• Tokens are becoming the currency of AI — and potentially of networks themselves
• Telcos face a choice: carry AI, or think in its terms by transitioning to an AICO (AI Infrastructure Company) model
• Control is moving up the stack, from networks to models
Buzzword, right ahead! Telecom is heading straight into its latest vortex of hype with the concept of AI tokens.
Tokens themselves aren’t new. They’ve been used for decades in natural language processing: small segments of text representing a word, part of a word, or punctuation. What’s changed is their elevation from a technical detail to the core unit of computation in modern AI systems.
This is now reaching industrial scale. T-Mobile, for example, has passed a benchmark of processing over 1 trillion tokens annually as it integrates AI deep into its operations.
Large language models like GPT-4 process everything as tokens — from input to inference to output. And companies like OpenAI, Anthropic and Google have built businesses around this, charging for AI not as software, but as a metered service: priced per token, or per unit of thought. This shift is creating a new field of token economics, where the value of a network is determined not just by capacity, but by how efficiently it can generate and transport these units of intelligence.
Given the growing importance of AI workloads, there’s a compelling case for telecom operators to extend traditional packet-based metrics — packet loss, latency, jitter — with token-aware capabilities. Not to replace packets, but to complement them.
In this new paradigm, traditional network latency is replaced by time-to-first-token (TTFT) as the key performance indicator, measuring how quickly a network can deliver the start of an AI response to a user. And instead of defining service quality purely in packets per second, operators could begin tracking tokens per second.
That’s a simple shift in framing, but it opens up entirely new dimensions of measurement and optimisation:
Token-based pricing models (the core of modern token economics)
Network optimization based on token efficiency and TTFT reduction
Visibility into model inference performance and hardware utilization (GPUs/TPUs)
Energy consumption per token or per application (a critical KPI for the new Net-Zero AI mandate)
This aligns with how modern AI systems actually operate: processing large volumes of context, reasoning across long inputs and maintaining continuity across interactions.
But this is where things get sticky.
Telcos expose APIs grounded in physical reality — identity tied to a SIM, verified location, live network state. Hyperscalers, by contrast, sell cognition: systems that interpret, generate and decide, priced in tokens. One layer authenticates the world. The other reduces it to something that can be processed — and acted upon. This is the core of the Telco-to-AICO transition: moving from authenticating a user to facilitating an inference about them.
Tokens won’t replace packets. They travel inside them. But they introduce a new abstraction layer — one that sits above the network, its services and applications — and increasingly defines how value is created. As we see with the T-Mobile trillion-token benchmark, the volume of this “cognitive cargo” is beginning to rival traditional data traffic in economic importance.
All in?
Because the real question is not whether telcos must adapt to token-based AI traffic, but whether they should adopt the same abstraction themselves.
In one model, tokens remain external: AI systems generate traffic, and networks evolve to carry it efficiently. In the other, more radical shift, the network itself becomes legible in tokens — its state reduced to inputs, its behavior to outputs. At that point, the distinction collapses. The network is no longer just infrastructure, but a substrate for “agentic networking” — systems that interpret and act upon network state in real time.
Today, AI traffic is still a small proportion of the public internet, even as it dominates inside data centers and the burgeoning regional AI factories. But if telcos want to participate in the emerging AI economy alongside hyperscalers, as peers — both commercially and architecturally — they will need to operate in the same units. This shift is the cornerstone of modern token economics; without token-level visibility, telcos remain blind to the value they are transporting.
“We’re working on this because customers are asking us about it,” says Stephen Douglas, head of market strategy at Spirent, now part of Keysight, who points to tokenization as one of the most important new opportunities Spirent is now focused on.
One major vendor is already in no doubt about the significance of tokens. Gurudatt Shenoy, senior VP, Datacenter and Provider Connectivity at Cisco Systems, puts it like this:
“The AI economy will be defined by token economics — the cost to generate and deliver tokens at scale. That’s why we’re building network infrastructure for a changing economic model that demands efficiency as measured by tokens per dollar per watt.”
To stay in the game, telcos may need to do more than make a token effort.
Note: On April 8th, I’ll be hosting a webinar with Stephen Douglas, head of market strategy at Spirent, covering tokens and other next-generation challenges. I strongly encourage you to join us — it’s going to be a doozy! Click here to register.
And for more of this kind of thing, click here.
Stephen M. Saunders MBE is a communications analyst and USPTO-registered inventor examining how digital infrastructure — 5G, cloud, and AI — is reshaping industry, power and society, as well as underpinning the emerging, ubiquitous global digital economy. As anchor of FNTV and a longtime industry insider, he focuses less on growth narratives and more on execution, risk and how hyperscale technology is distorting markets, governance and society at scale.
Opinion pieces from industry experts, analysts or our editorial staff do not necessarily represent the opinions of Fierce Network.