MWC Barcelona 2026 | Jeetu Patel, Cicso

Clips
More from MWC Barcelona 2026

Jeetu Patel

President & CPO

Cisco

play_circle_outline Altman, Jensen at AI Summit: 24/7 Agents Reshape Telco Infrastructure and Monetization via Interoperable Fabrics

play_circle_outline Token-generation and tokenomics: compute and network central to competitive advantage

play_circle_outline Building Trustworthy AI: Infrastructure, Context Enrichment, and Security Guardrails Against Rogue Agents

play_circle_outline Cisco's unified edge: hyperconverged network, compute, and security in single managed box

play_circle_outline Massive data center re-racking, repowering, and CapEx—demand still outstripping supply

play_circle_outline Networking as connective tissue/OS for AI factories: scale-up, scale-out, scale-across

Info
Transcript

Jeetu Patel, Cicso

Jeetu Patel

President & CPO Cisco

In this interview from MWC Barcelona 2026, Jeetu Patel, president and chief product officer of Cisco, joins theCUBE’s John Furrier and Dave Vellante to discuss the massive infrastructure refresh driven by the transition from simple chatbots to agentic AI. Patel explores how telecom operators are repositioning themselves at the center of a secure global connectivity fabric to become "token-generation factories." He breaks down the three primary constraints facing the AI revolution – infrastructure, trust and context – and explains Cisco’s strategy to provide t... Read more

explore Keep Exploring

What are the implications of the recent AI summit for telecommunications operators—how will AI (especially agentic AI and the unified edge) affect telco infrastructure, and does it present opportunities for monetization or threats they need to address? add

What will fundamentally change right now, and what will determine a company’s or country’s competitiveness as AI-driven token generation scales — including the roles of network infrastructure, security, observability, and telcos? add

How has your organization shifted to prioritize AI, and what do you see as the major constraints and safety/security challenges in deploying AI agents (and your company's role in addressing them)? add

How should AI inference, token generation, and policy enforcement be managed across distributed edge nodes and data centers, and what are the security implications (for example, does hyper-convergence reduce the attack surface)? add

How will the rise of AI change data‑center infrastructure and capacity needs — including power, cooling, space, networking and multi‑site GPU clusters — what are the expected costs and timelines, and is there sufficient demand to justify the large CapEx (or might this be an overbuilt, dot‑com–style bust)? add

How should networking evolve product‑wise to support the convergence of carriers and enterprises (as seen at MWC) — specifically, how do you enable coherent GPU networking, scale‑across “ultra‑clusters,” and the necessary chip/ASIC and data‑center interconnect technologies (e.g., deep buffering) so networking can act as the connective tissue for AI workloads? add

bolt Powered by CUBE AI

Jeetu Patel, Cicso

search

>> Hello, I'm John Furrier with theCUBE with Dave Vellante, kicking off Mobile World Congress, MWC, 2026. It's our fifth year. it's turning into an AI show, where telecom operators and carriers have an opportunity to reset with AI. It's turning into quite the AI show, super computing show. All the AI action's here. Jeetu Patel is the president and chief product officer for Cisco, CUBE alumni. Friends of theCUBE. Cisco, the leader in telecom service provider business over the years. Thanks for coming on.

>> Thank you for having me. I think I've never been to a Mobile World Congress and not been on your show, so it's great to be here.

>> It's great to have you on. I mean, obviously, you guys just recently had an AI summit in San Francisco. Jensen was there. Matt Garman from AWS, all the leaders. It was a great event. Check it out on the internet if you haven't seen it, it really set the agenda. It was an agenda-setting, North Star, very directionally, "This is where the world is sailing to." AI is here. The show here, the pretext to our coverage is the AI revolution's coming to the telco infrastructure operators and the carriers to, essentially, not to miss the window. They have all the infrastructure in place and it's poised for an upgrade. Unification is the term, you guys have a unified edge. All the discussions around how to shift from legacy voice-optimized infrastructure to data. This is a huge threat. If they get there, they will open up monetization, new levels of services, all network-based. Your opinion?

>> You said it all, so we can wrap it right there. This is an exciting time in the industry in general. The AI summit that you talked about, we were actually trying to test people's stamina for being able to watch fireside chats. And we realized that at 12 hours, there was still no saturation of appetite because you had the kind of caliber of people that we had. And we opened with Sam Altman, we closed with Jensen, everyone in the middle. And the reason for this kind of excitement right now for telcos is as you move to this second phase of AI from chatbots to agentic, the entire infrastructure is going to have to be rethought because these agents are going to be working 7-by-24. They're going to be very data hungry. They're going to actually be talking to each other. And the ability to interoperate and connect between the data center where the digital workers live and the campus branch where the human workers are is going to be done through a secure global connectivity fabric, which is, essentially, the service providers. And I think they have a whole different opportunity to monetize. They'll be able to have tiered services. They'll be able to make sure that they can deliver a degree of sophistication in the service that is going to be needed for the next wave. That is very different from what, I think, would have been their pre-AI. And by the way, you're seeing it right now in every region and every world, we have partnerships that are being announced right now with IndoSaf from Indonesia, and Geo, and Airtel, and all of the telcos around the world, T-Mobile. And so, I think what you're starting to see is this momentum is only going to compound over time.

>> I mean, connectivity has always been the lifeblood of the telcos. Now, you've got compute, network storage and AI unification. That changes the economics, that changes the operating model. What's your opinion on that? You guys are in the middle of it. This seems to be this hyper-converged trend that we're seeing.

>> Yes. And by the way, so if I were to think back and say, "What is fundamentally changing right now?" The global competitiveness of a company or a country will depend on their ability to securely and economically and efficiently generate tokens. So, everyone's going to want to become a token-generation factory at some point. And as you do that, you're going to need to have infrastructure and network connectivity is going to be central to that because a GPU, without being connected, can't really do much for you. You have to make sure that it gets connected. So, network is at the center, security's at the center, and then the observability of this entire stack from the GPU to the model, to the application, to the agent is going to be pretty important that we actually go out and look at in the full stack. And so, I feel like this is going to have a very different level of relevance. And I personally feel like this is a golden opportunity for telcos to take advantage of their size and scale and be able to just get to the next level.

>> So, you see a refresh coming, big time?

>> I see a massive refresh coming. I see 6G being pretty important, and I see them being very AI forward in the way that they think.

>> So, broadening beyond telco, because at your AI summit, which was phenomenal, by the way, you did a great job. I mean, you went end-to-end with very few breaks. It felt like CUBE Day. And so, but you went through the whole stack-

>> I'm taking inspiration from you guys.

>> Thank you.

>> You have good stamina.

>> You had silicon, the Andreessen interview was really fascinating because you guys talked about productivity. It all relates to tokenomics and the cost of producing intelligence. But before I go there, I wanted to ask you about Cisco. Three years ago, I was asked by a friend inside of Cisco who shall remain unnamed to evaluate Cisco's AI strategy. And I looked at it and I said, "Wow." Honest opinion, it's disjointed, it's fragmented, it's incoherent, and I don't get it. What changed?

>> I think we were working on a lot of these things, but the big shift that happened was we made sure that the organization got structured in a way where we could move fast, that was number one. And number two is AI itself exploded where the network became not a thing that was the ancillary thing. It became part of the core. So, if you think about like, what are the constraints of AI right now? There are three big constraints. You have an infrastructure constraint, you don't have enough power, compute, network bandwidth in the world to satiate the needs of AI. We play a central role in that by tying the GPUs together for both scale out and scale across. Second big constraint is a trust deficit of some sort. If people don't trust these systems, they're not going to use it. The reason that not everyone's using OpenClaw today is because you don't know if it's going to actually take over your computer and you want to make sure that you have guardrails on OpenClaw. And so, you should expect that companies like Cisco will provide safety and security guardrails for agents. And that's going to require two things. You have to protect the agent from the world, prompt injection attacks, data poisoning, all of those things that you could actually do harm to the agent. And second, you have to protect the world from agents. Agents can't just go awry and so you have to make sure that you put guardrails around them. And so, we will focus a lot on that and we are already focusing a lot on that. And that's an area that we're seeing a tremendous amount of compounding. And the third area is this notion of context enrichment. If you think about ourselves, the trillions of tokens of context that we gather on a regular basis through sight, through touch, through smell, through hearing, through speech, language, that's what allows us to be so efficient in society, right? Agents are going to need the same thing. And if you don't give it context, they'll still make decisions, they'll just be really bad decisions. So, we have to enrich the agents with context. Those three things of infrastructure, trust and context, Cisco is at the center of it all. And that's what's changed is our strategy is very clear. We're building things from the silicon that powers the network. ASIC, that powers the switch to the operating system that we have, to the platform and security. And so, what you have is the broadest portfolio and the deepest portfolio in areas that is completely complimentary to everything else that everyone else is doing. So, I think we are very fortunate.

>> It's interesting when you map that out, I think of a distributed edge as you have all these nodes out there, wireless, wireline, you got these AI factories. How do the agents enforce policy at scale and what does that surface area look like from a security standpoint? I mean, does hyper-converge reduce the surface area for attackers or does it...

>> Actually, everything in AI just continues to keep increasing the surface area. Now, by the way, it's a really important point you bring up because token generation should not just happen at data centers. It needs to happen at the edge and with robotics and with physical AI, you might not have the luxury to have a roundtrip made to the data center. And so, we launched a unified edge for branches. The reason we did that is because you wanted to make sure that the branch had the infrastructure, but didn't have the people to manage that infrastructure remotely. So, it had to be centrally managed, but made sure that it was available in a distributed way. And it wasn't just the network, it was network and compute and security all in a single box that was combined together. I think that's going to be the way that the future is going to be, is you're going to have inference at the edge, you're going to have inference in the data center, and you're going to have inference everywhere in the fabric.

>> So, latency is incredibly important in that scenario, right?

>> Latency, power efficiency and performance.

>> Okay. And so, as John's saying, your hyper-converged scenario at the edge, I mean it's essentially got to be real time. So, you have to be physically proximate to the LLM, I guess. And then, the network has to be of sufficient enough performance to be able to give me an answer and make decisions literally in real time. Not only good decisions, but fast decisions, or else you're going to miss the opportunity, are you not?

>> Ultra-low latency, at nanosecond. When you think about AI infrastructure, you have to rethink the way that the infrastructure actually operated. It's ultra-low latency, ultra-high performance with energy efficiency because we just don't have enough power in the world to satiate the needs of AI right now. And so, you need to have all those three and the cycle time for getting the next version of these keeps compressing. And so, not only do you have to have all those three, the next generation has to be infinitely better than the previous one, and then the one after that has to be infinitely better than the previous one and that's just the way that the world's operating right now.

>> Yeah. Jeetu, one of the things we've been monitoring, I want to get your reaction to this, my headline from the SiliconANGLE story, I'm going to post is the hyperconverged AI edge is telecom's next big bet or its last missed opportunity. And really, that speaks to the business model of data-centric architectures. If you look at CoreWeave they just had their first year as a public company, they're doing great. The neoclouds, the clouds, all the AI factories that are being built and the construction CapEx is off the charts and we can see the news there. So, it's a natural progression to think, "Oh, edge." And if you look at all the carriers, central offices, these are cabinets, they're infrastructure, they have power, they got space. Jensen speaks about energy being the low level of the Maslow's hierarchy of needs in tech. You got the energy, power, and space. This speaks to potentially a massive upgrade to create intelligence at the edge. What's your reaction to that?

>> I think the intelligence at the edge is absolutely going to be critical. I also think when you start thinking about... Jensen talks about this where he talks about the fact that the infrastructure that's going to be needed, you're going to have to re-rack every single data center in the world to be AI ready. And as you re-rack those data centers, you're going to have to repower those data centers because it's going to require very different levels of power that needs to be pulled in, which means you have to rethink the cooling systems, which means you have to rethink the space requirements and the shell that needs to get built. I think everything changes. And then, you need to rethink how the network flows and the packets flow. So, what used to happen was you had a single GPU where the entire model would fit in for training. Then you said, "Well, this isn't going to be enough." So, they had eight GPUs within a server, that tended not to be enough. So, then you created a rack of GPUs of a bunch of servers, that wasn't enough. And so, now what you have is a cluster of GPUs within a data center that need to get connected and network. What we are at now though is data centers that are hundreds of kilometers apart that need to act as one ultra cluster that can operate coherently because you might only be able to fit 100,000 GPUs within a single data center and I might not be able to pull all that power in one location. So, what I now need to do is pull power in multiple locations that currently are hundreds of kilometers apart. Over time, it's going to get to continental scale.

>> So, three years ago, we thought that was going to cost a trillion dollars. We're going to blow through a trillion, we spend $700 billion on that capacity-

>> Yeah, right now the estimate's you probably will spend close to $5 trillion on data center build-outs.

>> Totally. And so, okay. So, where do you land on this? Obviously, people are freaking out about the CapEx, looking for the return, the dissonance between this CapEx spending and the revenue growth. It's going to take a while for that to come breakeven. I mean, you're obviously a beneficiary of this CapEx build out, but from a productivity standpoint, this gets into some of the conversation you had with Andreessen, what are your expectations and what would you tell people that are saying, "Wow, this is really dangerous. It's not going to happen," or, "We're concerned it's not going to happen"? What's your take on all that?

>> So, my take on this is in the long-term, if you think about this as a 7 to 10-year window, we are grossly underestimating the capacity required to fulfill the needs of AI. Underestimating, not overestimating. The other thing that people always ask is, "Hey, is this going to be like the dot-com bust days, where you built out all this capacity and then all of a sudden there was no demand for it?" The big difference between that and right now-

>> Demand....

>> is we are still trailing the demand with supply. We are supply-constrained. Every single bit of token-generation capacity that's being sold is getting consumed right away. This is not something that you have to grow into the demand side. The demand is already there.

>> And the web had no demand. They were building it out-

>> They had no demand....

>> Prior to demand.

>> This has the demand. Now, by the way, will there be some short-term ebbs and flows that might happen? Yes, but in the long-term, I don't see how this thing is actually going to be associated with the current needs, especially when you start thinking about agents that are going to be working 7-by-24 that are going to conduct majority of the tasks, because everything can be taught computationally to these agents. And so, it's going to require computational power.

>> So, what about the SaaS apocalypse? I got to get your weigh in on that because everybody's freaking out about that. And basically, I don't care if you're a security company, or a SaaS company, or Microsoft, it's hitting everything across the board. Obviously, not everyone's going to make it through the knothole, but many are. How do you see that playing out?

>> So, I have a very specific point of view on this, and I think there's a bunch of arguments that get conflated together. So, let's just break it out. The need for independent software vendors to go out and solve a problem once that can then be applied to millions of companies consistently is not going to go away because AI came about. Why is that not going to go away? Because I don't think a hospital says, "My mission is going to change into building an ERP system." A hospital says, "I want to take care of patients." An insurance company says, "I want to insure my clients." So, they're going to want to make sure that they focus on their core and what's context is something that they're going to look for someone else to go out and focus on. So, the independent software vendor model I don't think goes away anytime soon, and it's absurd to think that it's going to go away. Now, the current software vendors that are out there, every company in the world faces this tremendous multi-trillion dollar opportunity ahead of them and an existential crisis. So, what is the existential crisis? If you don't reconfigure your business to be AI-first, you will not be relevant. So, if any company that's a SaaS company says, "I'm going to keep milking my cash cow and I'm not going to go out and forward invest in AI," then they won't be a relevant company for the next wave. If they do, then they've actually got a tremendous moat in their system of record, and if they can add an agent layer on top of that and make sure that they can actually tie it together, but what's one of the big things that's missing right now is we have trained these models with human-generated data publicly available on the internet. The next wave of enriching context is going to come from proprietary data, that is only their data for their own use cases and machine data, right? Independent software vendors are in the best position to be able to deliver that to customers. They just have to make sure that they're AI forward. So, I agree with anyone that says, "If you're not an AI-forward company, you won't be relevant." But the fact that independent software vendors will not be something that'll be needed, I don't even understand their-

>> There are a lot of them that are AI forward that are getting . It's a software.

>> It's a software.

>> Software value. Let's apply that to MWC because if you look at what you guys did on stage with Jensen following NVIDIA, they had Blackwell, they had the GPUs. The networking was the key component to make these AI factories, and now you see the results. The interconnect, InfiniBand, which is people , but the interconnects. Networking, Jensen said on stage is the OS for AI factories. Okay. Apply that networking concept to MWC, telecom infrastructure, carriers, enterprises. 30% of attendees and exhibitors are enterprise here. Okay, that's the convergence of enterprise and telecom. Networking is the lifeblood. What's your vision on that product-wise? Because you have to have coherency, you mentioned the data centers. The networking is going to be the connective tissue.

>> In the absence of having the GPUs coherently networked, you will not be able to go out and do with AI what you need to have done. Now, the big area of upside for the telcos and for networking in general, it's not just about the interconnect. The interconnect actually denotes a very loose connection where two people can have lightweight data that can go back and forth. What we're talking about is ultra clusters that get built out in a scale-across mode. And you literally have to start from the chip architecture. The chips and network ASICs we build for scale across have completely different technology like deep buffering that allows you to make sure that you can have very large volumes of data go from one data center to the other, and it looks virtually like a single data center.

>> So, to inference, because it's the killer app.

>> Okay. But explain this because you hear Jensen on this conference call talk about, "We're the biggest networking company out there." I'm like, "Oh, I thought Cisco was the biggest networking company out there." But interesting dynamic-

>> And we're partners with them-

>> Of course you are and that's my question is explain to the audience how you partner, how you compliment what NVIDIA is doing in their networking business and where Cisco is needed and necessary and relevant?

>> What NVIDIA does a lot is they do the scale up networking, which is within a rack, how do you make sure that you network the GPUs of racks within a single rack? When you have multiple clusters of racks, that's called scale out networking.

>> But that's what they do with their ethernet, right?

>> But so are we, right? And then, when you go scale across, we do that as well. By the way, Jensen is one of our biggest fans because we actually go out and serve the enterprise world and he actually wants to make sure he serves the enterprise. And so, there's a ton of synergy. The one thing I'll say to you guys about this is what's different in this era right now is you have to play with the ecosystem. And this notion of if two companies compete on a small, different area, then they're actually not going to be able to partner together, that's a very forgotten notion at this point. Every single major partner that we have, we have a competitive element with them, but that doesn't mean that prevents us from partnering with them. And in fact, a lot of our competitors, we want to partner with them as well because we can't be arrogant enough thinking that the entire world's requirements for everything that they need is going to be served by one company. You're going to be specialized in different areas and those things are going to come together and we just have to make sure that we tie those things together.

>> So, what does that mean in practice? Obviously, if you're buying a GPU cluster and you need NVLink 72 to scale up, that's going to be clearly NVIDIA's role. But if you've got a Cisco customer and a fleet of customers that want to do business with Cisco, they're going to continue to do business-

>> Okay. If you are a hyperscaler today, or if you happen to be a neocloud, or you happen to be a sovereign cloud, or you happen to be a service provider, or you happen to be an enterprise, Cisco serves all of those customers for scale out and scale across needs, right? We have our ASICs across five different families of products that are special-purpose built, they're programmable, they're energy efficient, they're sustainable, they actually work with the entire ecosystem of partners and players. We just happen to be-

>> We were at the launch of your new ASIC strategy, what was that, four years ago, maybe, in San Francisco? I mean, you guys made the decision to develop your own custom silicon.

>> We did.

>> And that wasn't a light decision. A lot of people criticized it at the time, thought it was a mistake. How have you turned that into a competitive advantage? What does it mean for the industry?

>> It was one of the most strategic things that we did in the history of Cisco. I take no credit for it because that was a Chuck Robbins decision. But if you really think about what happened over there is, right now, there's only like a handful of companies, us, Broadcom, NVIDIA, that actually build the network casings. And so, what hyperscalers want at volume is they want choice. What enterprises want is they want choice. And so, we happen to be one of those that actually not only builds the ASIC, we build the system that goes on top of the ASIC, the hardware itself, the physical boxes, and we build the OS. And then, we have platforms for security, and then we fuse security into the fabric of the network. So, when you start thinking about the full stack from silicon to system, to OS, to management plane, to security, to observability from one company, we're the only company that does that.

>> And I can buy that as a system or I can buy the piece parts if I want. Is that correct?

>> That's right.

>> You guys have the routes, you know the routers, got all this gear-

>> And by the way, we serve hyperscalers, neoclouds, sovereign clouds, service providers and the enterprise.

>> You can't take credit for that silicon decision, but you can take credit for my earlier question of maybe helping bringing that AI strategy together and make it more coherent.

>> This is a team sport, Dave.

>> I know it is, but that was the other thing that changed that you humbly didn't introduce into the conversation.

>> One thing NVIDIA can't take credit for is that you guys have that horizontal routing that they don't have.

>> That's right.

>> And so, I want to ask you, Cisco's known for policy-based networking. I put that in quotes because that's old-school packet moving around the internet. What does an AI factory at the edge mean for policy? Because now, you have, essentially, intelligence looking at everything happening in the interface coming in, access. And then, looking back in the network, you have full visibility on both sides. What's that look like?

>> I mean, here's the way that you should think about what the future is going to look like is when you have a policy change, the way that the policy change gets implemented in your infrastructure is think about a bunch of ambient agents that are going to be sitting there monitoring what's happening in your infrastructure. They find something that is an anomaly. As a result of that anomaly, you are now going to go out and set a policy that's going to go help mitigate the risk from that anomaly. You don't want to go out and do that in your production environment right away. So, what you will do is you'll create a digital twin environment and you will make sure that you apply that policy with data that was actually live data that you had in your system through that digital twin. See how that digital twin operated. And if there was something that went wrong, do a diff. And once you've done a diff and things look good, you can actually roll that out into your production system. What that basically says is it's a fully self-healing, autonomous network without downtime, and that's what the future is going to look like.

>> And how would you describe what an AI workload is versus network packets? What's the difference between the two?

>> I mean, look, the thing is is in these large training clusters that you do, dropping a packet requires that you actually restart your training run, so it's very expensive. It's not like you just dropped a packet in a video conference call. It's a very, very different kind of consequence. So, what you have to have then is technologies, like deep buffering, that allow you to make sure that the unevenness of packet loss that might happen in a typical network can be compensated for, which is what we've essentially done with our P200 chip that we have now with the 8223 router. And then, we've announced that that chip is also going to be available on our Nexus family of routers. So, fundamentally, what's happening is the architecture for networking and for security completely changes. Security's architecture becomes hyper-distributed. And networking architecture is self-healing and scaled across, hyperscale, energy efficient, low latency.

>> Yeah. Well, Jeetu, Great to have you on. I know you're super busy here at MWC. As always, great refresh cycle we think is going to be hitting the network at the edge this year and beyond. It's going to be a great build-out. Continuing having factories at the edge.

>> This is an exciting time in the industry. Thank you for having me.

>> Fantastic having you. Thank you.

>> I'm John Furrier with Dave Vellante here at MWC 2026, kicking off four days of wall-to-wall coverage. Stay with us. We'll be right back after this short break.