At MWC26, one theme triumphed above all others as the focus of the show — AI. The vast majority of tangible announcements focused on “AI for networks” — that is, the use of AI in planning, optimizing and running networks. However, while real announcements and progress was thin on the ground, there was also a level of excitement about consumer AI use cases — and how mobile operators should prepare.
Perhaps I’m jaded. Some reports have already predicted sizable traffic booms from AI. However, the current popular use cases of passive AI (e.g. AI embedded in search) or active AI (e.g. chatbots) didn’t seem to me like they would tax the network. I actually tested it — throttling my network down to some truly atrocious conditions: 160 Kbps throughput, 500ms lag, and a 50% packet drop chance. At this point, I couldn’t even load a new web page, but I could get a response from ChatGPT and Gemini. Some responses took longer than others, but the LLMs remained usable.
So why are operators talking about uplink revolutions and the need for ultra-low latency?
Here’s my take on some of the most hyped topics in consumer AI — and what they might mean for networks.
Smartglasses are the most tangible example of new AI devices
Given the speed of LLM adoption by consumers, it would be naive to think that other AI applications won’t also break through into the mainstream. The most in-focus of these are different flavours of generative AI — be that multi-modal AI on existing devices, or AI smart glasses. At the show, both my colleague and I tried different types of smart glasses. While I was surprised by how natural it felt to be able to interact with AI without a device in my hands, I can’t reconcile myself to a world where everyone walks around wearing cameras on their faces.
Let’s entertain the idea that this could be the direction of travel, at least in one form or another. For example, I was quite impressed with the example presented of smart glasses for tourism from China Mobile. I can see a world where users could rent smart glasses to receive bespoke tours of cities and tourist attractions, in their native language.
Understanding network requirements for AR/VR glasses
But what would this mean for networks? At the moment, onboard AI processing means that video streams from smart glasses can be handled over an upload connection of about 1.4Mbps. Some predictions for future AR/VR glasses even suggest requirements up to 10Mbps upload speeds, and as little as 20ms latency. So while we’re a long way from smart glasses entering the mainstream for the general consumer, this does represent a real difference from the way networks are designed today. It’s not an insane idea that telcos should plan for these different network characteristics as a potential future need.
Physical AI remains more hype than substance, despite being present at MWC26
Where we land squarely in hype over substance is so-called “Physical AI”. While not a focus of the show as they were at CES 2026, we heard talk about physical AI from vendors like Huawei and Ericsson, and even saw several prominent examples around the show. Each morning walking into the show, a digital billboard cooed at us about the world’s first robot phone. Walking the halls I saw AI pets, and even humanoid robots, such as Honor’s humanoid “service robot” for elderly assistance. Humanoid robots likely have quite challenging network requirements — assuming the robot actually needed to leave the house and venture out into the world. But regardless, I don’t see my grandmother opting for a non-human companion any time soon. Perhaps there are bigger fish to fry in the telecoms sector before we solve for this one!
What about “agentic AI”?
And then of course there’s “agentic” AI. The utopian vision here seems to be that everyone will have their own AI personal assistant, working in the background to improve their life. The reality here is that current consumer agentic offerings — like those announced by Samsung — are mostly iterative improvements on LLMs. But as these services become more anticipatory — and more integrated into service APIs — in theory, we could see more network traffic during times when users themselves are not online.
But as I said at the start, LLMs seem pretty lightweight in their network requirements. Even if an AI is working constantly in the background, I can’t quite see why in a consumer context we would need ultra-low latency or massive amounts of network resources. After all, it’s not that involved for an AI agent to call a bunch of API endpoints and then politely suggest my stress levels are rising, and that I might like to consider this Costco offer for a wellness retreat in the Caribbean that’s within my price range.
Where do we go from here?
So where does that leave us? Not in a world that needs 100Mbps uplink for your grandmother’s robot companion — at least not yet. But it would be short-sighted to entirely dismiss the direction of travel. My best advice? Plan for flexibility, but watch real adoption (not booth demos). And then if or when real consumer AI network demands — and revenue — actually materialize, well, that’s what slicing is for.
If you want to hear more about our analysts’ takes on MWC26, watch our wrap-up webinar on demand here. Or, to keep up to date with our latest thoughts and analysis, be sure to subscribe to our newsletter here.
The editorial views expressed in this article are solely those of the author and do not necessarily reflect the views of Opensignal.
Opensignal Limited retains ownership of this insight including all intellectual property rights, data, content, graphs & analysis. Reports and insights produced by Opensignal Limited may not be quoted, reproduced, distributed, published for any commercial purpose (including use in advertisements or other promotional content) without prior written consent. Journalists are encouraged to quote information included in Opensignal reports and insights provided they include clear source attribution. For more information, contact [email protected].
