Broadcom’s artificial intelligence (AI) infusion into its VMware edge offerings has hit deluge proportions as VMware’s erstwhile VeloCloud SD-WAN platform has now gained an AI-focused networking stablemate.

The new VeloRAIN platform, which stands for “robust AI networking,” taps AI and machine learning (ML) to help improve the performance and security of distributed AI workloads. These would normally be edge workloads that run across VMware’s VeloCloud SD-WAN platform.

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 Barcelona event that the VeloRAIN product can identify encrypted application traffic, which allows for it to prioritize edge AI applications that can support quality and service-level requirements. It also runs a new Dynamic Application-Based Slicing (DABS) dynamic policy framework system that allow for the prioritization of application traffic.

Uppal noted that VMware has more than 500,000 edge units deployed “and each of these acts as a sensor on the enterprise and service provider networks.” These edge units can measure traffic at the application layer and more broadly on overall network performance to inform on deciphering AI-based network traffic.

That insight shows a dramatic difference between traditional edge network traffic and generative AI (genAI) and AI-fueled traffic coming from systems like Google Cloud’s Gemini.

“In the enterprise context, when you have [retrieval-augmented generation] from all the endpoints of an enterprise, what companies are doing is they are using their own private data to pre-train or fine tune the models that have already been put in place and that requires an enormous amount of upstream traffic,” Uppal said. “But now you add in the fact that there is video and audio also going in the upstream direction, and once again you get this high asymmetry in the traffic for genAI workloads as compared to workloads that we've been used to.”

VeloRAIN also includes channel estimation intelligence to monitor and optimize connections over wireless links like cellular and satellite communications. This ties into the DABS system to feed the VeloRAIN platform’s ability to manage service quality.

“What we're doing here is we've taken a mechanism to use machine learning to do channel estimation,” Uppal explained. “What that means simply is because 4G and 5G, as well as satellite technologies, vary a lot, it's very important to know how that bandwidth, jitter, packet loss, and latency are varying. And if you have a model that is running and you are able to predict how the channel is working and you’ve identified the application, you’ve got the first two building blocks.”

Uppal explained that this channel estimation is becoming increasingly important in support of 5G-based fixed-wireless access (FWA) services targeted at enterprise customers. This can also allow operators to tap service-quality features to better monetize those network investments.

“The service provider telco community is really looking for that next breakthrough – it wasn't 5G – but what is the next thing that is going to use their networks to the best extent so that they can monetize and charge for it,” Uppal said of VeloRAIN. “The whole idea of VeloRAIN is to be able to get there, to use AI and ML to improve the performance of the underlying networks, but then to be able to improve the quality of experience for the users that are making use of those networks.”

The VeloRAIN launch builds on Broadcom’s VMware edge AI updates announced earlier this year. Those updates included new edge connectivity appliances, an integrated secure access service edge (SASE) product, and updates to its Edge Compute Stack (ECS) platform.

More VeloCloud devices Broadcom also launched a pair of high-end VeloCloud SD-WAN edge devices that are targeted at high-scale and AI needs. This includes the VeloCloud Edge 4100 that can support up to 30 Gb/s of throughput and connect up to 12,000 tunnels, and 5100 devices that can support up to 100 Gb/s speeds and connect up to 20,000 tunnels.

Uppal noted these are targeted at edge applications.

“What we are talking about is what happens outside of the data center,” Uppal said. “When data centers talk to one another, or when data centers talk to other locations, you can imagine a [large language model] talking to a [small language model], all fronted by an agentic AI model, you need higher levels of throughput that can be accommodated for by bursts. We're calling these as AI -ready appliances that are hitting the top boundaries in terms of the ability to have very low levels of latency to serve traffic that is used by agentic AI.”

Gartner recently ranked VeloCloud as one of the SD-WAN market’s leading platforms alongside rivals Fortinet and Cisco. VeloCloud was lauded for its deep product capabilities, strong market share, and strong market understanding.

However, Gartner did caution on Broadcom’s customer experience, product roadmap that was more catchup than market leading, and concerns over Broadcom’s future SD-WAN plans tied to the VMware integration.