Broadcom has announced new AI data centre and edge network platforms designed to unify cloud AI with telecom edge computing infrastructure.

The company, working with partners including Samsung and various telecom operators, is rolling out a portfolio of silicon and software aimed at creating a more cohesive fabric from the hyperscale data centre to the network edge.

Broadcom’s initiative targets the growing disconnect between centralised AI model training and the need for low-latency inference in enterprise and consumer environments. This is a direct attempt to provide the underlying hardware that can finally monetise network operators’ extensive fibre and 5G investments through distributed AI services.

Unifying a fractured infrastructure

Running AI workloads directly on networks, closer to end-users, presents a path to new revenue streams for operators. The difficulty has been the operational complexity of deploying and managing compute infrastructure across thousands of geographically dispersed edge locations. The toolchains, hardware, and management planes for data centre AI and edge AI are often entirely separate.

Broadcom’s strategy is to address this hardware and network fragmentation directly. The new platforms are not a single product but a collection of technologies intended to work in concert. The operational goal is to allow a single, logical AI workload to be partitioned, with different components running where they are most efficient (e.g. large-scale training in the cloud, and real-time inference or data pre-processing at a cell tower, in a factory, or within an enterprise campus network.)

The new platforms are built on several key technology upgrades:

• 50G PON: An upgrade to the Passive Optical Network standard, increasing backhaul capacity from the current 10G standard. This is the foundational layer required to handle the data traffic from thousands of AI-enabled edge devices.
• Wi-Fi 8: The next generation of wireless networking, essential for providing the necessary throughput and low latency for on-premise AI applications inside enterprise and home environments.
• Fixed Wireless Access (FWA): Enhanced FWA solutions to deliver fibre-like speeds over 5G, expanding the reach of high-performance edge computing to locations where laying fibre is not economical.
• Integrated AI accelerators: Purpose-built silicon designed for efficient AI inference, integrated directly into networking equipment to reduce system cost, power consumption, and physical footprint at the edge.

These components are designed to create a performance-matched chain. There is little point in deploying powerful AI accelerators at the edge if the local network (Wi-Fi 8) and the backhaul network (50G PON) cannot handle the data throughput.

Vijay Nagarajan, VP of marketing in Broadcom’s Wireless and Broadband Communications Division, explained: “The true potential of the intelligent broadband edge requires a fundamentally new foundation for the smart home and the smart enterprise.

“By deploying NPUs across our Wi-Fi 8 and broadband solutions, we empower service providers to secure user privacy, reduce network congestion, and deliver the multi-gigabit, sub-millisecond connectivity that enables the AI era.”

From disjointed components to a cohesive AI fabric

By creating a seamless networking and compute continuum, enterprises can deploy AI-powered applications – such as factory-floor quality control or real-time retail analytics – without the performance penalty of a round trip to a distant public cloud. For telcos, it provides the architecture to offer managed services like private 5G networks with on-site AI processing or enhanced content delivery networks that can perform video transcoding at the edge.

Samsung’s involvement points to the integration of these new capabilities into both network infrastructure equipment and end-user devices. The partnership suggests a top-to-bottom optimisation, from the network core out to the handset or customer premises equipment (CPE). 

The FWA and 50G PON platforms provide the pipe, while the new Wi-Fi and accelerator silicon deliver the processing capability at the final destination. This architecture is necessary for running complex AI agents that require constant interaction with local sensors and data sources while also drawing on larger models housed in regional data centres.

Broadcom will be taking on other silicon providers, like NVIDIA and Marvell, who are pursuing similar strategies of converging networking and compute. NVIDIA has focused on its ‘AI factory’ concept, extending its data centre dominance outwards with platforms like the IGX Orin for the industrial edge.

Marvell, meanwhile, has built a strong portfolio in custom silicon for 5G infrastructure and data processing units (DPUs). Broadcom’s distinct advantage lies in its incumbency across the entire network data path, from the data centre switch to the home router’s chipset. This provides a potentially very powerful lever to drive adoption of its full platform.

See also: Global operations depend on securing subsea cable networks

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the IoT Tech Expo and Cyber Security & Cloud Expo. Click here for more information.

Telecoms is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.