From the Dot-Com Crash to the AI Era: How Builders Survive Waves of Disruption

If you look at VMware's roughly 20 year history, the first decade was us disrupting and growing and the second decade was others coming after us to disrupt us. Right? When a company gets large, you get very good at the math of the business, but you've kind of lost the soul of the business, which is are you innovating at a very impatient velocity. Operate like the world's largest startup. Make sure that you're operating at speed with scale. Don't delegate the storytelling. The story is the strategy.

So, maybe just to kick it off at a high level, let's talk about running a large operation both as a disruptor and through a market transition or transformation. So I'll start with you, Raghu. When VMware came out, I remember it was the late '90s, and I could run Linux on my Windows machine. It was the most amazing, magical moment, and it really changed the industry.

I remember when it became a $30 billion company. Everybody was running it; it was a monopoly, but it also had to weather a number of transitions going forward. Maybe talk a bit about the mindset as a disruptor, how you evolve your mindset as a leader, and how you evolve the organization as you have to go through a transformation yourself.

If you look at VMware's roughly 20-year history, the first decade was us disrupting and growing, and the second decade was others coming after us to disrupt us. That's roughly the VMware history.

There are 3 or 4 weapons of mass disruption in our industry, especially on the infrastructure side. One is if you introduce a new piece of software that introduces a new abstraction or a new usage model, and everything starts to aggregate around that. Or you are able to successfully bring about a new class of users that were previously not consumers. This is the classic description of the innovator's dilemma. Or you introduce a new business model.

In the case of VMware, we did the first and the last one. We brought about a new abstraction for the data center, or for compute, to be more specific, which was a software-based virtual machine. Then we had a business model associated with software as opposed to hardware. That was the core of the disruption.

Because of the benefits of software, we were able to fundamentally change the usage patterns in the data center. Once you start to change the usage patterns and behaviors of the practitioners, you get locked in, and the disruption is almost impossible to remove. Those weapons of mass disruption were used against us—not intentionally, for sure.

The 2 big ones that we faced were, first, cloud. The second one was containers and Kubernetes. In the case of cloud, all the same elements were present. Initially, Amazon Web Services did not change the abstraction. It was still a virtual machine, but everything else—the business model—changed.

The most important change that AWS brought about was that it made infrastructure available to developers. That was the massive unlock of a whole new class of users, and we had no idea how to work with developers. We did not know the secrets of how to do that. I think that is the point. Then Docker and Kubernetes and things like that changed the packaging and the abstraction.

The interesting thing is that nobody in the industry made money on either Docker or Kubernetes. There were some startups, but really there was no big company that was created out of that. The real fundamental disruption we had to fight against was in the cloud, and the cloud disruption was so hard for us to fight because it brought in a new class of users for infrastructure.

Cisco has been such an iconic company for so long. It really rode the internet wave, and then it rode the data center wave. It's been incredibly successful at riding multiple waves. It feels like it's hard to make a wave—you can't do that—but once they happen, you can ride them and capitalize on them.

We're clearly seeing one right now with AI, so it seems like a great opportunity for Cisco again to figure out what to do. I'd love to hear how you're thinking about the waves that Cisco did miss, and how you're thinking about navigating this as a leader at Cisco.

The obvious one is that we missed the cloud wave. If you take a step back, every large company that's successful has to have gone through a startup phase, because every large company was a successful startup at some point in time.

What ends up happening, though, is that at some point, when a company gets large, it loses touch with the front lines. You get into this mode where you get very good at the math of the business. Everyone knows what the gross margin of the company is, but you've kind of lost the soul of the business, which is: Are you innovating at a very impatient velocity and a very fast velocity so that you can continue to keep discontinuously leapfrogging?

When that happens, you have to hit the reset button. About 5 or 6 years ago, that had happened for us, and we had to hit the reset button.

Let me tell you what needs to happen when you hit the reset button. You have to make sure that you get people who have a founder's mentality as executives in the company, or just throughout the company. Most of the people on my leadership team are former CEOs of acquisitions that we've had who are actually running these pieces.

You have to have a combination of CEOs and people who know how to operate within the machine that is Cisco. The CEOs can put pressure on the company by being impatient and saying, "This doesn't make sense. I'm not moving fast enough." Then the people who know how to navigate Cisco can guide them through it. If you can make a good team out of those 2 types of people, magic starts to happen.

Where we've really started to see the tempo pick up is in how we operate. Internally, we have a mantra that says, "Operate like the world's largest startup." Make sure that you're operating at speed with scale.

You have to have a great 0-to-1 practice, a great 1-to-100 practice, and now a 100-to-1,000 practice. We're getting very good at saying, "In 9 months, can you get a product from 0 to market? Then, in 3 to 4 years, can you get it to $1 billion?"

If you can do that 8 to 10 times, you've actually got a pretty healthy, additive book of business that you didn't have before. You can either create a category or make sure that you ride a category that's been created. You can do both of those; it takes a little longer to create a category.

I think large companies are very good at running many experiments. What they're bad at is doubling down when an experiment works. That's where startups are good. They'll just focus on one problem.

The other thing that you usually see where things fail in a large company is that if you read one too many TechCrunch articles, you will build a 0-to-1 product that does not keep in mind the route to market that you have available to you.

You have to make sure that you can actually ride the route to market that you have. You can say, "If I can build 10 products that actually leverage my route to market, then I'm going to have an advantage in getting a disproportionate amount of velocity to get to market."

But if I don't leverage the route to market, and if I try to build a new route to market for every single one, then the sales team is going to feel starved, and you're not going to get to that level.

Let's dig into this very specific issue. I think many people don't appreciate how complex this exact thing is for a large company. You said something that I think is exactly right: Large companies lose touch with the front lines.

Steve Sinofsky has this great model of this. For any disruption, you'll have incumbents. Let's say 80% of the dollars come from 20% of the customers, and those customers are large enterprises with very sophisticated needs. That drives so much of the business.

They've got these deep relationships with a few customers, and then a new disruption happens. They're with those customers; they're not with the front lines, which tend to be different. Now you've got to retool go-to-market, product—everything—for this new base, even though 80% comes from the existing base.

Raghu, I know you had to deal with this. How did you think about evolving the company, not around a technology disruption, but even just catering to a new base—from the market side, the marketing side, and the sales side?

What you said is spot on, because when you're talking only to your best customers, by definition, that's not where the disruption is coming from. Like any other large company, you get really good at talking to your best customers, and that's where all the incentives are as well.

Companies deal with it in one of 2 ways. One is they fence off a different team to approach this thing completely differently, allowing it to break every rule in the book.

Or they go and buy.

Yeah. Right. So we have, at various times, done both. There are certain classes of innovation where you can get by with what I would say are close adjacencies. If you're building a product that's closely adjacent to your original product, you can take it from zero to one through the existing sales force, because it's going exactly at the same user.

Yeah.

But if you're, for example, going after a different user, like we were doing with NSX in networking, then you need to fence it off.

You know what's so interesting about VMware? It's one of the few companies that actually did both pretty successfully. vSAN was storage—

Exactly.

You pulled it off; I don't know how. And then NSX was inorganic, and you pulled it off—I didn't know how. So, what are the lessons of inorganic versus organic and the complexities?

Yeah. I mean, I think you have to be very careful about where organic is going to work. Obviously, everything starts with the product. For so many of us in the business, that's sort of a truth.

And in the case of vSAN, to your example, it was a close adjacency to compute, but it's a different buyer, right? Is it? You tell me—I mean, you're the expert.

Yeah. In fact, we had 2 iterations on vSAN. Initially, we said, "Hey, let's go after the storage buyer." That did not work as well. When we said, "Look, we're going to expand the compute buyer's purview"—

I see. Right.

Then it started working. So, the go-to-market started to match together. The other thing about going into existing categories is that you've got to be 10x better.

You have to be 10x better.

So, in the case of storage, we were 10x better than having external storage, at least for VMware use cases. That's why that worked.

And by the way, that 10x better is a very counterintuitive thing for large companies to think about, because they always think about catching up, and no one wins by playing catch-up. The asymmetry of coming at it from a different angle and being at least an order of magnitude better is something that you have to keep pushing at and be the biggest skeptic about, because most people will tell you that they're 10x better when they're actually 15% better. And 15% doesn't create a displacement of an incumbent.

Just to finish that thought, in the case of networking, as you all know, it was a different way of operating a network. That was the value proposition—not necessarily 10x better. It was something that could not be done on a physical network at all. What do you mean you can program the network? What do you mean you can have virtual firewalls, et cetera, that scale well? I think you really have to understand the nature of the disruption that you're bringing to the market and then tailor the rest of the go-to-market around it. That's why we chose different approaches in both cases.

And you have to—the other thing that I would say, to what Raghu said—is you have to define very clear insertion points.

You mean in the market or in the company?

In the market, right? If you have an incumbent that's already in the market and is entrenched pretty well, a large company will typically try to say, "Okay, I'm going to go out and build something for the greenfield with an entire platform." But the reality is, the market's not greenfield; it's brownfield. So, you have to identify an insertion point where your competitor might be there, and then over time you have to make sure that you extract the competitor, which is not something that's intuitive to people, because they're like, "Oh, I'm going to compete with them." It's like, no, you actually have to coexist first before you can displace. That requires a very open, ecosystem-based mentality, which large companies sometimes tend not to have.

So, one thing I've seen fail consistently—again, you guys are way more expert than I am—is this notion that if an engineer spends 1 day a week experimenting, then maybe they'll come up with a great idea that changes the company. It just feels like that tends not to have enough momentum behind it to do something big. I've also seen fail this idea that you take 1 large team and say, "You've got to do 2 things: the new thing and the old thing." It's just too hard. What seems best is ring-fencing it a bit.

I think Cisco, in particular, has a lot of experience with this type of thing. If you're in the AI wave and doing something disruptive, and you've decided not to do it inorganically—you're doing it organically—how do you think about structuring it?

The way that we structure it is, you think of a 2-pizza team initially that you start with, really small. That team has agency and air cover from the very top, all the way up to the top of the food chain. What you have to do, because there are enough antibodies that will argue why that's not a good idea and that it should be part of the core, is make sure that that team is protected from that and is just focused on driving it.

Now, they might be good at getting a version 1 product out to the market, but they are not going to be good at getting the entirety of the sales force of 17,000 sellers, in Cisco's case, to go out and get it. So, what ends up happening is you have to be very prescriptive about an ideal customer profile that you want to make sure that you start this with. This is where I feel like not losing touch with the front lines is one of the best ways to not lose touch with the front lines.

You start with a zero-to-one project. Start from the bottom, because you never start a zero-to-one project from the largest financial services institution.

Exactly. Right. You start from the bottom.

And when you do start from the bottom, you give that team enough freedom and enough agency, and then over time construct enough incentives in the different teams for making that thing successful.

Yeah.

When that happens, you start to see a snowball effect start to occur. But it takes a while. One of the things we learned that failed the first couple of times we did this was we'd have version 1 of the product out, and then the field would just reject it, saying, "Yeah, it's not ready. It's not a complete product." No, it's actually ready for this segment of the market.

Would you do an overlay to help?

We would have an overlay sales team, but the core sales force would keep it out of the account. It would keep it out of the account, like, "Ah, it's not going to go sell at Bank of America." I'm like, "Yes, that's actually the wrong account to go sell it in as a version 1 product, because it's not ready for all of the capabilities that are needed there. I don't need to have a data center in China for my first billion."

Yeah.

Now, when you have an incubation team, they have 2 jobs: build a great product and define an ideal customer profile. They also need to make sure that, within that ideal customer profile, you have initial adoption and repeatability of a go-to-market opportunity-creation motion.

After you've got product-market fit very clearly defined, and once you've got that ICP nailed and you've saturated that market—or you're starting to saturate that market—expand the ICP, and then expand it again, and expand it again. I think that's a very counterintuitive motion for large companies, because it's very hard to go to 17,000 sellers and say to them, "You, 10,000 of the sellers, I'm only going to train you on it."

Yeah.

So, it gets to be a little hard to do.

Yeah. I think the term "ideal customer profile," especially in larger companies, sometimes misleads people into thinking, "Oh, my customer is JPMorgan. My customer is Home Depot," et cetera. That's an ideal use-case profile.

The ideal practitioner profile.

Exactly.

The ideal practitioner profile. So, you've got to narrow-cast it down to who's the person that's going to wake up every day using that thing that you're going to build, and then revolve all of your practices around that, from go-to-market, of course, to product.

But I will tell you this: once you start building products that start getting momentum in a large company, the company starts regaining a spring in its step. I think Cisco is going through this right now, right? Cisco's got its mojo back.

It's got its mojo back.

Stock at all-time highs—congratulations.

Well, you know, those things—you never know how these things go up and down. But in general, we're pretty excited about the reception in the market. The most gratifying thing is when you go into the internal audiences.

Yeah.

You can feel the spring in the step of the employees.

Yeah. And why is that?

Because we're starting to win again. Most people want to come into the office wanting to win. It's not like people are saying, "I'm going to phone it in." But what ends up happening is, when you have 95,000 employees, you can tend to dissipate a message very quickly.

Yeah.

So, one of the best pieces of advice I was given when I took over this job as head of product came from one of my board members. He pulled me aside and said, “I’m going to give you one piece of advice.”

Yeah.

He was the ex-CEO of Northrop Grumman. I said, “What’s that?” He said, “Don’t delegate the storytelling to anyone in the company. You go do it yourself because you need one voice.”

That’s great. That’s smart. Yeah, yeah.

It doesn’t mean that I’m the best presenter in the world. But if one person doesn’t tell the end-to-end story—someone who actually owns it—you start to fracture the story. Once you fracture the story, you start losing control of it.

Totally.

The story is so important. The best in the world at this is Apple.

Yeah, of course. Right.

You just want to be 10% as good as Apple. I think we’re really getting obsessed about that, and right now it’s starting to work. When you can galvanize 95,000 people with that kind of clarity, you can start to feel a tempo in the business that’s just very different.

In fact, I’ll go one step further and say the story is the strategy.

That is the strategy. Yeah.

Right. Because human beings—95,000 people together—are not going to understand it as 5 bullets.

It is the story.

Something that’s unique to this wave, this AI wave, that, Raghu, you’ve certainly seen—and maybe you saw a little bit of at Box—is really this kind of consumer-prosumer movement, which the internet was, and which Netscape certainly was to begin with. I mean, it was like—actually, VMware kind of was, too. I’m just thinking about that right now—but you paid $199 and then $249 and downloaded the product. I did it as a college student, right? And so how do you think about navigating a wave where the buyer is not just the ICP? It’s the actual consumer. Is it like you sell to the people who sell to them?

Yeah, yeah. I mean, I think, firstly, the biggest danger for enterprises or even for startups is to look at this AI wave through the lens of previous waves. Maybe there are some applicable lessons, but this thing is so big and so different that you’ve got to look at it from first principles.

Mhm.

Yes, for sure. It’s particularly dangerous because we have these old AI chatbot things that didn’t work in the past, but we created businesses around them. I think we’re used to thinking about the old AI, not the new AI. It almost rhymes with previous stuff, and so it makes us all think in those terms. OpenAI—whatever OpenAI did—broke every single rule of what had been done previously, and they’ve actually wildly succeeded.

So I think this is the new skill for companies to learn as well. You’ve got to, especially as infrastructure or application companies, almost ignore the old model, right? Selling to the seller who’s selling to the buyer internally—that chain is forever broken. SaaS broke it, but AI breaks it even further.

What that means is that your product managers and everybody else who’s trying to figure out what the product is have to really think about the end user from a direct-reach point of view. There are different buyers: You could sell to IT, or you could sell to security or marketing. But these are literally individuals with credit cards, and they’re not even developers. Developers are a central buyer that we understand how to sell to, but this is some random employee asking ChatGPT to write an email. It’s very different.

It seems like Cisco has been able to navigate this because the network is such a point of leverage. If every user uses something, the network TAM grows. I’m curious: One option for Cisco is, whenever you have a new data center, we’ll sell you a new switch, and it doesn’t matter that a lot of the market is— you could argue that the entire web is something Cisco did a phenomenal job with, the data center switching, and it didn’t go sell to consumers. That was a consumer phenomenon. Is that the way that you think about it?

Yeah, I think the way that we think about ourselves is that we are the critical infrastructure for the AI era. You have to look back and say: Where is AI constrained? Right now, it’s constrained in power, it’s constrained in compute, and it’s constrained on the network. If the packet is delayed getting to the GPU, the GPU is idle. A GPU being idle is like burning money, especially in the training runs.

You have to make sure that you actually have a very efficient packet flow going over there: low-latency, high-performance, high-energy-efficiency networking infrastructure. That’s super important for training. As you have more agents in the workforce, your inference demand is not going to be spiky. It’s actually going to be sustained, because an agent today probably works autonomously for 20 minutes.

Yeah.

In the next 6 months, it’ll work autonomously for maybe 2 hours or 10 hours. Then it’ll work for 2 months, then 2 quarters, and then 2 years. The longer it works, the more sustained and persistent your demand for inferencing will be, and your network capacity appetite is going to go up quite exponentially.

If you have 1,000 employees and you add 10,000 agents, that’s like having 11,000 employees. That means your network bandwidth has to be equivalent to what can serve 11,000 employees. You’re going to need more infrastructure, so we are a direct beneficiary of that.

The second area is keeping it safe and secure. Securing AI is going to be pretty important. We actually have a core foundation on the safety side, so those 2 become core foundational elements of the infrastructure.

The third one is data, and we’ve got Splunk. You have to find those foundational elements that say that, no matter what happens with the business model, people are going to need them. The beauty is that your business model is not that complicated: As there’s a spike in demand for applications, you’re just going to need more infrastructure, and you have to continue to sell the infrastructure to the people building it.

Yeah. I mean, I think infrastructure traditionally has followed usage models, right? So, back to your consumer thing, you really have to understand the usage pattern, whether it’s human beings typing in chatbots or—

I’d say with Cisco, it’s also just good to be king, right? I mean, if you’re in the network, whatever compute grows, the network grows.

But it’s not always that way. During the COVID crisis, I think it’s great to have share, but if you stop innovating, you can actually start to see customers get frustrated. For about a 6- or 7-year period, we had just stopped innovating, and I don’t think it was that productive for us.

Getting back out to innovation right now—for example, in the past 18 months, we’ve probably done more innovation than in the previous 10 years combined. Our biggest challenge right now is sitting down with the customer and giving them the story that says, “Here’s where the innovation has happened.”

When you sit down with them for 90 minutes, they love it. Doing that for 1 million customers is really hard. At scale, changing perception is hard, and that’s one of those things that took us about 1.5 years to do.

One thing that’s interesting about this AI wave, actually, is that Aaron Levie was sitting in that seat.

One of my best friends.

Yeah. He brought this up, which I thought was really good: During this wave, you have a lot of the original founders still running the companies. You’ve got Mark Zuckerberg, Jensen Huang, Ali Ghodsi—there are many. A founder, as we’ve seen many times with the Reed Hastings effect, can actually navigate a transformation because they know the team, they’re product-focused, they have moral authority, and they have support from the board. They can do that.

When you joined VMware, Diane was still running it, right?

Right.

So you were there, and you’ve actually seen founders do it. But listen, you were CEO as a non-founder, and you guys have worked with non-founders. Do you think it’s a different job?

Yeah, there are a couple of things where a founder has unique license, right?

Right.

But then there’s the rest of it, which is: How do you make the change happen?

Right. How did you think about it? I mean, you were a CEO—you’d been in the company so long, maybe you were virtually a founder.

Yeah. I mean, the first element of it is: What is the technical and product credibility and license that your team is giving you?

Right? In my case, fortunately, I'd been there for a long time and driven a lot of the waves that VMware had gone through. So, I had some of that. But the other important thing that you need is early conviction on what to bet on, right? Because founders get much more time.

The market gives them more time, their board gives them more time, and their employees give them more time. Non-founders, for various reasons—and I don't know if you agree—you don't get the same amount of time. So what you've got to do is develop very early conviction about what the bet is that's going to drive this transformation, and then what the narrative behind it is, to your point earlier, and how you drive that into every aspect.

The other part of it that you've got to do, which founders naturally do, is—the overused phrase of the year probably is “founder mode,” right? You really have to own that change, every little aspect of it, right? From every check-in to the way it gets to the market. Those things are all common, but you're absolutely right: founders have a better ability to do that because they are founders and they're natural entrepreneurs.

Yeah.

See, I have a slightly different take. I agree with everything Raghu said, but I also feel like I've never been a founder. I have never felt like I'm not a founder of any company I've worked for. Yeah.

I would say you sure feel like a founder.

I feel like I founded Cisco, and this is my company, and I am going to make sure that we make it successful.

I think that owner's mentality is actually what's more important in my mind than being a founder. The way that I think about the owner's mentality is that you've got to make sure you're extremely impatient. You're running out of time; you always remind yourself that you're running out of time. You always market in rather than company out, and you don't tolerate any level of mediocrity.

You don't try to win a popularity contest because you'll never win it, especially the larger the company gets. It's a very dangerous trap to fall into because a lot of people will tell you what you want to hear. So you almost have to work hard to surround yourself with people who are excessive critics, where they might have an extreme version of looking at everything that's wrong with you and keep nitpicking on it.

You need to have some of those people around you who can tell you that, so that you are constantly getting better. I've never been a founder, but I always feel like I'm an owner of the company I'm in.

The other thing is, I think finding the truth is especially hard in a large company.

Yeah.

That's another place where I think founders can get to the bottom of things more easily.

And by the way, that's a really important point. In large companies especially, we start to create versions of the truth and then start believing those versions of the truth. That happens so frequently.

Seeking the truth and going down to the facts of what is, in fact, not working is especially important. What ends up happening, especially at a company like Cisco, is that one of the things Cisco is really amazing at is that it's a very compassionate culture. One of the things that Cisco is not that great at, because it's a very compassionate culture, is that we might not actually have very direct conversations sometimes.

You have to make sure that when something is not good, you sit people down and say, “We're failing.” You have to use binary language: “We're failing in this area. These are the 3 things we need to do to succeed, and if we don't succeed, there's an existential threat. I need and expect you to do this to make sure that happens.”

I feel like it's very unnatural for people when someone comes in and starts doing that initially. You have to break every mode of hierarchy. I'll give you an example. When I first joined, I said, “Let's start doing design reviews with the Webex team.” They said, “What are you talking about? This is a member of the executive leadership team. G, you're too senior. We don't need to do design reviews.”

Then they started doing design reviews, and the senior vice presidents would come in and do design reviews with me. I'm like, “No, I need to make sure that the designer, the PM, and the engineer are actually doing the design reviews.”

And so then they started doing 8 meetings of prep before they came to me because every layer was in the prep.

That is a very common disease.

Yeah. And so then one of the guys who was there said, “Stop this madness.”

Yeah.

His name is Tai. He was a great engineer. And he said, “We're going to learn when G learns, and this is a safe space. All of us are editors. We keep our titles out. Let's just start to make sure that we edit the code that's going to ship.” Right?

The moment that happened, there was an unlock in the team, and everyone started thinking differently because then it became a safe space. We were critiquing each other's ideas. They were telling me I was wrong, I was telling them they were wrong, and we were debating back and forth. The titles were left outside the room, and that created this magical, purely best-idea-wins environment.

In a large company, best idea wins happens very seldom. You have to make sure you fight that urge to say rank wins. The best idea has to win.

Yeah. It just occurred to me—I don't know why it didn't occur to me before—that both of you are product people.

Yeah, of course.

Right. I don't mean anything against people with other specialties; they're all phenomenally difficult to acquire expertise in, in many of these corporate domains. But you cannot navigate this AI transformation, or for that matter any other transformation, if it doesn't start from the product.

Yeah, I think so. That's very interesting. Do you manage from product on out?

Absolutely. I usually think the product is the soul of the company. When we start thinking about the product side, Cisco used to be a very sales-led company.

Well, I mean, John Chambers is the consummate sales guy.

And what I give Chuck a lot of credit for is that he said, “We have to become a product-centric company.” Chuck's a sales guy by training, and he said, “G, we have to make Cisco a product-led company, and you have to start from the product.”

Every single time a product person started complaining about why the salesperson wasn't selling well, he said, “If you build a good enough product that has market pull, you're never going to complain about enablement.”

If you build a product that cures cancer—one of my mentors used to tell me this—if you build a product that cures cancer, you can sell it in the Himalayas from 2:30 to 3:30 p.m. on Tuesday afternoon, every third Tuesday of the month, and there will be a line out the door, right? Because you're curing cancer.

So what you have to do is build a great product and make sure that there's enough pull. I feel like we're probably 70% of the way there now at Cisco, where everyone originates by thinking about what's the product and how are we going to add value.

You know, I feel like companies go through 3 stages: the product stage, the sales stage, and the operations stage. There are actually leaders for each one of these stages. The same person can do all of them, but they have very different requirements.

The danger zone seems to be that you've got the operations CEO during a transformation, because it's such a product-driven transformation and they've got to go ahead and reset. So I do think that founders tend to always be product people, because you have to know. If you look across any large Silicon Valley company, or for that matter outside Silicon Valley, where the leaders of the companies have successfully navigated the transition, I don't think you can think of an example. Maybe not.

Except for there are a few people who may be touched by divinity, like John Chambers, but there are very few of them.

But the one thing that I've never understood in large companies is that the one formula that definitively does not work is saying, “I'm in a tech space, and I'm going to cash out this business and not innovate, and I think I'm going to do great.” That never plays out in the long term, right?

But it is so common.

But it's so common, and I've never understood that. It becomes a self-fulfilling prophecy 100%.

It does. It's like, just keep innovating. It's a very simple formula: just keep outdoing yourself and doing better every day. We have this line internally that we use: just get 1.27% better every day from what you were yesterday, and in a year you'll be 100 times better. There's just the power of compounding.

Would love your thoughts on the AI stuff, Raghu. I know it's pretty early, and I'd love to hear your thoughts, too. Do you think this changes infrastructure? Do you think it's largely independent of infrastructure? Do you think this increases TAM? How do you think about this on a macro scale?

I think it changes infrastructure in an enormous way, right? Infrastructure always is a follower of the workload.

Yeah, of course. I mean, go back to Netscape, right? When the browser came out, we were running on 4.8-kilobit modems or whatever it was, right?

Yeah. And then look at the internet infrastructure today.

I remember the transition from the wiring closet to the data center. Remember, it used to be—

Exactly.

These mega data centers actually switching.

Yeah. Yeah. I mean, the neural net was created on a gaming chip.

Yeah, that's right. And today, now you've got AI factories. Like you said earlier in the talk, the bottlenecks are progressively moving everywhere. So I think, fundamentally, every layer changes. It's not just compute. High-bandwidth memory changes, obviously the networking changes, and the storage access patterns are changing. Every part of what we think of as conventional infrastructure is changing, but the layer below that is changing as well.

I mean, this whole business, at the end of the day, is power to tokens.

And so everything that's in between power and tokens, starting from power generation to the token output, is dramatically changing.

Do you buy that this is a 10x step-up in market size?

Yeah, it's a 10x step-up in market size because the applications are a 10x step-up in market potential. They're replacing labor. They're replacing—

I mean, they are GDP-enhancing, whereas previous generations, while there were tremendous advances in productivity, were not to this scale. I think, actually, infrastructure, from a market-size as well as a scale perspective, is more like 100x to 1,000x—not 10x—in infrastructure scale. You will probably see 2 to 3 orders of magnitude. And I think one of the things that we are extremely obsessed about is that, when you do something like this, vertical integration is super important.

We build our own silicon ASICs. We have our own network infrastructure, our own security platform, and our own models that we're building in certain cases, because I think the models are going to get smaller and more bespoke. We have our own data platform, and we have our own observability stack. I think that having all of that work well together, while being completely open to the ecosystem, is super important in AI. Right now, I feel like those structural changes in how companies think about an open ecosystem will be challenged quite a bit, because you have to partner with your largest competitors.

Yeah.

And be unapologetic about it, and make sure that you don't let the internal forces stop you from doing that. One of the best things we've done in the past 5 years is that we've completely opened up to our competitors. For example, Microsoft Teams is a huge competitor for us.

Yeah.

We partnered with them and had them run natively on our devices. Hundreds of millions of dollars in revenue accrued because of that for us.

That's awesome.

And it wouldn't have happened if we had not opened it up. The one lesson I've learned in that is, when someone owns more than 20% share in the market and you don't integrate with them, you're just excluding yourself from the market. You're actually not doing anything to displace that vendor. I think in AI, that's going to get even more prominent.

That is great advice. That's why I asked Raghu whether I could insert into the hypervisor, and he ended up buying the company.

Yeah, that was the only way that was possible.

Because he said no.

Yeah. By the way, on your point about vertical integration, I think that's an interesting industry debate, in my opinion. If you look at the last 2 industry waves, the PC and the—

Maybe—listen, it's a timeline question. They horizontally started—

Google makes chips now, man. So then, over time—

That's where I was going.

Yeah. Okay, sorry. So now, in this wave, it's soup to nuts, and we'll see if this is the right model.

Yeah. Yeah, for sure.

And we are totally starting on cybersecurity. We're starting to etch security into the silicon.

Yeah. Cool.

And it'll just have a very different performance output than what you'd have otherwise.

Totally. Totally. I do think that there are signs of horizontalization also, by the way, in this AI wave. For example, the open-source AI models are actually pretty successful, and there are separate inference platforms. I don't think the vertical-versus-horizontal question has played out yet. It's just so fast and so furious.

But that's why, if you do the vertical, you have to be very horizontally friendly.

No, that was a great point. I love that. So, listen, we're coming to the end of the time. We've got 2 of the most storied, elite executives in infrastructure. A lot of the people listening to this are in companies navigating this transition. Maybe they're not as big as VMware or Cisco, but perhaps they're just a few words on how you think about navigating this and the opportunity. Do you have a bit of guidance?

I'll give it to the founders, who I think are kind of the heart and soul of America—and the world, actually. The startup ecosystem is really important, and I think it's important that we keep it vibrant. I would say that's a 6-part formula that I use for what's really important, in descending order, when you think about building a great company.

Number 1, and the most important thing, is timing.

Yeah.

Get the timing right. Right now, don't fight the megatrend. Make sure that everything you do actually has AI as a tailwind. Otherwise, you're not going to win. Timing is number 1.

Number 2, make sure you go after a very large market that you can attack a step at a time. If you try to go out and swallow the entire market, it's going to be really hard. You have to make sure you go after a large TAM, but address it a step at a time. Ideally, create the TAM. Don't go after an existing TAM. It's even better.

Number 3 is team. Oftentimes, people will say, “Does team trump market?” I actually think market always trumps team. This is not my idea; this is Marc Andreessen's. I think that's consensus, by the way.

It's consensus right now. Market always trumps.

Market always wins.

Number 4 is product. I think you have to build a great product. In my mind, there are 3 parts to a product. Build a product that people love, that they talk to their friends and family about, because that's the only way you get to hundreds of millions of users. Number 2, get adoption and really understand retention and why it happens. Number 3, get to commercial relevance. Otherwise, it's a science experiment.

So, timing, market, team, and product. Number 5 is brand.

Yeah.

You've got to build a brand that's actually identifiable, and don't try to have too much noise in the system. Just have 1 message over and over again.

Number 6 is scale distribution. Otherwise, it doesn't work.

I mean, we don't have time to explore it, but the last 2 are very, very different today than—

How you do it.

And by the way, this is again a challenge for large companies. In fact, this very medium that we're on turns out to be one of the biggest brand-building mediums ever.

And it's too scripted. That's the problem: people don't want to talk.

Authenticity is a big thing. And I will say the last thing for people who are looking at these disruptions and finding them a little bit scary: in my experience, there's a lot more opportunity than not. In a way, run toward the fire in this case.

I do actually think it's interesting when people say, “The AI thing is going to make humans irrelevant.” I think we're so far from a point where humans are not going to be able to add value to society. Once cancer is solved and, whatever, once I could simply pay my taxes, then we could have a conversation. I have to pinch myself and ask, are we actually having this argument? Because right now—

I can't write a full board presentation with AI just yet.

I mean, until I can go to the DMV easily, come on, man.

Forget a board presentation. Just even a careful email, right? A very careful email that you would send to a customer to close a deal or whatever it is, right?

There's a lot of upside going on. Well, listen, thank you so much for joining us. This was a lot of fun.

Absolutely.

Thank you for having us.