, implementation

2:55

that we're going after.

2:57

We've been on this journey for a long time

2:59

and it was a lot of the exciting pillars

3:01

that I actually spend a big amount of my

3:03

time lately on. And the fourth, which is

3:05

kind of goes hand in hand with

3:07

all of this, is business process transformation.

3:09

Many companies call it the digital office.

3:11

We don't have any office name to

3:13

it, but it's business process transformation. So

3:16

basically we look around in efficiencies in the

3:18

company, employee experience and whatnot. And we

3:20

go off to them and look at

3:22

various ways technology and solve those. And

3:25

your team does it reflect that

3:27

what you just described you have

3:29

a leader over each of the

3:31

areas you you just noted is

3:33

it is is it organized in

3:35

a different way entirely. So

3:38

the IT side of it

3:40

is fairly standard organization that

3:42

we have business partner model.

3:44

We're very, like, most idea groups, we

3:46

are a customer oriented. So we have

3:48

our leaders which are organized along with

3:51

the pillars that the company has. So

3:53

we have a person who's organized

3:55

or who's there to support

3:58

everything engineering, then everything. supply

4:00

chain. So we have it organized

4:02

that ways. We do have pillars for data

4:04

office for AI. But within the

4:06

team itself, we have this matrix

4:08

organization because AI, for example, you

4:11

can't just have one team do

4:13

all the AI in the company.

4:15

So if you're supporting finance, then AI

4:17

is one of the tools that

4:19

you want to bring in front

4:21

of finance to solve their business

4:23

needs. So that is a little

4:25

bit of matrix within, but yeah,

4:28

mostly the organization is aligned to

4:30

the business goals that I

4:32

defined. Very interesting indeed.

4:34

I mentioned the outset that you've

4:36

been in the process developing what

4:38

you refer to as a digital

4:40

foundation, and I wonder if you

4:42

could talk a bit more about

4:45

what that entails. Yeah. So, no,

4:47

again, as an IT person, as

4:49

a technology person, I need, I wanted

4:51

to get systems off for everything. you

4:53

do any function with the company, you

4:55

should have a system which automates that

4:57

for you. Now earlier, it was kind of

5:00

a nice to have thing. I mean, you

5:02

want to have an ERP system, you want

5:04

to have your collaboration or communication systems, but

5:06

pretty much beyond that, everything was optional,

5:08

right? I mean, you can do things in

5:10

Excel, too. Now, at Marvel, we are very

5:13

lean company for the revenue that we are

5:15

supporting, for the market side, we are supporting,

5:17

for the opportunity we are supporting. So

5:19

we were always very tech oriented, very

5:22

tech oriented. So we wanted to have

5:24

automation in every place that we can

5:26

do automation. So we were ahead on this

5:29

journey to have platforms and processes

5:31

which are fairly automated and can run

5:33

the business process faster. Now, there are

5:35

certain things that still blindspots

5:37

that many companies, especially old companies

5:39

like ours, have, right, where you,

5:41

where you're focused a little bit more

5:44

outcome based. You're more like, you know,

5:46

let's give out system for HR, let

5:48

out, give us systems for engineers, you

5:50

know, deaf platforms and what not. and

5:52

we stish through on the

5:54

backend. Now, with AI especially

5:56

coming into the mix, you

5:58

want to have... very strong

6:01

foundation. That is one thing. Second

6:03

thing is with all these

6:05

technology advancements and expectations that

6:07

you have, the workload size

6:09

improving, you want to have

6:11

good resiliency. So we are investing,

6:13

and of course, one key things that

6:15

are for security. I mean, we are

6:18

one of those target companies with heavy

6:20

IPs. We are always a target. So

6:22

looking at these three layers. which are

6:24

like more back-endish layers where the

6:27

architecture is in a perfect shape.

6:29

You have the end-to-end mapping done

6:31

well, all the technology that we

6:33

have in the company integrates

6:36

better. Below that is the right

6:38

infrastructure, a very resilient infrastructure,

6:40

something as simple as

6:43

having a perfect network

6:45

environment to move the data

6:47

across the company. And all of

6:49

this is properly secured with our

6:51

security there. Now... These are like

6:54

infrastructure in nature. Now the question

6:56

is when you stack up the applications on

6:58

top of it, how's the data moving? And

7:00

that is also one of the layers with the

7:03

data office that I talked about that we

7:05

are working through is do we have the

7:07

right governance in place? Do we have the

7:09

right controls in place? We have right compliance

7:12

in place. So that is also what I

7:14

call a very foundational work. The reason I

7:16

call it foundational is because normally people, I

7:18

mean if you're a user of a system

7:20

you don't see it. If I do my

7:22

job right. This is pretty transparent

7:25

to you. But at the back end, we

7:27

are stitching through multiple things to get

7:29

it together and what we are working

7:31

to us is a really clean environment.

7:33

Great overview. Thank you for that. You

7:35

mentioned earlier data office and I wanted

7:38

to delve a little bit more into

7:40

that. You've noted to me in past

7:42

conversations the methods you're using to leverage

7:44

data as a catalyst for growth. Can

7:47

you talk a bit about the methods

7:49

that you're using in order to do

7:51

so? I'll stitch two things. I mentioned

7:53

that Marvel is a very data-driven company.

7:56

So we don't have subjective conversations.

7:58

It's always in front. you come

8:00

and you come in with killer information,

8:02

somebody else comes with their information and

8:05

you didn't make a decision based on that.

8:07

So that has always been the culture of

8:09

the company. Now what we initially did

8:11

was to automate that culture, right? We

8:13

brought in analytics, we brought in information

8:16

in hands of decision makers so they

8:18

can quickly make decisions and form decisions

8:20

on the right data. So that was the first

8:22

layer that we draw and we call it

8:25

the data driven transformation journey. We have. good

8:27

analytics tools. I mean, we went

8:29

through a supply chain crisis in 2021,

8:31

22, then I invented a crisis in

8:33

22, and then 23, then again, I

8:36

supplied our demand crisis, you know, with

8:38

AI, later part of it, we

8:40

navigated very well due to

8:42

our data infrastructure. Now, what, again,

8:44

this was very outcome driven, this

8:47

was all about making the business

8:49

function well. Now that we are implementing

8:51

AI on top of everything, we

8:53

have to have the proper governance

8:56

in place. So we have done a

8:58

decent job on what you might call

9:00

the enterprise data, your ERP or sales

9:02

force, your work, the financial system

9:05

data. But when it comes to

9:07

the unstructured data, right, your IPs,

9:10

your technical documentation, they're all

9:12

pretty much living in their

9:15

isolated islands with not a lot of

9:17

reason for us to, you know, bring

9:19

into the center, but the AI, you

9:21

got to bring it to the main

9:23

place so that now you can feed

9:25

in that technology. and you get more

9:27

value out of it. So now we have

9:29

to put governance in places which are a

9:32

little hard to do. So we're

9:34

implementing a data mesh architecture where

9:36

you can put the governance at

9:38

rest. So before you actually bring

9:40

that in to derive information out of

9:42

that data, you know that you're compliant,

9:44

you're secure, it's clean. You have

9:46

50 versions of text specs to

9:48

pick from, which one are you

9:50

going to pick? So that is the journey

9:52

that we are on on the data set up

9:54

that. I get a great overview and indicative

9:57

of the creative thought process you with

9:59

the team. undergoing from that

10:01

perspective. Yours is an

10:03

organization that has grown luckably

10:05

through the course of your

10:08

tenure. Part of it through

10:10

major acquisitions, most recently the

10:12

infi acquisition, $10 billion acquisition

10:14

that closed in 2021, which

10:16

expanded the company's region data

10:18

centers and 5G network infrastructure,

10:20

for example. and then growth

10:22

beyond that as well. And

10:24

I wonder, first of all,

10:27

how you and the team

10:29

remain ahead of that growth.

10:31

The company you are today is

10:33

very different than the one you

10:35

joined. And likewise, with the continued

10:37

projected growth, just in a couple,

10:39

two or three years, it will

10:41

be a very different company again.

10:43

Talk a bit about how you

10:45

remain cognizant of this and ahead

10:47

of the growth so that you

10:49

can in fact effectively digest all

10:51

that growth accordingly. So Peter to

10:53

start, but the whole reason I joined

10:55

Marvel was because I know it's going

10:58

to be a dynamic company. It's a

11:00

company which has been here for a long

11:02

time when Matt joined the CEO. Matt came

11:04

for Maxim. I was working at

11:06

Maxim. It was very clear that

11:09

Matt is going to bring transformation

11:11

to the company. The key thing was to

11:13

define what this company is, right? So

11:15

if you look at, even go seven

11:17

years ago, we took the path of

11:19

being a data infrastructure company.

11:21

And we did wherever the market needs are.

11:24

That's what we focused on. So, you know,

11:26

we're a big automotive, I mean,

11:28

beyond AI itself, we're big in

11:30

automotive, we're big in carrier, everything

11:32

and anything which needs data. Now

11:34

the question was, you have these, you

11:36

know, what about being a storage company,

11:39

initially that was what this company was

11:41

found for, what are the different, you

11:43

know, pieces that are missing? And that's

11:46

what we filled through the acquisitions. Now

11:48

that is from the strategy

11:50

perspective. Now the key thing is, I mean,

11:52

a lot of companies make conclusions. Now you have

11:54

to make it work. And there are a couple

11:56

of elements to it, right? You have to have

11:58

the right strategy, so math. and his staff

12:01

has always had a great strategy

12:03

when it comes to the product,

12:05

very market focused, long-term

12:07

strategy. The second is

12:09

cultural. So when you're bringing

12:11

a company, can you integrate

12:13

that company culturally and Marvel

12:16

is in a phenomenal job? And the

12:18

third is what I, when I come in,

12:20

your systems and processes have to

12:22

just work. And what... What we

12:24

had perfected over, you know, five

12:26

major acquisitions that we did was

12:28

how quickly you can do it,

12:30

how cleanly you can do it, and when

12:33

you come in, how do you support the

12:35

top two layers that I talked about? So

12:37

the system side of the process

12:39

side is to be completely seamless

12:42

for everybody who's, you know,

12:44

working, you know, post acquisition within

12:46

the company. Speed was in

12:48

point. So, you know, sometimes people

12:50

like... Why do you need to

12:53

integrate this fast? When I talk

12:55

about speed, infi was integrated.

12:57

And when I say integrated,

12:59

their supply chain, their

13:01

CRM, their portals, all

13:03

of this was integrated

13:05

within two months of the

13:07

due process. Now that with 99.6%

13:09

plus accuracy in data. So

13:12

it's not like, you know, you

13:14

can all the shaven data in

13:16

your system. and integrate, right? But

13:18

we maintain the quality of our

13:20

systems after integration. At the same

13:22

time, we pretty much kept most

13:24

of the automation that the incoming

13:26

company conscript. So it's not like,

13:29

you know, this is the system, you

13:31

use it, this is what you have, right? And

13:33

that required quite a bit of collaboration.

13:35

So this is where the culture of

13:37

the company came into the mix. There's

13:39

not just IT coming in and solving

13:41

this integration problems. It's actually, it's a

13:43

site to behold, right? The whole company

13:45

comes together and decides, okay, this is

13:48

how we're going to do our part

13:50

and make this a success. Once we've

13:52

decided the goal post that, okay, you

13:54

know, March or April is when we

13:56

be integrated as one system or process,

13:58

the whole company came together. make it

14:00

happen. Fascinating. A remarkable

14:02

what you've been able to accomplish

14:04

from that perspective. The pathway to

14:07

making an acquisition accretive is no

14:09

doubt enhanced by the degree to

14:11

which you can successfully integrate

14:14

those organizations as you've described.

14:16

Another topic that you and I

14:18

talked about that your team has been

14:20

integrally involved in has

14:22

been helping move silicon design to the

14:24

cloud. And I wonder if you could

14:26

talk a bit about that and the

14:29

advantages of doing so. Of course, the

14:31

methods that you and the team have used

14:33

as well. To start with, when I

14:35

say we're going to move something to

14:37

cloud, everybody's like, yeah, cloud, right? It's

14:39

been there for so long, pretty much

14:42

everything at this point is in

14:44

cloud. So what's so special about

14:46

silicon design? Now, actually, this is

14:48

one of those areas which has

14:50

not been moved to cloud. pretty

14:53

much every semiconductor company does it

14:55

on prime. And the reason is, you

14:57

know, when you do silicon design, you

14:59

have a lot of legacy flows which

15:01

the data flow is not meant for

15:04

what other cloud should be, right? You

15:06

can't have, you can't have a

15:08

storage a mile away from your

15:10

server because you're moving data in

15:13

and out very fast. So architecturally,

15:15

the cloud was not meant

15:17

for silicon design type of

15:20

programs. Now what What we did

15:22

was we actually partnered with the

15:24

hyperscalers. And we went in and

15:26

we said, your architecture doesn't work

15:28

for us, but there's value, and I'll

15:31

come to the value part. And to their,

15:33

you know, we recently announced our partnership

15:35

with ALEs, right, to their credit,

15:37

they re-archated their solution to make

15:40

it work for Cloud. Now, comes to

15:42

the value part, why cloud, right? So

15:44

one is obvious, one, you don't want

15:46

to have data centers, and you, you

15:48

don't want to have data centers, But

15:51

the other piece is what you can

15:53

do on premises, what you

15:55

can do cloud in terms

15:57

of basic infrastructure. Now, silicon.

16:00

design is a very complex

16:02

workload. Think about the size

16:04

of logic that we deal with.

16:06

You know, we talk about two

16:08

nanometers. I mean, human mind cannot

16:10

perceive beyond a certain

16:12

dimension. When I was doing double

16:14

E, I was told a long time

16:17

ago, I was told like

16:19

100 nanometers where physics wouldn't

16:21

allow you to go any

16:23

further down. And we had

16:25

a two nanometers. So when

16:27

you're designing such complex dense

16:29

circuits, you have to have

16:31

your workloads, your tools that

16:34

run those workloads. They're extremely

16:36

complicated to run.

16:38

So we have this huge

16:41

high performance servers which are meant

16:43

for these tools to work on

16:45

a data center. I think about

16:47

how a data center is, right?

16:49

Now, it's like a series of

16:51

these servers running through. Now,

16:54

the way Silicon Design has

16:56

done now is everything runs

16:58

on a CPU. light workload

17:00

very dedicated workload whether it's

17:02

a very logical workload right

17:04

a lot of math involved

17:06

or it's a simulation everything

17:08

just runs on a CPU now what

17:10

made a I happen this GPU made a

17:13

happen right so when there's certain

17:15

type of workloads which are meant

17:17

for GPUs there's certain type of

17:19

workloads which doesn't need that kind

17:21

of performance which requires a

17:24

dedicated or very customized type

17:26

of setup for them so This is what

17:29

you call customizable compute. You

17:31

can just take a workload and you can

17:33

say, okay, this one, I'm going to run

17:35

on a CPU because it requires a

17:37

lot of math, very high performance math,

17:40

and CPU are perfect for that.

17:42

But there are workloads, which has

17:44

simulation in nature. So you want to

17:46

run it on GPU. I mean, you can literally take

17:48

four weeks workload and make it a

17:50

two-day workload, or maybe a few hours

17:52

workload. If you run it, try it

17:54

on GPU. This is what GPUs are

17:56

for. But there are some places where you want

17:59

to save on power. Because both

18:01

CPUs are tremendously power hungry

18:03

systems. CPUs are fairly in

18:05

high performance, CPUs are fairly

18:07

power hungry systems. But then you

18:10

have arm-based systems, which are very

18:12

easy on performance, take good on

18:14

power, very good on performance for

18:17

certain workloads. So you can take your

18:19

full flow and say, okay, this job goes

18:21

to arm, this job goes to CPU,

18:23

this job goes to GPU, and you

18:26

can have the most optimized workflow. Now

18:28

this is something that you don't want to

18:30

do on track. Can you imagine maintaining that

18:32

kind of ecosystem yourself? I mean, this is

18:34

not the business that I'm in, right? Now

18:36

that is something you can do in cloud.

18:39

So the first big benefit of cloud that

18:41

we saw and that we're aiming towards is

18:43

this customizable compute. We are bringing all

18:45

the ADA partners with us. Because it's not

18:47

just us talking to AWS or the hyperscalers

18:49

who can make an app and they are to

18:51

adjust their technology to work with customizable

18:54

compute. The second bit is

18:56

all these new tools that our engineers

18:58

are going to get. So cloud, you

19:01

know, talk about the AWS platform,

19:03

you get a lot of microservices.

19:05

Now we can activate those on-prem

19:07

too, but again, you can imagine

19:09

the headache that you had to do

19:11

and the time it consumes. Again, you

19:14

want to get your services fast to

19:16

your engineers. So cloud gives you

19:18

that new system. And the third back

19:20

to AI. Pretty much all AI exists

19:22

in cloud. Right. There are very few

19:25

companies, hopefully, you know, not one Z2Z.

19:27

Even though the companies who are actually

19:29

doing on RAM are mostly the companies

19:31

who have built GPUs. So beyond that,

19:34

pretty much everybody is using cloud

19:36

all the solutions that are in cloud.

19:38

So if you are to enable all

19:40

these air technology for engineers cloud of

19:42

the right place. So that's why you

19:44

want to move our design industry cloud.

19:47

We have started that full pace. We had

19:49

announced this big partnership with

19:51

AWS and that's the journey we are on.

19:53

Again, there are a lot

19:55

of companies in silicon design.

19:58

It's not like cloud. exist,

20:00

but most people use it for

20:02

burst capacity. So you still have your

20:04

data center as a primary workload driver,

20:06

but if you need more you just go

20:09

to cloud. So that's the first thing.

20:11

So we are taking that approach. We

20:13

are actually moving our entire floor to

20:15

club. So that is very unique. The

20:17

second thing is Marvel being, you

20:19

know, the kind of company we are,

20:21

we are in two nanometers, three nanometers,

20:23

highest end nodes. Our workloads are

20:26

way more complex than the

20:28

traditional workloads that other semiconductor

20:30

companies have. So A, we are actually driving

20:32

the cloud limits or cloud performance to its

20:34

limits. So they are working very closely with

20:37

us to make sure that the architecture is,

20:39

you know, can support us. And the second

20:41

thing is, ours is not like, okay, you know,

20:43

if you need excess capacity, you'll go there,

20:45

we're actually going the whole thing to cloud.

20:47

Very, very interesting. I appreciate you

20:50

sharing those perspectives. makes a lot

20:52

of sense as you describe the rationale

20:54

and the sort of steps that you've

20:56

undertaken in order to bring that to

20:58

life. I wanted to talk a

21:00

bit about artificial intelligence, which we've

21:03

kind of mentioned at the edge

21:05

of the conversation so far. As

21:07

you noted at the outset, your

21:10

responsibilities include the internal use of

21:12

artificial intelligence writ large. And I

21:14

know from again our past

21:16

conversations that there are four

21:18

areas that you're focused on.

21:21

relative to AI, you noted

21:23

those as moving the needle

21:25

on product development, overall company

21:27

efficiency, change management, and long-term

21:29

investments. And I wonder if you

21:31

could take a moment and describe a

21:33

little bit each of those just to

21:35

provide a bit of background and color

21:38

to what each is and what's intended

21:40

by each. Sure. Yeah. So we've been

21:42

on this AI journey for almost two

21:44

years now. That is, and before that

21:46

we were doing data science stuff, which

21:48

is just not generitive

21:51

way in nature. But then the

21:53

question was, where do we

21:55

invest? Where do we make our

21:57

big bets? You can always

21:59

buy. you know, of the shelf AI tool

22:02

and deploy for the company. The question

22:04

is, does it add value to the company?

22:06

And that is when we came up with

22:08

this four pillar approach. So semiconductor

22:10

projects or new product development

22:12

takes a tremendous amount of

22:15

time. Like I mentioned, it's very

22:17

complex flow with a lot of engineering,

22:19

a lot of science behind it, a

22:21

lot of use for AI within it. So the

22:23

first pillar that we picked or the main

22:25

pillar that we picked where you want to

22:27

make big debts. is to improve our product

22:30

development. I'm using the word improve and

22:32

not shorten the life cycle, because shortened

22:34

is obvious, right? I mean, you can

22:36

take efficiency or make things more efficient,

22:38

and it's not just AI, you can

22:41

do multiple different things to make your

22:43

flows efficient, and that is great. If

22:45

you can shorten your time to market, that

22:47

is perfect. But at the same time, you

22:49

can use AI to improve our feature sets in

22:51

the product. So within the product development

22:54

itself, we're looking at various vectors

22:56

on where we can actually bring

22:58

AI. to do big things. Now, generative

23:00

AI was a catalyst, but when

23:02

we actually went into the flow,

23:05

we realize a lot of places

23:07

traditional machine learning can actually be

23:09

leveraged, I mean, to make things

23:11

better. So that is one pillar.

23:14

Again, think about it, you know, if

23:16

we make it then there, the kind of

23:18

products we build, they're very

23:20

expensive, I end, take long time

23:23

to build type products, and if

23:25

you can make it faster, and

23:27

we can make. our product features

23:29

better than company wise, the kind of

23:31

company we are in or the kind of

23:33

business we are in, that will be a

23:35

huge impact to the market and tomorrow.

23:37

So that is our primary focus.

23:40

But at the same time, you want to

23:42

make sure that the whole company get

23:44

the efficiency out of this. So the idea

23:46

here is while we are

23:48

focusing on product development, let's

23:50

also focus on efficiency, right?

23:52

The whole company's efficiency. And

23:54

there you can look at

23:56

the corporate wide efficiency, you

23:59

know, what else? things that take time

24:01

for people. One common thing for us

24:03

is searching information. So you gave

24:05

out a system for people where

24:07

they can quickly search for information

24:09

within the company like a Google

24:11

platform or chat DVD for the

24:13

company itself. Then we are going

24:15

into each business functions and working

24:17

with the business process owners to

24:20

see how we can reimagine their

24:22

business functions with AI. And there are quite

24:24

a few cases that we actually went in

24:26

and said, instead of doing the process this

24:28

way, let's do the process in a

24:31

different way and plug in these technologies

24:33

to enable that process. So that

24:35

is basically going after not

24:38

just overall employee efficiency, but

24:40

business process efficiency, which ultimately

24:42

brings efficiency to the functions,

24:44

right? So that's the second pillar. So

24:46

that's very ROI base. That is. You look

24:48

at the time mapping and you

24:50

can do those things, whichever traditional

24:52

type of approaches on making your

24:54

things more efficient. Now, you can

24:57

build great tools. You can deploy great

24:59

tools. Now, AI even now is an optional

25:01

tool for you to use because we're not

25:03

building autonomous tools, right? It's not

25:05

like, and by the way, at

25:07

this point, AI tools are not

25:09

ready to be autonomous. It's an

25:11

assistive technology. I'm an engineer, I'm doing certain

25:14

things, I'm going to give, you know, so

25:16

I deal, or my team will give all

25:18

these tools, which will make them faster,

25:20

or they make them more efficient, but

25:23

they are the masters, right? They

25:25

are the ones using these assistance,

25:27

deal assistance that we're giving out to

25:29

there. Now when you do that, it's again

25:31

a very optional thing for them to

25:33

use. Now we want to encourage people

25:35

to use it. So that's why the

25:37

change management pillar came into the

25:40

being to the being. And it goes

25:42

from like people having basic

25:44

awareness of the technologies as

25:46

we're coming out at the same time,

25:48

you know, knowing how to use it. These tools

25:51

are as good as you use them to

25:53

be, right? I mean, think of something

25:55

as simple as co-pilot. Oh yeah,

25:57

you can do basic chat with it, but

25:59

then. the more you learn from that tool

26:01

to ask the right level of questions, the

26:03

more useful these tools are. So there's a

26:06

lot of training involved, a lot of awareness

26:08

involved, we're doing a lot of these,

26:10

what we call AI days, where we're

26:12

actually going into different sites and we're

26:14

bringing to get people together. We're building

26:16

this team of AI champions who are

26:18

belonging to business functions, belonging to these

26:20

sites, who make sure that people are

26:22

aware of what's happening, and there's the

26:25

right amount of usage in there. And if there

26:27

is an idea coming in. then it can come

26:29

to my, you know, our core group so we

26:31

can work on it, all the good ideas.

26:33

So that is the change management

26:36

side of it. Again, the one thing

26:38

to consider here is, you know,

26:40

these technologies are not like

26:42

your typical usual systems that

26:44

you used to have, right, where

26:47

you actually ask a specific

26:49

question and you always get a

26:51

specific answer. In AI tools, based

26:53

on how you ask a question,

26:55

you answer could be completely different.

26:57

So it's a very different type of

27:00

technology. It requires a rewiring of

27:02

a lot of people's thought process

27:04

on how to use a system. That's the

27:06

third pillar. The final is, or the fourth

27:08

pillar is about, okay, what do we do

27:10

long term? All of this are building technologies,

27:13

buying solutions, deploying them, training

27:15

them. That's like, you know,

27:17

every month, every week, we actually

27:19

come up with a plan, we

27:21

refine the plan. That's very short-term

27:23

nature, right. Long-term what do we

27:25

invest in. And we talked about that a

27:28

little bit. The first one is data, because

27:30

the key thing that fuels this

27:32

AI is your own company's knowledge,

27:34

your own company's data, your own

27:37

company's information, still have a right

27:39

data platform, which can be leveraged

27:41

by these AI solutions, was kind

27:43

of paramount for us. And that's

27:45

where we set up a data office,

27:47

brought in data governance team, you know,

27:50

formalize the data science team, brought

27:52

in some technologies to actually. created

27:54

data mesh in modern that I

27:56

talked about. So that is, and that's an

27:58

ongoing thing. Have the right. the data

28:00

controls. Think of it, your data might

28:03

be fully secure access controlled in

28:05

their own islands. Once you bring

28:07

that into the AI ecosystem, that

28:09

it's very hard to control the

28:11

access of data, you know, to understand

28:13

what the quality of that data

28:16

is. So we bring in, we're

28:18

building that infrastructure team to support

28:20

awareness and bring, you know, building

28:22

data stewards team to make sure

28:24

that our business understands the importance

28:27

of data. We are building

28:29

a relationship with a lot

28:31

of solutions providers as well

28:33

as service providers because there's

28:35

not something that we want

28:37

to do ourselves. We actually want

28:39

to build an ecosystem around

28:41

us who can work with us to

28:44

either build solutions or support solutions

28:46

or deploy new tools for Marvel. So

28:48

we're investing with some set of companies

28:51

where we either bring them in to

28:53

work with us. or we actually engage with

28:55

them on a regular basis to understand

28:58

where they're going. And the third is we

29:00

are also engaging with the universities. That's why

29:02

a lot of research has been happening. And we

29:04

want to make sure that we are aware of

29:06

those research where, you know, we are getting ourselves

29:09

ready for all of that. And yeah, that's why

29:11

the university engagement. And we have chosen

29:13

like two or three nurses that we

29:15

can engage more intimately with. And then

29:17

I hear echoes in what you've

29:19

described in something else you've mentioned

29:21

to me, which is the notion

29:23

of democratizing artificial intelligence to a

29:26

greater degree. I can, I can,

29:28

I can, I can hear the rationale, even

29:30

some of the methods that you're using

29:32

in order to do so, right? My

29:34

drawing of the right connection there. Oh

29:36

yeah, no, totally. Actually, that's a good

29:38

point. So one of the things that

29:40

we worked on very early was, you know,

29:42

how do we, how do we get

29:44

our marvels, engineering strength to play, So that

29:47

too, one is like we're a company of

29:49

engineers and we build machine learning

29:51

products. So we have a lot

29:53

of that knowledge setting with thousands

29:55

of people here. So we don't

29:57

want to have these 10 set of people.

30:00

are the only ones responsible for building

30:02

any AI solution in the company. So

30:04

we actually wanted to, we have wanted

30:06

to encourage our engineers to work with

30:08

us on building these systems. Now

30:10

the problem with that is then

30:12

you'll have proliferation of systems, you'll

30:15

have a same amount of systems, right? So

30:17

what we did was it built an

30:19

AI development platform which is supported by my

30:21

team, you know the right data is getting

30:23

in the like, you know, all the rest

30:26

of it legal compliance security, all of it

30:28

architecture. And we give them this

30:30

access and training so that they

30:32

can come in and build their

30:34

own applications. So at this point,

30:37

in fact, we have around 40

30:39

plus applications which are

30:41

live. And I would say 60% of

30:43

it is built outside IT. It's built

30:45

by these engineers. So that is

30:48

one. The second thing is

30:50

democratizing the thought process

30:52

on AI2. So we are not

30:54

domain experts. We are technology

30:57

folks with IT, right. We want

30:59

to make sure that the knowledge on

31:01

how the ideas on how AI can

31:03

be leveraged in a particular business function

31:05

can come from the domain experts.

31:07

They're the ones who really

31:09

really understand how the depth of their

31:11

function works. So that is the other approach

31:14

that we had with AI champions

31:16

coming in for various business functions

31:18

that they're representing and they are

31:20

bringing the ideas to the table

31:23

too. So both building as well as

31:25

idea generation. That's

31:27

exciting to hear about how pervasive

31:29

this is this has become. I

31:31

really appreciate you sharing each of

31:33

those issues. I wanted to also

31:35

ask you more generally speaking as

31:37

you look ahead. Any other trends

31:40

that particularly excite you

31:42

that are making their way on your

31:44

roadmap? Yeah, because yeah, so exactly funny.

31:46

Yeah, I just started. So I think

31:49

one thing that excites me the most

31:51

and is quantum computing and the way

31:53

I see it is, you know, you

31:55

had GPS could use. doing something very

31:57

different. And then there is all set.

32:00

of problem statements that needed GPUs

32:02

and CPUs are meant to solve.

32:04

And there's a technology which seems

32:06

pretty promising, which is actually coming

32:08

to Christian lately. Now, there are

32:10

various different thought process on, you know,

32:12

the practicality of it and how fast

32:15

this will be coming to the market

32:17

as a enterprise solution. But keeping that

32:19

aside, that is one thing that excites me

32:21

the most, and sometimes scares me the most.

32:23

By the way, I don't think about it,

32:26

all the security solution that you put in

32:28

mind. is meant to deflect, you know,

32:30

a processing speed of a GPO or

32:32

a CPU. Quantum computers

32:34

will have very different

32:36

realm altogether. So you have to

32:39

reimagine quite a few things

32:41

the way we do things, especially

32:43

to protect our environment with the

32:45

quantum computing coming in.

32:48

And the other question is, how do

32:50

we integrate that to our

32:52

infrastructure eventually? And what kind

32:54

of business problems it can

32:56

solve, right? So technology excites me.

32:59

I've been waiting for this to

33:01

come to life for a long

33:03

time and the possibilities are fairly

33:05

limitless with that and of course

33:08

the risks are fairly limitless too.

33:10

There are a lot of similarities in

33:12

terms of AI and quantum and as

33:14

much as the more advanced versions of

33:16

these was expected to be many decades

33:19

out and it seems like both of

33:21

them are getting pulled forward so dramatically

33:23

leading to a lot of excitement but

33:25

as you point out also a necessity

33:27

to. think more today about the risks

33:29

that maybe it may appear in the

33:32

not too distant future as well. I

33:34

appreciate you who's sharing some perspectives there.

33:36

And the sheet, thank you so much

33:38

more generally speaking for talking a bit

33:40

more about your tenure at Marvell, your

33:42

tenure as Chief Information Officer. the remarkable

33:44

progress that you and the team have

33:47

made in with an organization that's been

33:49

growing so dramatically, as well as a

33:51

number of trends, perhaps primary among those

33:53

centered around data and artificial intelligence, that

33:55

you're helping ensure the organization is taking

33:57

full advantage of. It's been a great conversation.

34:00

region. See you are feed it.