Broadcom reoriented its VMware software-defined edge posture to take advantage of what company management claims is a surge in artificial intelligence (AI) use in those edge environments that is altering the demand and need from enterprise customers.

Sanjay Uppal, VP and GM of Broadcom’s Software-Defined Edge division, explained in a press pre-briefing ahead of this week’s VMware Explore event that the focus on edge AI use cases has changed dramatically, which in turn required the vendor to break its legacy vertical market mindset. Uppal added that Broadcom’s view of edge AI is both the access to AI and the AI applications that are running out to edge deployments.

“We felt that software-defined edge was very vertically oriented, so you had a solution for each one of those verticals,” Uppal said. “But with the advent of edge AI, we're finding a horizontal play in the infrastructure, and that is this play that we are getting to market.”

The reorientation includes new edge connectivity appliances, an integrated secure access service edge (SASE) product, and updates to its Edge Compute Stack (ECS) platform.

From the appliance standpoint, VMware’s VeloCloud SD-WAN Edge product line gained three new models, adding connectivity options that can better support changing throughput demands in edge locations. This includes greater support for 5G-based fixed-wireless access (FWA) deployments that are increasingly being adopted by enterprises and new satellite connectivity options.

Uppal explained that those connection options help support a change in data transmission orientation.

“What's happening out there with fixed-wireless access is that it's outgrowing what's happening with broadband,” Uppal said, acknowledging recent trends of FWA connection growth outpacing that of new wired broadband connections. “And when you converge it with fiber and you converge it with satellite you're actually able to get a network that is much more symmetric than you could get with any one of these technologies individually.”

This need for more symmetric data transmission support is being driven growing demands from AI use at the edge, something that Uppal said Broadcom is able to ascertain due to its deep VeloCloud appliance footprint.

“What we're finding is just simply chat bots themselves, the nature of traffic just with chat bots has changed,” Uppal said of just one use case. “It is no longer the asymmetric traffic that we were expecting. There's a lot more traffic that is coming from the end points.”

Uppal noted that this increased traffic is tied to large language models (LLMs) that are needed for training and inference models that need to be run to and from these end points. This traffic increase and the actual data being transmitted require a more diverse connectivity palate that can be managed with SD-WAN in real time to deal with fluctuating demands and availability.

“Now you can see that both the underlay, the diversity of what's happening with wireless, the programmability of it, and then this overlay that we've constructed all work in the context of this new class of workflow, which is this edge AI workflow, and that's what we're really excited about, because bringing together this edge AI with the network capability of responding to these new workloads is what we are doing from a software-defined edge standpoint,” Uppal said.

VMware SASE and ECS Broadcom is also integrating its single-vendor SASE platform that uses that VeloCloud SD-WAN and its Symantec security service edge (SSE) apparatus to help secure these edge AI use cases. Uppal noted this is needed to help those enterprises to monitor and manage the resiliency and security of those edge AI applications.

“We've introduced not just the ability to respond to these new workloads, but also extending our SD-WAN into software locations and then combining it with the [points of presence] that are common between Symantec and VeloCloud,” Uppal said.

VMware is also integrating recent updates to its ECS to help manage the shifting traffic patterns and additional insight provided by the VeloCloud SD-WAN to support small language models (SLMs) that are typically housed or accessed in edge locations.

“The amount of change that is happening with SLMs is actually very rapid,” Uppal said. “That is much smaller models being deployed at the edge that are trained and are oriented toward the decisions that need to be made right there on the job. And what Edge Compute Stack does is that it has this pull-based architecture that pulls it all together.”

Uppal added that by combining ECS with VeloCloud, “you can have the network as well as the stack that supports these new applications all together in one package.”