Semiconductor Insiders - Podcast EP358: How Custom Silicon Development Allowed Cisco to Fuel New Innovation for Its Customers

Episode Date: July 31, 2026

Daniel is joined by Nick Kucharewski Senior Vice President and General Manager for Cisco’s silicon development organization. In this role he is responsible for the end-to-end strategy, cross-functional execution, organization alignment, and product roadmap for Silicon One, Cisco’s uniquely scalable and programmable … Read More

Hello, my name is Daniel Nenny, founder of Semaywiki, the Open Forum for Semiconductor professionals. Welcome to the Semiconductor Insiders podcast series. My guest today is Nick Kuchreski, Senior Vice President and General Manager for Cisco's Silicon Development Organization. In this role, he's responsible for the end-to-end strategy, cross-functional execution, organizational alignment, and product roadmap for Silicon 1, Cisco's uniquely scalable and programmable networking product family.

Nick has over 25 years of experience in networking spanning the evolution of the communication Silicon market over the past three decades. He has worked as a chip design engineer, engineering manager, product manager, and general manager. His work has contributed to silicon and software production innovations across AI and cloud infrastructure, wireless networking, optics, enterprise switching, carrier routing, and broadband access. Welcome to the podcast, Nick. Thanks very much, Daniel. It's great to be here. I'd like to first start out to just ask you quickly, what brought you to Cisco? Do you have an interesting story you can share?

Oh, absolutely. So as you mentioned, I've been involved in networking semiconductors for over 25 years, actually, beginning in the late 1990s. And I had the opportunity to work in semiconductors providing new solutions for carrier networks, enterprise networks, and data center in AI and cloud. And through that entire time, Cisco was always a key customer or a key partner or was driving a number of new innovations happening in those market segments. And the opportunity to work at Cisco and to get involved directly in Silicon Development for such an industry leader was a great opportunity. And I'm glad to be here.

Great. So what strategic imperatives drove Cisco's transition to custom silicon development? I remember, you know, I'm from Silicon Valley, so I remember Cisco's beginning and used to work with ASIC companies so people would design the chips for you. Yeah, absolutely. That's right. So for most of Cisco's 40-year history, the company has built silicon, but primarily that had been done with ASICs, meaning any given product product would work with a third-party company to develop some custom silicon specifically for that product line. And what changed was several years ago, Cisco embarked on a strategic initiative to develop. in-house silicon, a unified product family that can serve enterprise, service provider, and cloud applications with in-house developed silicon using customer-owned toolset. That would give us the ability to control our own roadmap, to drive to our own market timelines, and to get a high amount of software reuse across different market segments. And that has largely

been a great success. We have five product lines in production now, spanning the enterprise provider and cloud markets. So how does Silicon One's architecture address the evolving needs of today's infrastructure, whether it's on-premise or in a data center? Yeah. So what's interesting is when we look at the Cisco Silicon Strategy, the fact that we're able to do in-house silicon development has proved to be really critical in the new transition to AI applications, both in enterprise and campus networks and also in cloud and hyper-scale.

And the reason for that is actually different for those two markets. When we look at the cloud and AI market within hyperscale networks, we see a need for very, very high performance silicon that is used to connect large numbers of GPUs or XPUs operating at scale. And here you have to balance high performance silicon that's very efficient, power efficient, very high performance, with a high degree of flexibility and programmability. And Cisco's implementation is programmable, but it enables that high performance with software programmability. And that allows you to actually change the functionality of the silicon infield after it's deployed. And we're finding that that's really critical as we look at new AI buildouts because there's a wide variety of different applications when you talk about training versus inference. In building up networks comprised of GPUs or XPU solutions from different providers, there's a real need for the the network to be very high performance and very flexible. And the Cisco approach is proving to be

very important for that. And it's really, it's bringing us a lot of new opportunities. Now within the campus network, we're also seeing a great application for our unique architectural approach, which is programmable and high performance. And there, we're seeing AI within the enterprise means that you're actually running agentic AI within the campus network. You're generating a lot of network traffic because you have eight. that are running and they generate more traffic than a human would and they're running 24-7 and this means a change in the requirements for those campus networks the silicon has to be running at high performance it has to be very adaptive so that you can respond to the performance needs

and also the new security challenges that are posed by AI running within the campus network interesting so what unique value do customers derive from the full stack programmable and what open approach yeah it's it's a great question. So one of the advantages that Cisco has in designing its own in-house silicon is that we can look at a full-stack solution, starting first at the silicon, and then the optics, the system hardware that brings that together, the network operating system software, and then the higher layer orchestration software for network management. And that's a very unique approach. There are very few others who are looking at it in this very broad way. And in building that full stack, that fully vertical, integrated solution. We can look at the requirements for next generation network management in anticipating that drive silicon engines that can work very closely with that software in order to provide new security, new visibility, telemetry, or adaptability for the network. And that translates into real performance gains for the customer in terms of being able to map these AI

workloads across multiple devices. It allows our customers a simple view of the network, which is essential when you talk about security policy or visibility in terms of what's happening in the network and making adjustments. So with that, we're really able to drive next generation features looking not only at the software, but also how do you change the silicon in order to enable that software to perform better? Interesting. So as AI shifts towards inference and agentic workflows, what critical infrastructure bottlenecks are customers facing? Yeah, so the answer is different. you're looking at cloud AI versus within the enterprise. When you're looking at cloud and

hyperscale applications, the challenge here is connecting massive numbers of compute elements, GPUs or XPUs. The performance of the network is critical in that it allows all of these compute elements to perform as one. The workloads are spread across multiple devices in the latency of the network, the throughput of the network actually has a real impact in terms of ultimately the compute speed that these devices can achieve. And that translates into the profitability of that deployment. So here, Cisco has a number of innovations in the silicon, which enables higher performance. In particular, the packet buffer architecture is optimized for these AI workloads. We have a number of features that provide better load balancing

across the network that can react to congestion events or link losses within the network. And we provide the ability to upgrade that network after it's deployed. And we're finding this, this is increasingly important in cloud and hyperscale because deployments are not homogenous. They are heterogeneous where you may have in a large-scale cloud deployment a mixture

of silicon from different vendors. You may have multiple different. different generations of device. In here, when you're looking at a new deployment, rolling out new equipment, you would like to have your existing equipment actually up-level its behaviors in order to match your new equipment, for instance, in telemetry or load balancing.

And the programmability that Cisco offers really gives you that ability to upgrade the performance of your existing equipment. We call this re-greening. Instead of viewing it as a brownfield or greenfield, you can actually take your existing equipment and actually upgrade some of that functionality through the programmability that we provide.

And we're seeing that this has a lot of traction with our customers as a concept that enables them to extend the lifespan of their equipment and ultimately get higher ROI from their deployments. Now within enterprise and campus, our unique architectural approach is again, this programmability and silicon that is optimized to work in tandem with the network level software.

In here, AI in the enterprise campus is constantly evolving, And the software needs to evolve with it in terms of the telemetry you can provide in terms of the security capabilities for new threats that are posed by agentic AI running within the network. And so here, it's really a vertical stack solution where the software is working in combination with the silicon in order to provide new telemetry or new security responses that are needed by these new applications as they run in the network. And how does collaborative co-design with hyperscalers shape the product roadmap? Yeah, it's a great question. When we look at the cloud and hyperscale application, clearly the end customers are really driving a lot of the new innovations in terms of how they want their network structured. Each hyperscaler has a different vision, a different view for how they would build their topology and the management, the workloads that they're supporting. And then also what is their long term plan for managing that network? So here, it's really about a partnership working with the hypers to understand their requirements.

both in terms of the network performance and also in terms of how they want to manage that network and make sure that the silicon has the right features to enable that, and then working with their software teams to make sure that they can develop full system solutions, full network solutions to take advantage of what we're providing in the silicon. And last question, Nick, how does Cisco strategy with silicon and optics as the foundation ensure the company is best positioned for the future AI challenges? Yeah, it's a great question and it's very compelling because when we look at the next generation of AI, it really is about the performance and the visibility of the network.

Worldwide, if you look inside the cloud or if you look inside global networks, we're looking at a significant increase in the compute capacity that's required for these new AI applications. And the network plays a critical role in that in terms of making those compute devices behave as one with low latency, because they're sharing the workload across a very diverse network environment. And so the network infrastructure, that network fabric is really critical. And there are two really key elements to that. One is the packet processing or the switching that I've talked about at length. And the other one is the connectivity, copper connectivity and also optical connectivity.

And really optics is key for the bandwidths that we're talking about here. And Cisco has a product line for packet processing silicon and also CISCO. systems and software, and also has a very broad product line for optics, both optics inside an enterprise, optics inside a data center, and then the optics running between the data centers across a wide area network. And so having that solution, both for the switching and networking, as well as for the optical connectivity, is really key to providing a broad portfolio, both for our enterprise customers, as well as our cloud customers. great. Hey, it's great to connect with you, Nick. You know, I've been in the semiconductor industry for over 40 years, and I grew up with Cisco. And I remember when you did transition from ASIC to in-house silicon, and, you know, now a lot of system companies do it, but Cisco was one of the first ones. And I did have the pleasure of working on one of your projects and just an incredible project and incredible team. So congratulations on your success. Great. Thanks very much. I appreciate it.

That concludes our podcast. Thank you all for listening and have A great day.