
By Claus Hetting, WiFi NOW CEO & Chairman
GenAI-based services are pre-destined to sooner or later land in our homes but what exactly will such services look like and what might be the architectural requirements for CPEs? Charles Cheevers – a frequent WiFi NOW congress speaker and Vantiva CTO – offers his view of what the future connected home might look like. Here’s a summary of the highly recommended new paper available for download here.
A new predictions paper by Vantiva’s CTO Charles Cheevers says the market for home broadband service provider solutions in 2024 moved from performance-led to value-based purchasing. In other words: Most CSPs don’t see a lot of value in speeds beyond 400/60 Mbps but do see value in improved QoE, Vantiva says. This is not news as such but underscores the continued need for QoE solutions including the emergence of new specialised vendors such as Aprecomm and Righ, or even Domos.

Although high demand for Wi-Fi 6 boxes will continue in low-ARPU countries, Vantiva says that we’ll see a lot more Wi-Fi 7 deployments in 2025. “For top tier services – meaning services above 1 Gbps – we will see the introduction of single box 10G-XGSPON Wi-Fi 7 tri-band gateways, DOCSIS 4.0 tri-band Wi-Fi 7 gateways. Additionally, Wi-Fi 7 will emerge for two-box NTU and access point plus mesh solutions in both tri-band and dual-band configurations,” Charles Cheevers says. So far five European CSPs have launch Wi-Fi 7-based services – for more read our coverage here.
And now on to the inescapable subject: GenAI for the home. Charles Cheevers says there is a heavy analysis phase underway to “to map the use of AI based solutions into the entire ecosystem of a service provider” in order to “leverage GenAI solutions for customer support and consumer experience.” Part of Vantiva’s vision is to offer GenAI ‘assistants’ trained on consumer and CSP data sets to enhance the user experience and more. It is expected that a lot of this will happen at the edge (in the gateway) instead of in the Cloud to save power and Cloud costs, Vantiva says. The work to make this happen is already well underway.
To make such use cases happen Vantiva says that CPE architectures will need to support what they call ‘hierarchical AI solutions’. As clients, set top boxes, gateways, and potential Home AI Servers are enabled with AI inference abilities, hierarchical AI solutions will evolve in such a way that every device will contribute its AI element based on its available resources, Vantiva says.
“For example, today’s CoPilot PC’s and smartphones they are enabled for future AI services locally. Similarly, set top boxes and gateways shipping now incorporate integrated NPUs to allow them to perform functions like video object recognition, camera-fed facial and pose detection, and they can even run small language models,” Charles Cheevers says.
So what will such services look like? Some of them will be related to the physical layer, such as discovery and prediction of connectivity disruptions and errors, even security and service fingerprinting, Vantiva says. Others are more futuristic: “So if you imagine an avatar-based assistant with camera as eyes, microphones as ears, and TV screen (powered by an AI set top box) it can engage with home occupants knowing who they are, what they are doing, their facial emotions, body pose to be able to have as close to a human engagement as possible,” Charles Cheevers says.
As CPE device roadmaps evolve to do more, they will need to incorporate more DRAM as well as higher specification NPUs. Vantiva also points out that chipset companies such as Broadcom, Amlogic, and Synaptics offer CPE chipsets with integrated NPU and the ability to do AI inference today while new NPU enabled solutions from Qualcomm and Mediatek are in the pipeline. See our coverage of new GenAI chipset architectures from Qualcomm here and from MediaTek here.
/Claus.