Thriving in the Era of AI: Essential Innovations with Cisco’s Jeetu Patel

Patrick Moorhead: The Six Five is On the Road here in Los Angeles with Cisco. It’s been a great event. And shocking, Daniel, we’ve been talking a lot about AI whether that’s networking, security, data. And we even heard compute today.

Daniel Newman: Yeah, we saw Tom Brady which by the way was quite a bit of fun. And, of course, we are seeing this evolutionary period. You could call it revolutionary because it’s happening so fast that we’re really seeing the practical part of AI entering aggressively into enterprises. And companies like Cisco, who we’re here with, are really focusing on that enterprise. I mean, talk about a company that has deep roots inside of big enterprises. And yes, we hear all about the cloud and the hyperscalers but where’s all the AI going to need to happen? Where’s all the data? So this is a really great week, a lot of learnings here.

Patrick Moorhead: So let’s talk about the product strategy. And you all know I’m a recovering ex-product person in the industry. I’d like to welcome Jeetu. Jeetu, great to see you.

Jeetu Patel: Good to see you, Patrick.

Patrick Moorhead: You rolled out your philosophy of products. That speaks to all us product people and ex-product people, recovering product people like myself. Thanks for coming on the show.

Jeetu Patel: You can take the guy out of product. You can’t take product out of the guy.

Patrick Moorhead: It’s hard, it’s very hard.

Jeetu Patel: You will always be a product person, my friend.

Patrick Moorhead: Yes.

Daniel Newman: I don’t know what you’re talking about when you say recovering. We only track the entire industry, every technology company, and platform, and spend most of our life looking at it, talk … Evaluating it, sharing our perspectives on it. You’re going to need a new line.

Patrick Moorhead: It’s like a product smorgasbord being an industry analyst.

Daniel Newman: Not when you always tell people you used to have a real job.

Patrick Moorhead: That’s right.

Daniel Newman: That’s funny because you’re an analyst.

Jeetu Patel: A little bit of trivia. I used to be a product person now an industry analyst. I used to be an industry analyst now a product person.

Patrick Moorhead: Gosh.

Jeetu Patel: It was just swapped.

Patrick Moorhead: We just did the high five mid-

Daniel Newman: I don’t know that I’ve ever really seen that evolution, especially with such a rapid ascent into such a big job. Gosh, that’s maybe-

Patrick Moorhead: Love it.

Daniel Newman: Something we can do some real work. Just kidding, Jeetu. So your passion’s contagious, I’ve listened to you. You did the cameo yesterday, you did the big keynote today. You did some great work in front of, actually, the analyst crew in the Q&A yesterday. All in on AI is my sense. I wouldn’t say it’s a pivot, but I’d say that Cisco has been thoughtful, it’s been a bit patient, it hasn’t rushed to AI wash everything it’s doing. The moment is now. Talk a little bit about how you’re thinking about that.

Jeetu Patel: One of the lessons that I’ve learned in this is, when there’s a massive platform shift and a megatrend that happens in the industry you should never fight the megatrend, you should always use it as a tailwind. You have to know the difference between when is it a megatrend versus a hype cycle.

Patrick Moorhead: Yes.

Jeetu Patel: Right? Web 2.0, it was completely ridiculous what was going on there. You could sit down and talk to someone for 45 minutes and I had no idea what they’re talking about or what Web 2.0 is. With AI you didn’t have that. The reality is, is when there’s something that’s a seismic shift that’s going on you have to be the first to market, right, you have to make sure that you go all in. But you have to do it in areas where there’s permission to play for you and that … Where you’re going to be going in is going to be logical for customers for you to participate. And so for us, that was in a few areas. Can we make sure that we one, use AI embedded in the fabric of every single one of our products? That’s the obvious one.

Number two. We are in the era like how the gold rush was where the people that made the money in the gold rush were the ones that actually sold picks and shovels. If you think about Cisco, we provide AI infrastructure to power AI. And so that’s a pretty important dimension that we had to get perfected and right. And we didn’t want to announce it until we actually had something meaningful to talk about rather than just … Because otherwise you lose credibility when you just start spouting things that are going to be three years out. You lose a lot of credibility. So we wanted to make sure that we stayed patient and we were … We kept building. And we said, “Now when we are ready to go out and make sure that we release it let’s make a big moment out of it” which was today.

And then the third area, which I think is going to be so consequential to society, is around this whole notion of securing AI itself so that AI safety is a pretty big deal. And I think that’s going to get bigger and bigger and bigger of a problem as time goes on. And I think you’re going to have to put some guardrails. There’s going to have to be some level of intervention … Not intervention but a degree of involvement from the government for policy setting because it can’t be a free-for-all. You have to make sure that there’s some degree of consistency so AI doesn’t take a life of its own. AI is in service of humans, not the other way around.

Patrick Moorhead: Yes. A very wise and pragmatic approach, particularly when you’re addressing enterprise, that they aren’t the first movers on these things. They never want to be first which is why right now our research suggests … And I think I heard glimmerings of this on stage this week is that there’s a lot of experimentation going on that are moving into POCs but nowhere close to mass adoption of these technologies. I want to dial this out a little bit. First of all, congratulations on the Chief Product Officer role, this is great.

Jeetu Patel: Thank you.

Patrick Moorhead: Nothing integrates multitudes of different technologies into a consistent platform than having a Chief Product Officer. I think this is a super important role. The product development process in rational, it’s a long-term. You have to make long-term bets, especially when you want to change architectures. What is it that you’re doing today to deliver AI value right now?

Jeetu Patel: Most of the things that we announced today are all available in market either now or within a matter of a month or two. Like I said, these are not overly futuristic. And that was actually part of the reason why we waited for this long. We didn’t want to go out and announce something that was going to be a year away.

Patrick Moorhead: Right.

Jeetu Patel: So what are we doing? At the basic level, we’re saying, “Let’s make sure that AI infrastructure is plug and play.” We simplify AI infrastructure so that when organizations need agility in their data center we can provide that to them with having the right level of compute, the right level of networking. There’s been conventional wisdom in AI for the past few years but three things are really important. The first one is compute, the second one is models and algorithms, and then the third one being data. We’ve always known this but the market’s now completely turned around on this one and believes exactly what we believe which is networking and security are just as important as compute models and data. And so there’s a five core foundational components that are needed, compute, model, data, networking, security, in order to make sure that you can power AI to its fullest potential. And so we need to make sure that we deliver that in the biggest way possible.

What we announced today was a low latency, high performance switches that are power efficient. We announced today that we are going to make sure that we have training servers for AI. But what we’ve also announced is we’ll have inferencing workloads that can actually be serviced with AI, with AI pods, with a varying level of classes of configuration that we can go out and deliver. That’s a pretty important area to get into because most people want to spin up capacity within a very short order. And if you think about the delta between the training dollars being spent right now in inferencing … Inferencing is going to have a huge hockey stick curve. Because right now there’s about $200 billion spent in training. Guess how much is being spent in inferencing? Five billion.

Patrick Moorhead: That makes a lot of-

Jeetu Patel: Not spent, actually getting derived from inferencing. So how much revenue is being derived? It’s about five billion, five to 10 billion. And so I do feel like there’s a tremendous opportunity for having the right level of simplification in AI infrastructure so that as that inferencing volume goes up we are right there to go take it. I don’t think we are in any way, shape, or form over here late in the market.

Daniel Newman: Our data suggests something very similar. The enterprises are adopting at a much more conservative pace which I think all of us have alluded to. They’re all very interested. A lot of them are experimenting, some in the cloud.

Jeetu Patel: That’s right.

Daniel Newman: But the thing, Jeetu, that makes so much sense about what you’re trying to do offering inferencing pods that are available … Can be put on data center infrastructure of these enterprises is … That’s where a lot of the data is.

Jeetu Patel: That’s right.

Daniel Newman: We all know that energy costs, egress, and moving data, is incredibly expensive. It’s not sustainable at scale. And then on top of that, we also know that there is going to be an inverse of that training and inference number at some point. The training will stay but inferencing is going to scale at probably a pretty rapid inflection, right?

Jeetu Patel: Yeah. I think right now if you think about a GPT-4o1 class model, it costs between three to $5 billion to train, right? It used to be about a billion dollars for a GPT-3 class model so it’s actually gone up quite a bit. We are now at a point where we are actually running out of publicly available data, that’s available for free, to go out and train these models. However, there’s about 150X scale proportion of enterprise data that is not training these models yet that is available in the market that every company in the world is going to want to go out and want to use for their own purposes so that they can actually create a unique differentiation in the market for themselves. And that’s going to require a bunch of scaffolding, and a bunch of infrastructure, and a bunch of plumbing that you need to make sure that you deliver to the market. I think a company like Cisco, we do the boring work so that our customers can do the really exciting work.

Daniel Newman: Picks and shovels. So they all maybe won’t-

Jeetu Patel: They’re the picks and shovels folks.

Daniel Newman: Hit the goal. But the funny thing-

Jeetu Patel: That’s right.

Daniel Newman: Is we are going to see, and both of us I think have written about this, smaller models. RAG and-

Patrick Moorhead: Fine-tuning.

Daniel Newman: Fine-tuning,

Jeetu Patel: Exactly.

Daniel Newman: Lots of optimizing. And companies want to take that right to not use all that unnecessary data. Lower energy, more efficiency, better answers. And we’re seeing it with what’s in the market right now with these open-source models for sure.

Jeetu Patel: The future as it gets more agentic, as there’ll be multiple agents that talk to each other and actually get work done. You’re currently at a point where AI is conducting certain tasks pretty well. But what we’re not doing right now is conducting full end-to-end workflows and jobs really well. That’s going to happen over time. And as that happens you’re going to need more and more infrastructure. And you’re going to need that infrastructure to be plug-and-play.

Daniel Newman: I want to pivot. You talk network security. Big concern, safety, security. You’ve got Hypershield, you’ve made lots of investments. We’ll be talking to Gary Steele with Splunk. You brought him and we’ll talk about that. But give us the posture on the security side and its role in AI.

Jeetu Patel: So security’s role in AI is going to be a very strategic one. Because right now, if you think about it, one of the biggest concerns that most enterprises have … The reason why enterprises haven’t gone all in and taken 15,000 apps to production within the enterprise right away is because they want to see what the adverse effects might be of this thing, and they want to make sure that they have some guardrails in place that might actually … For companies that have to be highly compliant for regulatory reasons, the companies that have to have things that they need to worry about for a multitude of different reasons, they just have to be cautious as they’re moving forward. The way that we see it is AI is going to need to have some guardrails. Security will play … Will have to get pretty deeply embedded. When you’re building an AI application, I don’t think you can rely on some base level of security from the model providers because you’ll have a dozen models.

Patrick Moorhead: Sure.

Jeetu Patel: So just like you have multi-cloud you will have multimodel architectures in place. And as you have multimodel architectures you need to have a common substrate for security that goes across all the models. And you don’t want to have the application developer thinking about building security stacks every single time they build an application. So the only option remaining then is if you … If you say that the model providers aren’t the ones who provide security as the best place, and if the application developer doesn’t have to build security every single time, then you need to have an independent security company that actually can provide security across all the models. You should expect Cisco to do more of that.

As data center architectures evolve, because of the scale proportions of AI, you will see that you will have to rethink security architectures from the way that they used to be, from highly monolithic and perimeter centric to highly distributed. What we’ve done with Hypershield is a first of its kind and it’s a category creation moment. And we are being very cautious in how we move forward because we want to make sure that we take a few customers, get them hyper-successful, and then you go from there. I could tell you that the feedback and the response we’ve gotten from the market … In my 30-year career, I have never seen this response to any product that I’ve launched.

Patrick Moorhead: No, it is provocative-

Jeetu Patel: It is.

Patrick Moorhead: For sure in terms of what it does. It’s hard enough to work on products that are delivering value today and in the near future. How do you look at the future? Things change every six months, three months. Some people say every month. Enterprises want to be able to rely on companies who’ve got a good vision for the future and can move back to protect their investment. I’m curious how you’re looking at different opportunities and threats to AI into the future.

Jeetu Patel: I think there’s a difference between being reactive to market movements and being responsive to market movements. And the way to think about this is you have to play the long game. And I think Jeff Bezos has a great learning that he shares with people which is “Don’t ask what’s going to change, ask what’s going to remain the same over the course of the next 10 years.”

Patrick Moorhead: Right.

Jeetu Patel: And so when we start thinking about that and say, “What’s going to remain the same?” There’s going to be a continued demand for more compute and more infrastructure for training models and doing inferencing on models because those volumes are going to go up.

Patrick Moorhead: Right.

Jeetu Patel: That seems like a pretty safe bet.

Patrick Moorhead: Sure.

Jeetu Patel: So being in the AI infrastructure business seems like a pretty safe bet. There’s going to be continued level of sophistication and compounding of the sophistication of threat actors wanting to do bad things at a nation-state level, at an individual level. And so going out and making sure that you can put guardrails around AI seems like a pretty safe bet. There’s going to be a continued level of development of AI that’ll simplify user experiences. So having AI natively baked into the products, a pretty safe bet. If we just take those three things, that’s a pretty long-term strategy for us to pursue. And yes, the mechanics might change a little bit but the core thesis doesn’t change and that’s what’s so important.

Patrick Moorhead: I appreciate that.

Daniel Newman: Yeah, you’ve got to keep moving data, you got to keep processing data, right? I do think we’re going to see some major changes in the way we interact with data, and the applications, and rendering, and generative tools but it’s going to be really fun, really exciting. It’d be really awesome to have you come back and chat to us some more about this, Jeetu.

Jeetu Patel: Thank you for having me.

Patrick Moorhead: Yeah, it’s great to see you.

Daniel Newman: It maybe your first time but let’s not let this be the last.

Jeetu Patel: I’m a big fan of your show, and I’m a big fan of both of you on-

Patrick Moorhead: Thank you.

Jeetu Patel: Twitter, and on social media, on LinkedIn and so it’s a pleasure to see you. And we’ve known each other for a while so it’s not like we’ve not known each other. But I’d love to come back again. So thank you for having me.

Daniel Newman: Let’s have you again soon, Jeetu. And let’s have you back again soon. We appreciate all of our community being part of all of these episodes here in Los Angeles with Cisco and, of course, on all of these Six Five shows with Patrick and myself. But for this episode and for Patrick and I, it’s time to say goodbye. See you all later.