娇色导航

Hello. Good afternoon, and welcome to the new season of 娇色导航Leadership Live. I'm your host, Maryfran Johnson, CEO of Maryfran Johnson Media, and the former editor in chief of 娇色导航magazine.

Since November 2017 this video show and podcast has been produced by the editors of cio.com and the digital media division of Foundry, an IDG company. Today's episode is sponsored by Broadcom, a global technology leader specializing in semiconductor, enterprise software and security solutions, headquartered in Palo Alto, California.

Broadcom's Enterprise product portfolio addresses several critical markets, including cloud data center, networking, broadband and wireless connectivity. Broadcom's VMware division, which was acquired last year, also offers a comprehensive private cloud platform called VMware Cloud Foundation. To learn more, please visit vmware.com and now onward to today's guest.

I'm very pleased to be welcoming to leadership live. Alan Davidson, who is the 娇色导航of the global technology organization at Broadcom. He joined the company in 2016 as its director of IT infrastructure, and was then promoted to 娇色导航in 2020 Alan reports directly to the CEO.

His broad range of responsibilities encompass the strategic direction and evolution of the company's technology landscape, as well as its e business application suite, hardware and software product, hosting and the security operations.

Prior to his current role, Allen held a variety of key leadership roles in three of broadcom's legacy companies, Hewlett, Packard, Agilent and Avago, where he oversaw everything from enterprise infrastructure and data center management to software product hosting, SaaS operations, customer support, solutions and more. Alan, welcome.

Thanks for joining me here today. Thank you. I didn't realize I was that busy, but yes, thank you. All of a sudden, you're feeling a lot busier than you were feeling an hour ago.

All right, I want to start out with a very kind of a high level perspective on the most notable IT industry trends that you've seen impacting broadcoms business and customers, not probably not just during your last couple years as CIO, but with your long tenure in the industry itself, you've had kind of a catbird seat on all the different changes that have taken place in IT leadership.

So give me some perspective on what is emerging now that you're hearing a lot about from your own 娇色导航customers. So you know, obviously there's the there's the AI elephant in the room that's certainly front and center of the topic of the day.

But I you know, there are a variety of trends that are happening, both in the SaaS application area as well as the infrastructure area. And there's always the perennial, you know, where do you run your workloads?

Do you run it in public cloud versus private cloud, and then the adoption of the different technologies. So I think the way we are, we're tackling that really, is looking at that from a workload characterization perspective.

So really, you know, we view the only thing that's constant over the multiple years that I've been doing this is business problems are still business problems. So business fundamentals don't change. Technology evolves, typically, every 3, 5, 7, years, whether it's hardware or software too.

So how you solve those problems may may evolve, hopefully get easier, simpler, quicker.

But I think the the variety of technology today is something we I probably couldn't even have imagined even 1015, years ago, in terms of just the capabilities where, as well as the public cloud and obviously the private cloud will bring to that, that landscape, I think the other key challenge really is, is the regulatory space is significantly evolving a high degree of pace, where you know the problems of two three years ago are certainly not the problems of Today, and how you address them are going to require a higher degree of discipline, both in your your data modeling, your data privacy, your localization.

So it really focuses on the data and how you protect that. And then obviously the third one is, is security and. No real change. Everybody is still healthily paranoid around the level of both external as well as internal threats.

As we are expanding our employee base, as we're expanding our breadth and depth that becomes a bigger issue as we go forward, just to make sure that, you know, we're protecting the right things at the right time.

So you know, security, the regulatory stuff, but I think you know, the more interesting spaces is technical evolution and how it really addresses business problems. So I think, like I say, the AI mantra, I would say in the media, has been very well publicized.

I don't like the word to use is hype. There is some reality behind it, in terms of the technical capabilities, but I go back to my business problems to solve. So what problems are you really trying to address with that type of technology?

So some of that may be, I don't think there's probably a good easy button for it in terms of, hey, it's a panacea to solve all problems, but I think it can have some astonishingly good results when applied directly.

So those are the areas that we are focusing on. I think how we address them is going to evolve over the next, you know, one to two years.

You know, as you you have stated, We acquired VMware last year, which has been an interesting year that's kept us slightly busy.

So you know, one of the, one of the key foundations of that, obviously, as you mentioned, is the VMware Cloud Foundation, or VCF, that we are implementing as part of our private cloud offering. So again, you know, we're really taking a look at our cost portfolio.

We are highly cost sensitive. As you you know the lineage of Broadcom. It has around 26 different business units. 17 of them are hardware, nine of them are software or infrastructure software. So the lineage of Broadcom really came from the hardware side of the business.

So again, there is definitely a DNA of focusing on margin, focusing on cost effectiveness, because with a maniacal focus, I would say on that, and we've really tried to install that DNA in the software side of the business as well.

So as we go into managing the breadth, the depth and breadth of the portfolio. Now it really comes to we need to develop standardization and platforms to accommodate the variety and complexity of the workloads.

Well, when you mentioned that the AI elephant in the room, it's that is certain that is certainly a topic that every 娇色导航that I've interviewed on this program for probably the last two years has had a lot of thoughtfulness going into it. There's a lot of discussion.

Sometimes there's action happening on staff.

I've had some CIOs tell me that, well, at the end of the day, it's really just more math, but the excitement that it is generated out in the public and I noticed the the old eminence Greece among the research houses, Gartner recently declared that aI had entered the trough of disillusionment, where you know, which you know, those, all those sine waves that Gartner will put out.

It's is ultimately like everybody thinks, Oh, my God, this is the solution to everything, and then we're back down into the reality of it. Is it now? Is it mainly in the work you're doing with your VMware integration that you've gotten more directly involved in the AI strategies?

Or is that something that Broadcom has been doing in different areas over the last couple of years. Anyway, where do you, where do you stand on that, in terms of where it falls on your priority list?

Yeah, so, I mean, I think there's, there's two, two aspects to that one, from the product side, obviously, we are, we produce some of, you know, world leading inference and AI related chipsets, as well as networking components on the hardware side to support the large scale compute and networking capable capability that's necessary to run some of these models, certainly in the hyperscalers.

And some of the larger, larger companies out there. So, you know, from an infrastructure perspective, we've been engaged in AI for a large number of years before AI was was part of the psyche. Because these things don't happen overnight.

You know, typically a three nanometer or two nanometer design or chip design merchant silicon happens over two to three years, and then you need to manufacture, then you need to so we've been probably looking at this problem for the past 44 to five years to figure out how to really enable some of the high throughputs and the TPUs that are required on the software side.

I think it's certainly more recent in terms of, how do you then take advantage of the large scale processing power to really solve business problems? What issues are you trying to resolve? And I think I'll go back to my previous statement of it's not one size fits all.

I think there are. I think we've got some very interesting use cases that are coming to market now.

Some of the product areas, semantics of one that springs to mind, where they have, you know, the their, their issue in the past has been security, the output of some of the results from a security standpoint has been very technical, so to all to understand, hey, am I really?

Am I really having a problem? I think the ad that the implementation of AI in that model, and that inference model in terms of really being able to generate a more a tactical and action driven outcome that makes to more people.

So, you know, the they had a very technical output that they've used generative AI, they've used their inference models. They're really trying to to make this more effective so that people take action quicker. So is it, is it? Are you having AI generate your, your, your threat model?

No. I mean, the threat model still there. The core foundation is still there. But how you take action? What? What are the next steps. I think AI is helping them significantly to go to make that quicker. So because, from a security standpoint, it's all about time to action.

So if you can get a quicker action, it helps. So I think that's a good, you know, a good example. From a product standpoint, there are a number of other things.

You know, there are more, I would say generic things roll out around customer service, call deflection, things like that. But, you know, part of the part of the challenge, as you stated, It's just math, you know?

I mean, it really is a lot of matrix multiplication and some good guessing.

So But ultimately, you can't get away from the fundamentals of your data needs to be, as you know, it's the garbage in garbage out scenario of really wrong answer very quickly if you give it wrong information.

So you know, we have to we can't lose sight of the fundamentals of your data needs to be relatively accurate in order to get best use of whatever aspect you're trying to get of AI.

So again, we're trying to, we're looking at, I would say, my, my acronym for AI, over the last couple of years has been always investigating we've tried a number of different as as most other companies are doing.

We're kind of dabbling a little bit in terms of where things are. I think we're moving more into the actually implementing things.

That may be the 20 the story for 25 and 26 where we can really take a look at again, but we view it from a business problem perspective. It's not really technology, it's are we, you know, we'll measure it on call deflections, we'll measure it on revenue increase.

We'll measure it on some customer satisfaction metrics, but it needs to be fairly hard tactical measures in order to justify the investment. Because right now, the investment in AI is not for the faint of heart. If you do, if you are doing multiple POCs without good return.

It's, it's costly. It's not cheap. So, I mean, there are, we are, I would say it's not so much disillusionment, but I would say you're, you're now getting a if you're, if you ever run cloud, public cloud, you have a thin ops. A team.

So who really look at the financial operations of cloud? I think those, those same people are being better used now in managing AI models as well. Because sure, how cost effectively are you? Are you making use of the model?

So even if you're solving a big problem if you're bankrupting yourself at the same time, that's not business outcome. So you just need to make sure you have a more cost effective model to figure out what real problem you're trying to solve.

I think in that regard, again, I'll go back to workload characterization. So some of the, some of the things that we are trying out, we will probably try out on premise.

First, we have a fairly robust model from a private cloud perspective, on the VCF thing, where we can have a relatively safe to feel playground in some of that area. So, you know, we let software do the, do the hard work.

But I think, you know, at least we can, we can try, in a cost, contained way, to implement some of the things that are trying to, trying to achieve. But I think at scale, we have a fair amount of scale.

At scale, probably 25 and 26 is when we're we're really trying to make that transition. Well, let's talk a little bit about that scale of the company itself.

Because you had mentioned that, I think it's 26 different business units, or is it 2726 different business units, 17 hardware focused.

I think around the world, VMware or not, VMware, Broadcom, has something like 40,000 employees, and when we talked previously, I was, I was, I guess, surprised and impressed to learn that you're managing this global technology organization with Just shy of 300 people, around 292 on staff, and that it is you're also operating at less than 1% of revenues for your IT budget.

And so I want to talk a little bit more about that and how that plays into the changing culture of the company, those 26 different business units, the history of hardware and the the broadcoms background as a wireless connectivity provider, all of that is changing a lot.

So let's start with how is that changing the culture of the company? From your point of view, I think the, you know, like, I say, our DNA was always brought up in the hardware business unit, so we were highly focused on margin.

You know, we weren't so focused on the top line revenue, but we were focused on how much money you make out of the dollar you're making. So that has really been a part of our DNA moving forward.

And it's, it's slightly, you know, I say that's more ingrained in the hardware side. I would say, typically from a software business, the they're more focused on top line. They're more focused on, hey, can I get the next customer? Can I get the next revenue?

And just less focused on, on the bottom line. So I think from a management of change or a culture. It's more impactful on the software side, or the infrastructure software side of the business units, which are the companies we've acquired over the past 456, years.

But I think once that, once that DNA is installed in that in that group, it's very effective in terms of, just, you know, focusing on the business problems, focusing on the cost, but, you know, from a from a scale perspective, you know, I'll, I'll probably use VMware as a more recent example of what we're trying to do.

So we, we, we acquired them back in November 2023, we had a we were very simple people in it. We like to count low numbers. So we had a day one, day two, day three. Day one was everybody on boarded.

And you know, the additional people came on board to Broadcom. Everybody got a login ID. And, you know, they got paid, they got email type of thing. So all good. Then in May, which was our day two, we consolidated all our business applications.

So we retired a significant number of legacy VMware business applications, and we consolidate to one, one common ERP system, one common CRM financial system, and then day three is probably still in process. So day three will happen at the end of this calendar year.

So day three is our infrastructure model, where we are consolidating. We inherited around 31 data centers from VMware, Broadcom, had 11. Will end that the calendar year with seven data centers.

So we'll have around and logistically, that is a it's a lot of things, a lot of moving parts, so we have to be highly focused on on what we're trying to do.

I think the at scale, when we're talking about at scale, will probably end up with around a around 30,000 compute nodes, around 200 petabytes worth of storage. But really the key to this is standardization.

So how do we manage 20 cent 26 business units plus four functions, still need to, I like, I still have to play nicely with HR and finance and legal.

They are my my friends at this point, and my company so but the 26 business units, they have a significant variety of workload characterization, and we, we may not be the biggest company.

I mean, the hyperscalers are, you know, you add more decimal points on to the end, so they've got way more scaling than than we're trying to manage.

But I think the diversity of workload is significant, because each of these business units are wildly different in terms of what they do in order to generate their end product from, you know, highly customized ASIC chips to, you know, SaaS operations for security, to pipelines, development stuff.

So it gets fairly, fairly interesting very quickly. I think the key to operating this, like you see, at a significant cost envelope of less than 1% plus, with less than 300 people on staff, is standardization.

So the way that we do that is we have what we term, three platforms, so we run mainframe. The mainframe group is fairly unique.

So the operations piece is run in the actual mainframe, but we support the we have three mainframe, mainframe pieces of equipment that we support them in. You know, it's water and electricity. Is always fun in a data center.

So, you know, it's always good, but we manage that connectivity there, and again, we view that as one platform. The other one is we have a significant footprint in our chip design grid.

So we run around 15,000 nodes, and, you know, 100 plus petabytes worth of storage on our chip design so that is bare metal. That's 15,000 pieces of bare metal equipment, high degree of compute, high degree of high memory capacity on these, these machines.

It takes a lot to generate some of these, some of these chips, the smaller the number is, is nice from a theoretical physics perspective, but it means, you know, you go from a five nanometer to three nanometer chip.

You quadruple the size of the computer and the stories required to generate that chip. It's get it's getting up there. I think our chips now have round about anywhere between 50 to 80 billion transistors on a chip.

So plus the next, the next, we always go for the next one as well. So we're never designing the stuff that's already in in play. We're always designing what's coming next. So, yeah, it's not getting any any quieter we run around. We just hit a fairly interesting milestone.

So we track runtime hours on our grid. We just hit 2 billion runtime hours on an annualized basis for our grid. But the which sounds sounds impressive, but it's actually more impressive that four, only four years ago, were running at 1 billion.

So in that four year time, we've increased the billion. So it's the rate of change that you have to manage, and again, you can only manage that if you enable fairly strict standardization within both the compute step as well as the story set to really enable that scale.

Well, it's interesting too, because what you were saying about data and operational standardization. I've been around the industry maybe as long as you have.

I was at computer world before 娇色导航magazine, and a lot of the things you're talking about, I can remember having conversations in the 90s with CIOs about this.

And it could be that there's just a natural sine wave of what's in vogue, but it also could be that as companies get more complex, you have to head back to what sounds almost old fashioned to be talking about data simplification and operational standardization, but there's no other way to do it and run really financially sleek.

Is there? Right? No, you if I was to so I think, you know, part of this, part of the part of the way that we, we make this work is, is our organizational structure.

I think I was going to ask you about that next, yes, I'm in a very, I would say, privileged position within the company where it drives, it is not driven by the business unit. So I report to the CEO. It's a good news bad news story.

So, but it's mostly good news because you get very clear direction, very clear decision making, because it's a one to one relationship, if I was to try and negotiate with 26 business units on a variety of topics, my budget would be four times what it is today, because I'd have to accommodate.

Everybody has a good idea. Everybody has good justification as to why they're things more important than somebody else, and typically in certainly in companies we have acquired that has been the case.

We've made a significant degree of cost optimization as we've acquired the companies that we have done simple and it's not because people don't know what what to do. It's just organizationally, they've not been allowed or aligned to the business objectives to allow them to execute to that.

So I, you know, I'm in a fairly, fairly privileged position to be able to say, no, yes, that's not and if they have a problem, they have nobody else to go to other than CEO, which is never a typical, no, no.

Nobody wants to kick it that far upstairs. Do that well, I remember you said that when we talked earlier, that you're there to make everyone successful, not to make everyone happy.

Yeah, that surprised me a little bit too, because for a long time, CIOs have really felt pulled in several directions. You know, you have to keep the business unit presidents all happy. You have to keep the CEO happy.

You have to get out in the field and talk to customers, and it's such a complex job.

And when I you know when we were talking about the shields, sheer scale of what your global technology organization is running, plus the fact that you acquire, you guys gobble up very big companies like VMware, just last year, it seems like at some point the organizational approach would need to change for that, but it seems like you've just adapted your kind of straight down the middle approach to that, and it's been working pretty well.

No, I think you know it, it's one of those ones where, like you say it's we, you know, I, we used to, I would say about 10 years ago, maybe 1012, years ago, we actually did run a CSAT and try and figure out if the if the business units or our internal customers were satisfied or happy.

We stopped doing that because it actually didn't. Didn't matter if, we were four out of five, three out of five, it was more. We then changed the focus to, can I manage it on a on a what cost constraints have I got? What capabilities can I?

Can I enable and I can make you successful, even though it might not be exactly the path you want to go on, but I will enable your capability in a standardized way and drive that level of standardization and capability.

And the way we measure success is I do have a very low burden on their margin from an individual business unit. So our budget splits 26 ways.

So it their, their cost of running IT operations, or it as a as an allocation, is significantly lower than it would be in any other organization, and if they run it themselves. So it is definitely the GMs do see the value. But it takes time.

It takes time to build that trust. It takes time to but I think, like I say, the relationship is this only works because it reports directly to the CEO, and the CEO manages the provides air cover.

It's hard to imagine that could work as well if it was reporting into the chief financial officer or the chief operating officer. Have you? Have you been in leadership positions before where it wasn't reporting to the CEO? And that brought its own raft of difficulties.

I am a I am an HP alumni. I'm an Agilent alumni where, yeah, it definitely wasn't that. It wasn't that way.

And as we, as we had, I gone through it back in the, you know, in the 90s, in the interesting days, you know, I worked for a multi billion, you know, HP, multi billion dollar corporation, lots of business units, lots of issues going on, a lot.

So fuzziness going on, which is, I think, you know, HP seem to lose their way in terms of the focus, too much the over engineered, too much on the feel good factor for people, and lost sight of the fact that if you really go study bill and Dave's kind of mantra on the HP way, the first, the first rule of thumb is, make money.

Okay, make money. Then, then you can go do the other things that you need to make. So it definitely got lost in translation a little bit.

But, you know, a, you know, we went down to, we started, I would say the real focus, and maniacal focus on cost and standardization really happened when we split out from Agilent onto Avago around about 2006 so that was where we, we really started honing operational efficiency and effectiveness over time.

So that was, that was when the DNA really, really kicked in, and it's scaled so we haven't from 2006 to now, not really changed, because if you're if your fundamentals are correct, if your fundamentals on data quality, if your fundamentals on standardization and focusing on what business problem you are, you're tackling and your measurable ROI, and you can execute to that, you're allowed to execute to that.

Then it scales. We are. We are living proof of that scale.

And I think it's actually coming more to fruition now as we go into our you know, as we get into day three, because I mentioned two of our three platforms, which is mainframe in our grid, the third one is where we're having, I would say our more interesting a impact now.

So as VMware, we are now generating our own private cloud. Transcribed by https://otter.ai