From the newsletter: AI tokenomics is unlikely to rescue telco revenues; history shows metering models tend to commoditise. Long-term value will come from edge services, distributed AI, and networks embedded in broader AI execution environments.
If history tells us anything it is that unit economics always goes south. Right now, the tech industry is talking like AI revenue will follow AI demand – a little delayed, but smart-ish, and almost exactly: more prompts, more tokens, more value. But that is not the way this goes. The great infrastructure booms of the digital age – bandwidth, compute, storage – were metered and monetized, and none of them hockey-sticked into perpetual profit growth. Demand spiralled upwards and value spread outwards, and went elsewhere across platforms and applications – while the price for an internet session or a CPU cycle plateaued and declined.
So we should take Huawei’s address about telco tokenomics with a pinch of salt – and maybe just as a pitch to rally the troops ahead of the 6G investment cycle.
Because if that’s the grand plan – billing by the token – to revitalize telco economics, then it seems pretty lame. Yes, it is something telcos should do, as a way to meter unpredictable AI traffic on their networks, and it sure sounds clever (novel and alliterative), but it is only another unit metric for connecting workloads – which in the end no one will care very much about, just like their minute, SMS, and megabyte bundles. It is not a long game, and it assumes, somehow, the token itself is valuable – that AI creates token demand, which can be monetized by telcos in token transport and access. That is the bull case. For frontier model builders, tokens are the whole business model.
They largely / hardly-enough monetize through token-based subscriptions and licenses. But we have also been here before; this stuff usually gets commoditized. Will tokens be abundant and cheap like bytes of data and CPU hours or more scarce and valuable, and fought over, like oil. More than that: which will they be for telcos, which have a history of confusing traffic growth with revenue growth. More probably, they become an internal wholesale metric, while customers pay for agents, workflows, solutions – and outcomes!!! That’s the bear case. Ask the IoT crowd, or the private 5G one – which have re-learned how to sell critical tech, in ways funny talk about tokenomics does not address.
Telcos should be careful about assuming that just because model companies charge by the token today, tokens are where long-term value resides – especially while the whole AI market is being continually optimized for cheaper inference, smaller models, local execution, and token efficiency. In fact, every major force in AI is pushing against token scarcity. I mean, that’s the real play for telcos, right, as we have written in these pages for a year at least – how they are embedded into the edge-creep, as inference moves to devices, machines, appliances in edge infrastructure – and as the token model starts to look like the old cloud model: initially valuable, quickly commoditized.
See the interview last week with Orange from FutureNet World, and the review this week of Nokia at DTW – about cross-domain edge-to-cloud telco platforms anchored in critical services, sovereignty, and trust – rather than just connectivity. And actually, Huawei and China Telecom are on firmer, if more familiar, ground when they talk about AI-native orchestration. Strip away the token rhetoric and what remains is not really a new monetization unit at all, but the same drawn-out shift: connectivity absorbed into a broader services fabric. Which is where the story is: not about a “token economy” that rescues telcos, but a subtle re-rating of networks as part of the AI execution environment – hosting inference, enforcing policy, enabling locality, and binding distributed compute together.
