Managing director | Ex-Google | Ex-Ericsson | Founder | Author | Doctorate Candidate | Follow my weekly newsletter
The Telco AI Big Bang We still don’t get it. AI traffic is not “more video.” It is a completely different physics problem. AI agents will generate unique payloads, communicate directly with each other, and demand deterministic SLAs. This is not about ChatGPT prompts. It is inference bursts, uplink firehoses, stochastic variance, and agent-to-agent protocols that our networks are not designed to handle. Every layer of the stack will be exposed. Radio networks, built for downlink dominance, are unprepared for persistent uplinks from drones, glasses, and sensors. Transport layers, designed around fairness, will collapse under inference-driven microbursts. Cores, optimized for long-lived sessions, will struggle with millions of ephemeral agent-to-agent flows. And edge sites without GPUs will simply drown in uplink traffic they cannot process in time. The economics are no safer. Selling gigabytes per month is irrelevant when agents are buying latency guarantees, jitter bounds, and compute cycles. Value shifts from capacity to deterministic performance, and if telcos don’t adapt, that value will move to hyperscalers who already bundle bandwidth and compute. The news is already here. YouTube has announced that it will prioritize AI-generated videos over those created by human creators. A platform responsible for nearly 20% of global internet traffic is shifting to synthetic, non-cacheable content. Could you try running that through a CDN and see what happens? This is the Telco Big Bang. Machines, not people, will be the primary customers of our networks. If we continue to design for humans while the traffic shifts to AI, we will miss the inflection point entirely. It is time to start dimensioning for the world that is already arriving.
Look to the M&E Industry to see what happens when an Industry changes due to a fundamental technology shift (digital) and the leading firms don’t transform their core technology architecture. (The Hyperscalers are Thankful!) This is why Telcos must reinvent themselves as horizontal platform companies. A redesign for “compute + communications” is the only path forward.
⚠️ This is BS. YouTube has not announced any policy to prioritize AI-generated videos over those created by human creators. Instead, recent updates focus on providing AI tools to help human creators produce content more efficiently, such as generating Shorts, editing videos, and creating music. For example, at the "Made on YouTube 2025" event two days ago, they introduced features like Veo 3 in YouTube Studio.
Sebastian Barros YouTube is doing the opposite, low-effort AI generated video (vs. using AI in the creative process) is a serious problem for them, it's not engaging and negatively impacts ad revenue. On your bigger point, I see the challenge differently - regardless of how AI impacts usage and network demands, telcos have effectively outsourced strategic product decisions to technology suppliers and aren't equipped to adapt to *any* significant or rapid shift.
"YouTube has announced that it will prioritize AI-generated videos over those created by human creators" Do you have a reference for this? I think this is explicitly false. Would love to see evidence to the contrary.
Yes, machines will be the new actor in the game. Current networks are somewhat designed and sized for a model that has humans as end consumers,and as such many metrics are a consequence (there's so much your eyes can digest as content, beyond that, you might not need more). The only caveat with machines being new actors is to determine where it makes sense to have AI far from the device, or AI in the device. E.g. A new phone from any vendor, embedding GPUs, the advent of SLMs, it could probably render pointless to change the ways networks are designed. Good stuff in the air :)
Another Dot.Com bust in the making. How difficult it is to learn. Seems AI is being led by natural stupidity. AI is important, do not get me wrong but these cure all aspects of AI just do not apply...for some time now. We should be spending more time on the legal, social and intelectual property impacts this all brings with it...
AI traffic breaks the old internet model.
I’m still trying to get DTMF to register a “0” to reach the operator in my Octel OS/2 Powered VM system. 386 SX 25 4MB Ram is struggling with the open source model 🤣
Forbes Council Member | Sales & Account Management Executive | AI & Digital Transformation | Expert Network Advisor | VP-Level Experience
3dThis is spot on Sebastian!! The challenge isn’t just technical, it’s existential. Telcos have a narrow window to redefine themselves as platforms for machine-to-machine economies. If they keep optimizing for human-centric consumption models, they’ll be relegated to utility pipes… while hyperscalers capture the premium layer.