Palo Alto Networks CEO Nikesh Arora explains what makes a startup worth buying and why he puts the founders who beat his team in charge.
Arora joins Eric Newcomer to break down his acquisition strategy, what he looks for in cybersecurity startups, and how he keeps acquired founders motivated. He shares why buying a better product can beat building one internally, plus leadership lessons from his time at Google and SoftBank.
The conversation also explores how AI is accelerating cyberattacks, why securing AI coding tools remains an unsolved problem, and where Arora sees the next wave of opportunity. Plus, his case for OpenAI and Anthropic potentially becoming $10 trillion companies within five to ten years.
[00:00:00] AI is going to fundamentally change the speed of attacks that we see in cybersecurity, which requires a whole rewiring of the defense mechanism, both from an infrastructure perspective and sort of AI in defense perspective. In cybersecurity, every bad actor is trying to attack you in a net new way. They're not trying to do it the way they did yesterday because we probably found an antidote to the way they attacked you yesterday. So we've got to be constantly on our toes looking for where does innovation live in the realm of cybersecurity.
[00:00:29] Yeah, I mean, what is your reaction? I mean, you're a leader of an enormous company. You're someone who shares, you know, driven, I think people think aggressive, their personality traits people see similarly to Jensen. And then there's also this acquisitive deal-making attitude. Like, what is your reflection on sort of losing him as a leader at NVIDIA? Everything that's happened in AI has surpassed everyone's expectations.
[00:00:55] And I've had the pleasure of Sam coming and visiting me every few months at my house. And every time he'd come and tell me that he wants a bigger data center, and he got to a point where I saw that he's lost sight of reality. Like I said, I think we're going to have a $500 billion data center, and we're going to need all that compute. I'm like, dude, this guy was 10, four months ago, then he showed up at 50, then he showed up at 100, now he was on 500. I don't know, he's delusional. But what he's done, what Dario's done, what Jensen has done is just… Wait, but was he delusional or not? We had that period.
[00:01:24] I think so. I think so. He was visionary. I think him, Dario, Jensen, all of them have had a vision around how big AI can be. And I think we're all coming to it that this thing is way bigger than we expected. I think now my belief is that demand is insatiable. I have an interesting point of view. Like in 2004, I had the privilege of being at Google. And when I joined Google at that point in time, search was a consumer-driven technology.
[00:01:53] And there used to be 2% of advertising used to be kind of like digital. Now it's probably 75 to 80%. And you sat there and you couldn't anticipate that 20 years from now, this is going to be 80% of the world's advertising revenue. And every technology you've seen in the last 20 years has been consumer-first. Search was consumer-first. YouTube videos were consumer-first. Facebook, you take your pick. And these were built trillion-dollar companies, children's dollars of other companies.
[00:02:19] What's fascinating is we haven't seen the consumer revolution of AI yet. We're still seeing a bunch of enterprise coding applications. I think this is going to be so big from a consumer perspective. It's going to upend everything we do. It's going to upend what Amazon does in e-commerce. It's going to upend what Google does in search. It's going to upend what happens in the social media space. So all that's going to... Does that mean no bubble? That means we sort of keep going strong for the next few years? I think your bubble is a question of trying to separate the
[00:02:49] consumer demand, insatiable demand, and equity prices. You know, equity prices go in and out of bubbles. Or people get overenthusiastic at some point in time, then they get underenthusiastic. The market is always right once in a while. But every other time it's either over-exaggerating things or under-exaggerating things. So from that perspective, could we be in a bubble? I don't know. If you look 10 years out and look back, I think we all say these are all minor blips that we saw. Now, it is true that some companies will fundamentally vanish. And that happened in the Internet era, right? If you go back, some of you are not old enough,
[00:03:19] but if you go back to the late 90s, early 2000s, a lot of companies started the bang, some vanished. Google was the 13th search engine in the world. It wasn't the first. And it's now a $4,000 or $5,000 company. The first 12, you don't remember their names. We, the last conversation I think filled a void that we had, which was the physical world. There's a theme that I really thought we'd be talking about more today that we haven't so far, which is just security and the fact that there are, you know, I think you joke to be backstage.
[00:03:49] You're running a security company. It's not a joke. It's actually what I do. It's true. It's like this is an AI conference. I'm running security. I mean, but AI obviously poses perhaps like the largest security threat we've ever seen. Or I don't know. How would you score, you know, are the state-of-the-art models, is it fear-mongering that people are saying that they could find all these zero days? Or what's the level of security risk we face today thanks to the models? As you've seen, the sort of bleeding edge of the models are very good at solving hard problems,
[00:04:20] as you saw with Navier Stokes and similarly likes that stuff. And what you discovered, these models are really good on concentrated tasks and they're very good at 80% of a broad task. So you can focus them and put them on a very specific task, give them tons of compute persistence, and they'll go and find the solution for that problem. Or you can go try and do 80% of something, right? Like try and get to do 80% of your email and stuff like that. Get 80% right, 90% right, it's not 100% right. But you can focus on a very specific task. Now, cybersecurity is kind of interesting because
[00:04:49] you've got to find that one breach, that one vulnerability, that one attack bot they can go attack and attack somebody. So if you put enough amount of compute, put enough intelligence behind it, these models are pretty good at finding vulnerabilities. They're pretty good at finding vulnerabilities that they can daisy chain and try and attack somebody. So that's fundamentally changed the game from an attacker perspective. Historically, attackers and defenders somewhat stayed in sync because attackers were individuals who would have to spend a lot of time trying to come up to something.
[00:05:18] Usually you could catch them because they were trying to sort of lurk in infrastructure, trying to daisy chain stuff they couldn't get through as quickly. Now, in 18 minutes, enough compute and persistence can actually go find a vulnerability, daisy chain it, get inside, exfiltrate data and leave. 18 minutes is not long enough time. So yes, you know, AI is going to fundamentally change the speed of attacks that we see in cybersecurity,
[00:05:43] which requires a whole rewiring of the defense mechanism, both from an infrastructure perspective and sort of AI in defense perspective. What should startup founders do today? It depends on what you're trying to start up. What? Depends on what you're trying to start up. Like, but yeah, just advice for defending against the fact that attacks seem inevitable. I mean, it begs a sort of fatalism, which is like, I guess I'm going to attack. What am I going to do? Depends on what you're trying to do.
[00:06:11] Well, the fatalism probably is more in the realm of reality for people who have old infrastructure. They haven't upgraded for many years and they've been sort of sweating the asset. If you are running with new infrastructure, relatively new infrastructure, in the last three to five years, you actually have a chance of trying to consolidate all the data. So look, AI is a data problem. You hear that on a constant basis. If I wish I had better data, I had better synthetic data.
[00:06:37] There's tons of companies creating synthetic data because you're trying to train your models using data to be able to respond as quickly as you can in the defense world. So the first thing enterprises will have to do is they're going to have to put all the data in one place so we can train the data on saying, this is what normal behavior looks like in the enterprise. And then AI will be smart enough to figure out what abnormal behavior looks like. And you have to literally train it for each enterprise one at a time. And if I can train AI on how my enterprise operates traditionally,
[00:07:06] it should be able to suss out abnormal activity. The moment it susses out abnormal activity, it should be able to go out and investigate saying, what's going on? This looks different. That's where we need to get to from a North Star perspective. That does require a bunch of replumbing of infrastructure. That requires training models to be on the cyber defense side. I think that's going to be a huge opportunity. What are the attacks that you're most worried about over the next? Any.
[00:07:35] Do you think it's nation states fundamentally? Or what is the type of attacker that concerns you most at this moment? You know, the attacker that concerns you most, should concern you most, is who wants nothing in return. So if you look last year, there were north of 2,500 attacks. And majority of them ended up in ransomware. People said, I'm in your infrastructure. I've exported data. I've locked your machine down. Give me 50 million dollars. Or give me 30 million dollars.
[00:08:03] Now you know there's a price to solving the problem. Now you can get into a whole ethics conversation, get into a conversation about what's the right thing to do. What if I pay them and they don't unlock my laptop, so they don't give me my data back. Or they don't explain to me how they got in and help me remediate it. That's a decision every ransom attack company makes on their own. But if there's a price, you know. The problem is that there's no price. If somebody says, you're locked, you're done. Hmm.
[00:08:31] That can have a tremendous business impact. And that typically, the bad case scenario is that when it's a nation state, which is trying to create disruption. There's a colonial pipeline attack where people were worried that they could shut down the oil supply. Well, that was ransomware oriented so you could pay money and get it to be reopened. But there are others where the Ukraine war and Russia war happened that everything in Ukraine was attacked. They were shut down because they're trying to impair the operations of the Ukrainian army
[00:08:59] because that was what they were going to do in terms of sowing chaos in that situation. What's your read on the rate of improvement on state-of-the-art models? Like how much are we gonna have another shock where it's a sort of private sort of mythos level model next year? Like what's the rate of improvement you're seeing here? You know, we're already seeing that the most recent drop of Google's carbon, which was like last week. I think they launched hydrogen yesterday. They were launched carbon the day before.
[00:09:29] They're actually way better than their last generation of models from a cyber perspective. I think they're roughly at par with what we see from GPT 5.6 and GPT 6 at OpenAI as well as the most recent mythos 5.2 at Anthropic. And what's fascinating is when you put these models against checking for vulnerabilities, they have not that much overlap. Like we ran all three of them in a harness to test code and we
[00:09:59] discovered this is roughly accurate, not precisely because they keep moving, but roughly a third, a third, a third, right? And about 10% from open source. If I said take all the code Palo Alto has, point these models at Palo Alto and can they find all the vulnerabilities? What's funny is if you point each one of them, they find roughly about 30% of the vulnerabilities and then open source finds 10. So you have to run all four to understand what your entire sort of expanse of vulnerabilities is. Is it creating this have and
[00:10:28] have nots world where there's sort of, I guess to your advantage, a set of companies that have access to these models that are so threatening and then you know some of the startups in the room can't get access to them to build their own products? I think everybody's rate limited on these models for the most part, but I think they've become a lot better at access. I remember mythos in the early days, right? You're very restrictive list of about 15 or 20 companies had access and literally had to win the lottery or write a
[00:10:55] big check to get access to it. I think those days are sort of behind us because what's happened is you have multiple Frontier LLM companies which have the ability to give you the best model out there. So you have the ability to get access to the model. It's expensive because these are specialized models that they'd like to extract enough sort of do I need to recoup their training costs. So it is expensive. It's rate limited. So you can't do all, you can't like flood it with tons of requests, but there's enough access out there. Changing gears slightly, what is your
[00:11:25] philosophy or someone who's acquired a number of companies thinks about that as sort of a key component of growing Palo Alto networks? Like what is your philosophy both from someone here who's like, oh might he be interested in me? Like and someone who wants to learn from you in terms of how you think about it? What is what is your guiding set of principles in acquiring businesses? So look, the funny part is in cyber security every bad actor is trying to attack you in a new way. They're not trying to do it the way they did
[00:11:54] yesterday because we probably found an antidote to the way they attacked you yesterday. So we've got to be constantly on our toes looking for where does innovation live in the realm of cyber security? It's very hard to believe that my 7,000 developers at Palo Alto have every idea and can knock everything out of the park. There's approximately two to three thousand cyber security companies funded every year from everywhere from Silicon Valley to India to Israel and these are really smart people and they actually understand the problem really well
[00:12:23] and they're very focused on trying to solve that specific problem. So more often than not, I can only have one thesis on how to solve the problem. There's probably 30 pieces in the market in terms of how the problem could be solved. So my job is stay vigilant and say wait, if we haven't cracked and somebody else has cracked it I have a choice. I can rush as hard as I can to try and then beat them at it or I can think about it and say how long will it take me to beat them or should I go talk to them about being part of Palo Alto and that's usually our calculus. Our calculus is
[00:12:51] can I win by building faster or am I going to win by embracing them because we got it the wrong the first time? So it's more of a how quickly can I get products in the market? How quickly can I build a team around it? Is it better to buy than to build because it's going to take me a lot to build because I'm behind or whether I can buy and integrate and get out there much faster? That's usually the logic that goes in. Our industry is extremely innovative. So in the last eight years, I want to say we've bought approximately 40 companies
[00:13:21] and when we do we're fundamentally buying not just their progress so far we're buying the team and their strategy in terms of where they're going to take that product because you know and it's kind of interesting in enterprise you the hardest thing is customer acquisition as all the founders will find it's very hard to go convince a customer to buy something from you and keep spending money with you. So once you've acquired a customer eventually the debate becomes you want to be really out to enterprise company Is you have to be seller be able to sell a lot more to them? And if you've exhausted your current product line
[00:13:51] The question is do you have more product you can sell? So my job my product leaders to make sure we bring enough product which is cohesive which makes sense Which appeals to the same buyer and works better together as soon as my customers have exhausted purchasing my existing product? So it's kind of like a bit of a build your product portfolio to both generally you're trying somewhat uniquely to acquire companies where you're saying the management team of that company can stay and really run
[00:14:17] That product line or how do you decide or sometimes you keep your people running it or how do you decide? generally 99% of the time my people got their ass kicked by somebody out in the field who raised less money was extremely resource constrained It'd be silly to have them come and say now work for the guys who you beat So generally it's the other way around we acquire the founders they get to drive the product vision and the execution They may be part of an organizational structure can't have 40 founders report to me
[00:14:46] But usually they work for very smart people who understand the the desire for me to make sure These people get supported and they can execute fast But usually is the founders are running these things for us and we we do interesting things to keep them motivated interested Empowered to get stuff done. Is there a size? Valuation headcount progress do you prefer or obviously it's been a range, but how do you think it's a range? historically from a product perspective we
[00:15:14] Especially new innovation areas you think about 25 to 30 customers 50 customers proves product market fit shows that it's not just a feature that Somebody may not want in the future the fact that you can be able to convince real practitioners that you want to use they want to use your product is usually good um anywhere from 60 to 200 people give or take and Anywhere where they haven't gotten so much air out that i'm paying for customers Which I don't need because I have all the customers in the world They just need the product so I don't need to go acquire the customer again
[00:15:44] So then we roughly that's where it works. But again having said that we did buy Cyberark which is a 28 billion dollar asset where we literally Bought our way into a large category called identity, which we didn't play in So people look at it and say how do we build a footprint in a category? We don't live in I know i'm about to ask the most Invested person in this question, but how do you size the cyber security market today? I mean it's we're growing into it Yeah What should investors look to when they think about the the total size of this market?
[00:16:13] Look what's fascinating is the size of the market is about 400 billion dollars in revenue give or take it grows at about eight to ten percent a year I think it's not just the size of the market, but if you go back 10 years The share shifts are huge, right? There used to be a bunch of companies you want the endpoint category most of them don't exist anymore Right there's a bunch of companies in the firewall category from spala alto was born that don't exist anymore So what happens is in that market that grows at about eight to ten percent here? There's tons like over 10 year period
[00:16:40] You can see somebody go and the revenues don't grow at all and somebody's revenues grow at 20 to 30 percent So there's a tons of share shifts that happen in the market that's roughly so the way the market works um I think ai is going to create a whole new So space in the market in terms of creating new winners and losers because it's still a fully unsolved problem
[00:17:04] We haven't solved how to secure ai generally speaking. I think 90 of the coding Use cases out there are not secured. So everybody mostly who's using you know claw or cursor or codex or pick your favorite, you know AI coding tool they don't actually have a security capability around that Which is kind of a big it's a totally unsolved market that is an absolute market Uh, you know, you were an investor at softbank
[00:17:32] You've thought about the value of some of these ai businesses. Are you? Yeah, are you buyer on anthropic at 1.5 trillion? I think who said I think ali was sitting earlier talked about this Again, if you look at it from a five to ten year perspective I think the market will be served both by frontier llms and I see them not just as if you think about it If you look at the stack, there's compute There's services that sit on top of compute which are called them, you know, rag harnesses, etc
[00:18:01] And tools and then they're on top of that. You have applications, right?
[00:18:11] I think the market will be served by both kinds there are customers who prefer the full stack Delivered to them so they don't have to worry about the elements of the stack there are customers who believe I want more control I want to own the stack in some way shape or form. I want to make it sovereign
[00:18:40] I want to I want to manage the weights so you'll get different versions of people who want different versions of stack I think there'll be enough companies out there where there are small medium-sized companies with our companies which don't have the resources to Do their own diy on it who are going to consume a lot of these tokens I don't see any reason why both open AI and anthropic are not 10 trillion dollar companies five to ten years from now. Yes Look, I Sat with a very renowned investor
[00:19:08] Seven years ago at breakfast and he told me at that time there's no trillion dollar company He said to me in case in five to seven years There'll be multiple companies close to five trillion dollars And I sat there and said hmm weird It was a bit of the Sam Altman moment or I said what's he talking about? And he was right and there are about four or five companies in the four to five trillion dollar range right now and You know, I think it's perfectly feasible that we will see multiple companies nearing ten trillion dollars In our lifetimes and hopefully in the next five to ten years
[00:19:37] I I led off with you know sort of the tongue-in-cheek about You know you wanting to be the CEO or not, but what we do is this Jensen moment a Destabilizing moment for the tech industry and what what level of a loss is it or not? Like you think this is something that can be weathered? I think society weathered a pandemic Which could be in the end of all of us But we somehow all came out stronger we all sit here and we're talking about the next big revolution or lifetime So yes, I think time heals most things but
[00:20:07] Is there a short-term instability in the moment most likely because there's a whole bunch of parts of the ecosystem Which need to sort of re-examine, you know where their Champions are I think Jensen to his credit has been a phenomenal champion for almost everything AI He was equally generous at this time for people who wanted to build cyber security To people who want to build closed models to people want to build open models So I think from that perspective. Yes, you know We have we have sort of any I champion has gone sort of
[00:20:36] gone away for a bit and For that we'll have to make sure that the existing champions in the I world Continue to carry the baton and make sure that AI continues to thrive And I think there's many of them out there all the way from the leaders of the hyperscalers or people who run frontiers other lives when you think about your guiding principles as a leader like yeah, just What are they like for startups who are just getting into building their companies? What are the things you sort of wake up in the morning and you say I need to
[00:21:04] Embody this this and this or or does it change as your business changes or what what are those sort of principles? When I was at Google Larry turned to me once and said I know you're really good go-to-market and sales guy But Google is going to be great company because we have great product And that stuck with me since 2004 So when I joined Palo Alto I said one thing we're going to be amazing at is product Because if you continue to have great products, we will continue to win and if you look at silicon valley, there's a whole
[00:21:35] sort of Litany of companies which have died because their products became irrelevant or customers didn't want them So I think first and foremost the most important part is you have to have the best products in your category Whatever you choose a category to be you cannot build the best product without the best people You have to find the best people right? You know I have no shame in going out to a founder who's got a 50 people company where I believe they're doing something way better than I am
[00:22:02] And I'll take the meeting myself. I meet five to ten founders every week In different ways shape or form off the 45 companies 40 odd companies I've personally sat across the founder and bought 37 of them Because They're smarter than us. They got it right. I need to get inside of the head I need to understand their creativity and understand their motivation And then I need to find a way of getting them to pal auto and keeping them motivated That means that you know, I have to adapt the company
[00:22:29] I have to adapt the culture to embrace them in some way shape or form I should how have you tried to change the culture at the company? Look when I came to pal auto I was very blessed that my predecessor was a high integrity west point graduate He'd set up the culture in a very phenomenal way and I always believe companies take on the form of the leader Right when I was at google it felt like it was a company run by Larry and Sergey when I soft bank It felt like masa's company and you know Hopefully pal auto feels a bit more like my predecessor and mine because we somehow people absorb your values
[00:22:58] So what I did was I had 5,000 employees come back to me and say Tell me about the culture. What do you like about it? What are you pride? What are you proud of? What are things you do differently? I had 11,000 responses. I read them all We came back and said everybody overwhelmingly wants five attributes of our company It was around integrity and customer focus around disruption around collaboration, etc. And that became the company's values because The company values cultures are not just me
[00:23:25] It's how every human being at the company represents themselves to everybody else outside So I got a beautiful note three days ago from a customer Who decided to write a eulogy to this guy who works for us like this is an amazing guy He did the following things and when I read it and I reflected It reminded me of the notes which are written when I joined Palo to eight years ago So if you can get the team to represent the values the way the company describes and that's success Thank you so much for joining us. Thank you very much Eric pleasure
