event last week, didn't you? A little nice event in Napa.

I did.

We had a bunch of CIOs, not just one bank CIO, like Chamath said.

But before we jump into the end of days, I do want to say something to what you guys were talking about earlier, about the open source versus the sort of the hyperscalers providing influence.

So we had this peculiar problem.

We'd like to embrace AI as quickly as many of the startups.

And we've gone through all the understanding, all the analysis, open source versus cloud models.

When did you first learn about this? And how severe is it? Is this an acute issue? Or is this a harbinger of things to come? And is this going to happen every 10 days for the next year, as these models have gotten incredibly powerful at what we call cyber, which really means cyber security, not just cyberpunk, but cyber security?

Right.

So that's not a practical example of how models use.

So what you do is you take a small language model, you compress it, you make it task specific, you keep, keep training it until it gets to 99.9% accuracy and say, "Instead of using it, using machine learning algorithms, we're gonna use generative AI, small language models, and they're gonna do that" yeah.

So my point is what we're doing is we're trying to figure out what tasks requires what level of intelligence, and what is the cost of that intelligence for that particular task? And you'll suddenly discover you don't need a Ferrari for every task.

And by the way, the customers don't want to pay for a Ferrari for every task.

I'll give you the flip side.

Flip side is let's assume you're undergoing a breach.

[laughs]

What are best practices around navigating

this? There are clear rules of engagement in the industry.

The clear rule of engagement is, if you find something, the first rule is you've got to tell the person whose code it is.

You've got to go back and tell them that, listen, I found something wrong in your infrastructure product or your software product.

Here's a chance to go fix it.

If you find it in SaaS software, you call the company.

You find it in stuff like this.

But what's your view?

You agreed with what he

wrote? You said as much.

Yes, of course.

Look, I also endorsed it, and I said, look, we cannot deploy AI if we don't get the observability, if we don't get to look at how AI is operating.

It cannot be, trust me, I fixed it.

It's not going to work.

Because, look, every industry has this notion of verifiability, the notion of explainability.

They're thinking about the worst-case scenarios that could exist, and I don't want to enable that with something that could take my job down the road in my backyard.

I mean, do you not take them seriously when many of these safety people inside these labs seem to be very concerned?

Well, look, I've got to-- First of all, thanks for having me, David.

I've got to go reinforce what Jay just said.

The focus needs to be on increasing the pace of deploying this technology safely and securely.

The challenge is not to slow the tech- technology down.

The challenge is to make sure that we put the right safety and security harnesses around this technology, and I think that is immensely doable.

It's as much of an engineering problem as it is to build these amazing frontier models.

Yeah, go ahead, Nikesh.

When I

saw it come out, I figured that clearly the AI labs understand this is a problem.

And I think we all do.

It is a possibility that if not managed right, we should not be in a haste to release these models.

So I think that's a good idea to make sure you get it right.

I think the notion of a ninja move I thought was, well, maybe they're going to try and get the sympathy of the regulators and sympathy of lawmakers saying, hey, wait a minute, this thing's going to be dangerous.

Let's make sure we jump in and we put some guardrails.

But are you concerned about the regulatory capture side in any way? I think

What does that mean?

Well, Jim, I've been listening very carefully to all the rhetoric about what should happen, should AI be slowed down, and what are the bad effects of AI.

AI is an amazing technology.

I think what we have in front of us is once in a lifetime opportunity to embrace something that is so, gonna be so impactful across the board.

And I don't see how it's gonna slow down.

I don't see researchers not wanting to win, as I've said.

There is no competitive sport where I found people saying, "Oh, wait a minute.

Right.

What actually changed? What was, what was the conversation like and what what, How did that process go? How much of a boogeyman was open source? Because it was discussed early, but there were a bunch of different takes.

Look, I think what has happened is even in the last six months, you've seen the capabilities of these models continue to rise.

I read something that two days ago, Sam said that Astra...

Their new model is so powerful that they're going to have to clip its wings and trim it down.

That gives you a precursor of things to come.

I'm sure those capabilities will be caught up with, whether it's with other frontier LLMs or the open-source community, the open-source models or Chinese models, for that matter.

I think the capability is going to come out as hard and fast.

Yeah.

How can we stop them? What is Palo Alto doing to stop them?

Good news, Jim, that we have been very consistent in our approach of making sure that we build an AI-first product.

You and I have talked about this over the last few years.

We built a brand new product called XIM.

We bought a company called Chronosphere, which focuses on AI-driven observability.

We are focusing the AI security operations and IT operations space.

We've always had this ethos that AI was going to be big.

Since then, what has your life been like?

How does that come together in this role here so well?

If you look at what the markets reward, if you look at what, how business gets rewarded, business gets rewarded effectively in metrics like market cap, perhaps.

What is the market cap? The market cap is the sum total of the expectations of world about the strategy, execution, and the potential for your business, right? Nvidia trades at 5 trillion because people believe it has trends of demand that's gonna happen.

Jensen is a great executor, and they're doing a whole bunch of stuff right.

So you can actually work your way back from market expectations, what the market expects, that makes for a successful business.

Now, the market is pretty straightforward in its expectations on some level.

It says, "If you have a durable business that grows at a robust rate, which you can run profitable and generate tons of cash flow, we like you." Now, that's great, which means you have to run a good business, must have good cash flow, and must grow well.

[laughs]

You didn't know

... you didn't know it was bad until it got into your infrastructure.

So the question becomes, if you know it's a known bad, you stop it at the perimeter.

If it gets through, how quickly can you find it and stop it before it creates harm or damage? So from a cyber perspective, you wanna be in the perimeter business.

You wanna be on as many perimeter endpoints in the world as you can because that becomes a sustaining business.

The more perimeters I'm on, the longer my tenure for my business is.

So I'm on endpoints, I'm on, you know, devices, I'm on servers, I'm on firewalls.

They'll pay almost anything.

I, I really f- wanna know what you feel about it.

Look, as you can see, a lot of, lot of frontier model companies are flexing.

They're showing the immense capabilities of what AI models can do, and this is true that we had an incident where an AI model using agentic capabilities was able to escape its sandbox and be able to go ahead and attack sort of infrastructure of, effectively, Hugging Face to go look for, uh, exploited sort of capabilities.

So, I, I think what it tells us is, one, AI models are gonna be s- more and more capable over time from a cybersecurity perspective.

I think, two, it's a lesson for our friends at Frontier LLMs to make sure that before they create capture the flag exercises to go find something in their external infrastructure, they should probably make sure that their sandboxes are secure and they are not suffering from any vulnerabilities before they start pointing the models outside.

I think from a more industry perspective, as you saw, what we have to do is we have to start fighting this notion of AI with AI on the defense side.

Nikesh, what is the token economics consideration where AI is increasingly used in, in the, in the form of swarms of agents in defense?

Look, tokens, I've said, I said a few weeks ago that I think tokens are overpriced.

the catches david i know you've also said in the past you feel as though the consumers being subsidized by the enterprise when it comes to token economics i'm curious as to how you see that playing out over time

Well, if you think about it, David, as I said, the demand for AI is infinite.

I don't know how many times I interact with an AI chatbot in a day.

Probably a lot more than I was six to 12 months ago.

Who's paying for it? I'm not.

And those models are not advertising to me.

So as a concept, all this compute that is being used by consumers around the world, when we tout 2 billion or 3 billion people using all this AI capability, whether it's embedded in our social media applications, whether it's embedded in our efficiency applications, workforce applications, or we're using them for deploying OpenClaw across all of our different things we do.

I just wonder, from your perspective right now, with the CapEx budgets that we're seeing from the hyperscalers, with the raises of capital that we're seeing from the likes of companies such as Alphabet or SpaceX or Amazon the other day in the bond market, how do you view this in terms of the overall return on invested capital that we're going to actually see from this investment, the likes of which we've never seen?

Can you explain that to me?

Yeah, look, I, I've been r- listening to some of your podcasts, and I've been paying attention to what happens to market, because I want to understand where all this settles down, not just because, you know, I wanna understand it, but it also impacts how I build my own business.

So you got these phenomenal frontier models, and they keep sort of leapfrogging each other.

Every few days there's a net new model that's delivered by OpenAI or by Google or by our friends at Anthropic.

And the question becomes, okay, fine, these models are moving at this pace.

What do I need to build? What do I need to do with these models? And as we came to that Mythos moment when everybody was busy chasing Mythos, and that was kind of important, you realize even the best model has a high false positive rate.

But for some reason, in the consumer space, we don't seem to care.

What is true and what is right? Do you have to have FDs to sell into enterprise?

Go, Nikesh, start.

Well, first of all, thank you for having me here.

I think AI is exciting.

I think it's exciting to see all the stuff that's gone down in the last possibly 24 months.

I think Sarah just said it.

They were right in anticipating the huge amount of compute that was going to be needed.

So all that stuff's going on.

Do you think that there's a world in which these models become so good that you could see yourself advocating for more nationalism around how they're controlled and how they're managed and where we point them? Or do you think there should be maybe a set of these models that never see the light of day, that only the NSA and other folks could have access to, or guys like you?

[laughs]

[laughs] Well, uh, first of all, thank you for having me here.

I think AI's exciting.

I think it's exciting to see all the stuff that's gone down in the last pos- possibly 24 months.

Um, I think Sarah just said it.

They were right in anticipating the huge amount of compute that was gonna be needed.

So all that stuff's going on, but you can see that, you know, this, this notion which we talked about briefly last time, that AI is really democratizing intelligence.

How do you maintain the discipline, or do you see yourself at some point considering things that are not nearly so much right down the middle of, of cyber?

Yeah

...

I can use a smaller language model.

I can use an open source model to solve that problem because it's task specific.

So we're trying to be judicious about the use of AI, but at the same time, I think we're not using it as well as we should be using it.

I don't think anybody in the world, except unless you're a frontier AI model company, has got it perfectly right in terms of how we need to use it, so we've got to get up the learning curve, and we'll get there.

I think the, the part which I was trying to say to you earlier, I think it's a very important point.

Yeah.

We, we got machines all over the place, and they've gotta be stopped, and I don't know who else has a solution like you.

Well, Jim, look, I think what we've seen, a lot of people have discussed Mythos, uh, in different public environments, and I think what is very important to understand, AI has the capability to understand bad code.

AI has the capability to take vulnerabilities, stitch them together, and figure out a way to attack cyber infrastructure.

Now, we have to find antidotes to that.

There's no easy way around it.

The antidote in the short term, there's a lot of testing, a lot of blocking and tackling, which I think a lot of people are trying to do as we speak.

Lots of consulting companies that are involved, cybersecurity vendors involved.

there already are?

Casey and I have talked with people at these companies for many years, usually because something terrible has happened, and I feel like the vibe we get is like, "This is the worst, most dangerous time ever in cybersecurity." What is your subjective experience as someone who's been in this field for a long time?

I'm a little more perhaps relaxed than what you're trying to ascribe, that people come here tell us it's the worst moment.

Historically, what's happened is, in the last seven years, you've seen the time from somebody breaching an organization and being able to a- extract, we'll say, crown jewels, has been measured in days.

Unfortunately, with the emergence of AI, the arrival of vast technologies, that timeframe has shrunk down to minutes.

And when that happens in minutes, your defense systems have to be able to be activated and defend yourself some minutes.

And fundamentally, the cybersecurity infrastructure was designed for days.

Some parts of it are making it to seconds, the good parts where you know how to stop them.

[chuckles]

Right.

But remember, cybersecurity is something different.

What we do, what the industry does, us and other players in the industry, we build the constant scaffolding required to deliver cybersecurity.

So we are sitting on the endpoint.

We're sitting in your data centers.

We're sitting in your applications.

We're sitting in your cloud instances.

What this maybe feels like is that someone who may not be all that skilled may not realize they can click on something, and that could end the practice of whoever they're working for.

How do you respond to that?

Well, first of all, you know, I'm really excited that more people are trying to solve the cybersecurity problem.

So, uh, I do think there's a bit of hubris around AI, that people believe that AI is the panacea, will solve everything that we haven't been able to solve in the past.

Um, I think all players in cybersecurity, if you see at RSA, the entire rage is AI.

Everybody's putting agents into AI.

Everybody's trying to put generative AI into work so we can solve the problems in training our models.

So I think what he's telling you is that the customers are demanding more AI be used to solve this problem because we need to shorten the time to remediate these attacks because the old method's not gonna work.

Isn't that a harder competitor to fight than another cybersecurity firm?

Mm-hmm.

But it becomes part of your sort of muscle memory, and I think that's what happens.

I'm a huge student of business.

Spend a lot of time listening, reading, talking to people about this stuff, and you just get muscle memory over time.

You just kinda get a sense of what the right thing to do is.

And I think the part which we sort of underappreciate, people who do well are people who go back to first principles and try and abstract and say, "From a first principle basis, is this important? Is this not important?" For example, I don't have to manage my stock price.

Mm-hmm.

There's a lot of players out there.

So what is the role that sovereign AI plays in cybersecurity?

So I think there's two questions.

I think if you look at the evolution of foundation models, if you look at the amount of money that's being spent, it looks like it takes tens of billions of dollars to train the next impressionable model.

If you look at it, there's actually three large companies that are doing that sort of activity, and they're spending tens of billions of dollars and raising tens of billions of dollars.

So if you read the math, let's say $650 billion are going to be spent in CapEx in the US from technology companies, half of that to deploy these models, and a majority of that to build and train these models.

So if one wants to be in the models race, you have to make sure you're willing to commit that kind of capital.

The idea that you can build models with a lesser amount of capital, maybe you can be 20% more efficient, 40% more efficient.

That's

They see value clearly in what Palo Alto Networks is offering.

Well, remember, observability is beyond the model, right? The model, LLM people are focused on making sure their model is amazing, it does all the things that you expect it to do.

But observability is to look at it from an end-to-end perspective to see how do they make sure that the customers of theirs who are using these things and the usage continues to double almost every quarter, How do they make sure that all their systems are low latency, are working perfectly, and there's nothing in the entire end-to-end delivery chain that is underperforming? So that's what observability does.

The reason they're coming to us is because they're busy building the best models, applying their AI research to the models.

We're busy with Chronosphere making sure that the end-to-end proposition works, and there is very low latency, by the way.

This is true for not just LLM.

This is true for every consumer app that is out there, whether you take Uber or DoorDash or YouTube.

The idea is greater visibility into AI threats, but just succinctly, what problem did that solve for you on the team? What are you getting?

Is that strategy working to convince software companies in particular that it's better to come to Palo Alto Networks than do business with a handful of different players?

Yes, of course, it is working.

We call that platformization at Palo Alto.

We continue to add 50 to 75 customers every quarter and sometimes more in about the fourth quarter.

But I think more importantly, if you look at the evolution of technology, almost every industry vertical in technology has started that way.

from the application perspective.

People had 15 or 20 different vendors that did the entire solutioning for CRM.

Are we or are we not in an AI bubble?