Token traffic — network carriers are getting their slice of the AI revenue cake

The blockbuster, biggest stock market listing – or initial public offering – in the history of the world by Elon Musk’s SpaceX was a big test of the retail investor market’s AI appetite. The payoff is that South Africa’s prodigal son will move AI data centres into space to free us on terra firma from the resource-­intensive monstrosities.

The kicker, though, is the rapid onboarding that the listing enjoyed on the Nasdaq-100 index fund, and the limits on selling off placed on eager retail investors. Not only did the SpaceX listing vacuum up pension funds and life savings, but it also showed the likes of OpenAI and Anthropic that there is still life in the AI horse as they prepare to list shares.

This gobbling of capital is happening while all AI suppliers are raising prices for users and taxing the very enterprises to which they sold the automation idea in the first place – which is, ironically, making the AI solutions more expensive to run than the workers they replaced. Just this week, Microsoft hit its business users with a 16.7% increase on the basic package and an insane 33.3% increase for front-line workers. Why? Because it’s bundling in advanced AI capabilities.

These increases are coming off the back of capital expenditure that has almost doubled at all the mega tech companies building out AI infrastructure. The cost is making its way into steep premiums on using the latest AI models, for which companies and other users are billed via tokens – units of raw text or data processed to make up prompts and generate outputs.

Developers are facing peer pressure to burn through as many tokens as possible in a cash-fuelled pissing contest, and private users have noticed how quickly they hit their usage limits lately.

It’s an AI world and we’re all just working to pay off the investment.

A token gesture

The Mobile World Congress Shanghai showcased how the scale of token generation is growing at a truly staggering pace. Mark Lian, technology sector partner at Deloitte China, presented data explaining that “from 2024 … token consumption globally has been rising from 114 trillion tokens to the estimated 400 trillion tokens in 2026”.

In China alone, token consumption was just 100 billion tokens in early 2024, but this is projected to skyrocket to 100 trillion when the final numbers through to the end of 2025 are in.

As one can imagine, transmitting and processing these tokens place immense strain on network infrastructure, particularly on uplink speeds, because AI devices constantly send data to the cloud. And these tokens are riding the same data microwaves as TikTok videos and WhatsApp video calls, so naturally the network operators are starting to see dollar signs too.

Telecoms operators are realising that AI must evolve from an internal operational tool into a direct revenue stream. Shiwei Sun, deputy general manager of the market operations department at China Mobile, had the best quote about this shift: “As AI enters the era of scaled commercialisation, the token is becoming the foundation of the AI economy.”

What has followed this hunt for revenue is something called multi­dimensional token monetisation mechanisms. Eric Yang, president of Huawei’s carrier business, gave lots of advice during his presentation, saying network providers must “reshape products and services with AI to unlock new growth, while up­­grading compute-network in­­fra­­structure to enable efficient token monetisation”.

How does he suggest they do this? By “transforming traditional services into AI companions and offering enterprises tiered, integrated compute-network services that combine private lines with digital employee platforms”. This means slicing up the 5G network and selling dedicated bandwidth – at a premium – for token traffic.

Bringing it home

It is well known that South African telecommunications operators are investing heavily in physical network infrastructure to support these AI-driven workloads. For instance, Vodacom has committed to an R85.2-billion Vision 2030 programme, and MTN is transforming its network to extend 5G coverage and scale AI-driven operations.

However, regional research from enterprise data and AI platform Cloudera warns that these multibillion-rand upgrades risk hitting an AI wall due to fragile data foundations. The core issue is not physical capacity – Huawei has closed the gap on 5G infrastructure to match the speeds and capabilities of even the forbidden US technology fruit – but rather something called data friction.

Athul Prasad, global director for AI industry solutions at Cloudera, has a great take on the problem: “South African telecommunications operators are successfully scaling their physical networks to handle AI-scale workloads, but physical readiness is only half the battle. Visibility over your data estate is not the same as utility.”

For South African operators, strict compliance with the Protection of Personal Information Act (Popia) means that sending massive data sets across public clouds introduces regulatory and security risks.

To resolve these issues, local business is being pushed to embrace a private AI strategy that brings AI models directly to where the data is stored (on-site or in private clouds) to protect data sovereignty, satisfy Popia and avoid cloud egress costs, which are charged per gigabyte on one’s AWS bill.

This, of course, aligns perfectly with the deployment of mobile private networks (MPNs). Local operators are already demonstrating the value of this approach. Sasol and Vodacom Business, for instance, recently installed a 5G MPN at the Secunda energy site, connecting 3,000 employees and thousands of IoT devices over a highly secure, low-latency private network that keeps data localised and out of public cellular congestion. (IoT refers to the Internet of Things, the collective network of connected devices and the technology that facilitates communication between devices and the cloud, as well as between the devices themselves.)

The transition from measuring network capacity in data volume to now counting tokens is reshaping the relationship between businesses and the telecoms industry. As mobile networks increasingly support AI agents and robots using agent communication protocols, the focus shifts from simply transporting raw data to ensuring reliable delivery of AI computations. Everyone at MWC Shanghai was talking about it and how the network operators can finally monetise this seismic shift in technology.

It’s nothing personal, it’s just business. And more money. Just not for you and me. We just pay for it. DM

Lindsey Schutters attended the Mobile World Congress Shanghai as a guest of Huawei South Africa’s enterprise business.

This story first appeared in our weekly DM168 newspaper, available countrywide for R35.