Broadcom Advances Edge AI Infrastructure and Application-Aware Networking at MWC25

Broadcom boosts edge AI infrastructure for faster connectivity - SiliconANGLE

Building AI from the core to the edge: Broadcom's approach to connectivity The evolution of artificial intelligence is fueling a surge in demand for cutting-edge edge AI infrastructure. Companies such as Broadcom Inc. are stepping up to deliver open, scalable and power-efficient solutions to meet this need. Broadcom has made significant strides in AI infrastructure over the past year, according to Vijay Nagarajan (pictured), vice president of strategy and marketing for the Semiconductor Solutions Group at Broadcom. The company's commitment to these principles has driven its product decisions and led to advancements across its infrastructure portfolio. "In the last 12 months, I would say we've delivered to this vision, and we've also delivered to the product promises that we've made across all of our infrastructure products," Nagarajan said. "There's XPUs, there's ethernet, [peripheral component interconnect express], optics and foundational technologies like SerDes and [digital signal processing]. We've had tremendous success." Nagarajan spoke with theCUBE's Dave Vellante and Savannah Peterson at MWC25, during an exclusive broadcast on theCUBE, SiliconANGLE Media's livestreaming studio. They discussed Broadcom's evolving AI infrastructure strategy, from high-speed connectivity to edge intelligence. (* Disclosure below.) SerDes technology plays a crucial role in AI infrastructure, facilitating fast, reliable communication between components, according to Nagarajan. These high-speed connections help maintain strong signal-to-noise ratios, ensuring data integrity across various products. "SerDes needs to make sure that your signal gets across from one end of the link to the other despite all the noise that you would see in the electrical circuits," Nagarajan said. "Every piece of connectivity silicon that we do requires the SerDes, and the quality of the SerDes makes those products that much more impactful and valuable." Broadcom recognizes the central role of SerDes in powering high-performance edge AI infrastructure. It has invested heavily in advancing its capabilities to keep pace with the growing demands of AI and machine learning. But SerDes is just one piece of the equation, according to Nagarajan. "SerDes is important, but it's equally important that other pieces of our technology are ready as well," he said. "That includes advancements on the ethernet front and ensuring XPUs are equipped with the necessary capabilities." Broadcom is extending its AI expertise beyond large-scale infrastructure, expanding its capabilities in edge AI infrastructure through its broadband access offerings, according to Nagarajan. A key development in this space is the company's unified DOCSIS 4.0 chipset, which integrates AI and machine learning to enhance real-time processing in home networks. "In the last 12 months, we had a couple of very interesting announcements that we made in the edge AI space," Nagarajan said. "[DOCSIS 4.0] has a neural engine that is capable of supporting a lot of the AI use cases in your home. Think of your broadband gateway as your edge device, handling video processing for a great video experience." Beyond home connectivity, Broadcom is embedding AI-driven neural engines across its broadband ecosystem, including modems, smart applications and passive optical network devices. These advancements improve network optimization, power management and real-time decision-making, according to Nagarajan. "From an edge perspective, we've made sure that the edge access network actually has neural capabilities ... [and] we've integrated these neural engines not just in the modems, but also in the smart apps and nodes," he said. "Or you take our ... 50-gig PON devices. Today, they have neural engine spec, so the two sides of the link -- the [optical line terminals] and the [optical network units] -- have the neural engine. Think about what it can offer service providers and operators [such as] network management capabilities [and] power management capabilities. The possibilities are endless." The relationship between edge AI infrastructure and centralized AI infrastructure can be thought of as analogous to the spinal cord and brain, according to Nagarajan. Just as the spinal cord processes reflexive actions before signals reach the brain, edge AI enables real-time, localized decision-making before data is sent to cloud or data center compute. "When you go touch a hot object, your neural sensors drive your information back to the spinal cord and to the brain," Nagarajan explained. "The spinal cord first does the processing and lets you know, 'Hey, take your hand off.' The information goes back to the brain ... and the brain then triggers the pain sensation. You have a reflex that's driven by the spinal cord, and then you have the brain doing the ultimate processing." Here's the complete video interview, part of SiliconANGLE's and theCUBE's coverage of MWC25:

How application-aware networking is transforming telecom and AI connectivity - SiliconANGLE

Broadcom drives application-aware networking for the AI era Telecom operators have struggled to monetize their networks beyond basic connectivity. With the rise of AI-driven applications, new optimization models are emerging to improve performance and revenue. Application-aware networking allows providers to allocate resources dynamically, ensuring critical applications such as AI workloads and healthcare systems get priority access. A key innovation in this space is technological convergence. Unlike bundling services into a single bill, this approach integrates multiple networks -- 5G, fiber and satellite -- at the application level. This allows enterprises to optimize connectivity based on specific needs rather than relying on a one-size-fits-all model. Broadcom Inc. is at the forefront of this transformation, introducing solutions such as VeloSky, to help telecom operators seamlessly implement application-aware networking and unlock new revenue opportunities, according to Sanjay Uppal (pictured), senior vice president and general manager, VeloCloud Division, at Broadcom Inc. "We announced our VeloSky product, which is our new product for convergence," Uppal said. "Essentially, what it does is for telecom operators who are supplying technology and services to enterprises, it allows them to go on an application-by-application basis and combine or converge these networks together." Uppal spoke with theCUBE's Dave Vellante and Bob Laliberte at MWC25, during an exclusive broadcast on theCUBE, SiliconANGLE Media's livestreaming studio. They discussed how Broadcom's VeloSky enables telecom operators to leverage application-aware networking and AI-driven optimization to enhance connectivity, improve performance and unlock new monetization opportunities. (* Disclosure below.) Traditional telecom models prioritize network stability, but the next wave of innovation demands dynamic adaptability. Enterprises require networks that not only remain reliable but also intelligently allocate resources where they are needed most. The ability to combine different types of networks -- such as 5G, satellite and fiber -- ensures mission-critical applications remain operational, even in high-demand environments. "You can combine two 5Gs together, you can combine satellite with 5G or with fiber -- in fact, any type of network," Uppal said. "What's also really cool about this is that even a single link you can make it perform better. But of course, the whole idea is about converging different links together on the basis of applications." This concept extends beyond traditional high-availability models by focusing on performance optimization. By leveraging real-time data and AI-driven network adjustments, telecom operators can ensure that applications requiring low latency -- such as AI-driven assistants or augmented reality tools -- are prioritized. This application-aware networking also introduces new revenue opportunities for service providers, as businesses may be willing to pay for higher levels of guaranteed performance, Uppal explained. "I think one of the things that's very interesting ... what we've been doing is measuring the gen AI applications on the network for some of our largest customers -- those that have more than 1,000 endpoints," he said. "What we are finding is that the traffic stream of these applications is very different from normal web applications." The surge in AI applications is forcing telecom operators to rethink how networks are built. Unlike traditional applications, AI-driven workloads often require significantly more upstream traffic than downstream. This shift presents new challenges, as most networks were originally designed with the assumption that downstream traffic would be the dominant factor, according to Uppal. "Now, in certain cases you find that the upstream is 100 times more," he said. "Now you think 100 times more, how can that be possible? But here I have my pair of Meta Ray-Ban glasses. Now these are meant for the consumer, but these glasses, they have video in the upstream direction and audio in the downstream," Uppal noted. As AI-powered devices and agent-based computing become more prevalent, networks will need to adapt to support the evolving traffic patterns. Real-time processing, machine-to-machine communication and data-intensive applications will require flexible network solutions capable of dynamically optimizing connectivity based on usage patterns. This shift is not just about efficiency but about enabling new capabilities that were previously impossible, according to Uppal. "This agent is talking to that one, waiting for the response, and then behind the agent is sitting a small language model or a large language model," he said. "When you do that, then the interaction between these agents is dependent on the network. No network; no agentic AI. And that's really critical." The telecom industry is at a turning point. The convergence of AI, application-aware networking and dynamic connectivity is redefining how enterprises leverage network infrastructure. As operators explore new monetization strategies, the ability to intelligently allocate resources in real time could be the key to unlocking the full potential of next-generation connectivity, Uppal concluded. Here's the complete video interview, part of SiliconANGLE's and theCUBE's coverage of MWC25: