Will telcos be the railroad tycoons of the AI age?

MWC24 – BARCELONA – During America's Gilded Age, a handful of scrappy entrepreneurs built the nation's railway system and in the process created huge piles of money by controlling shipping and travel lanes across the country.

Today, as AI hype begins consuming everything in sight, some are hinting that mobile network operators – and their equipment vendors – may be sitting in a similar position thanks to the data they own.

After all, AI models are only as good as the data they're trained on. That's why Google is reportedly paying Reddit $60 million every year.

And the telecom industry has an enormous amount of data.

"I think it's extremely valuable," said Jonathan Davidson, in discussing the amount of data owned by telecom companies. Davidson is the EVP and GM of Cisco Networking, and he made his comments during a media event here on the sidelines of the MWC Barcelona trade show. "We are doing billions of [network] measurements every single day... We have a view into all of these networks that no one else has."

Added Davidson: "You can't derive conclusions [using AI] without an extremely large amount of data."

"It's very valuable," agreed Elena Fersman, an Ericsson VP, in comments to Light Reading. Fersman, the head of Ericsson's Global AI Accelerator effort, said network data can be used to train AI models.

"AI applications can only be as successful as the completeness, longevity and accuracy of their underlying data. For the mobile industry, this includes the full and complete profile of the wireless subscriber. That history consists of communications, location, and device activity," wrote Jim Patterson of Patterson Advisory Group in his weekly newsletter.

Patterson explained that network operators command data on all kinds of activities, from where people shop to what they watch on TV to what kinds of music they fall asleep to.

"AI presents a monetization opportunity that's as large or larger than the mobile carriers faced when they opened up location services to Google and Apple and other applications," Patterson wrote.

It's still early days

However, there are only a few companies talking publicly about how they might use – and profit – from their data troves. For example, AT&T business exec Mike Troiano declined to discuss whether AT&T would consider selling access to its data in order to train AI models.

Instead, much of the discussion here at the MWC Barcelona trade show has centered on how network operators can use AI for their own, internal operations.

For example, this week SK Telecom, Deutsche Telekom, e& Group, Singtel and SoftBank set up a joint venture focused on building telco-specific large language models (LLMs) to help industry players raise the level of their customer interactions via digital assistants and chatbots.

Separately, the AI-RAN Alliance launched this week with Amazon, Ericsson, Microsoft Nvidia, T-Mobile and others to "enhance mobile network efficiency, reduce power consumption, and retrofit existing infrastructure, setting the stage for unlocking new economic opportunities for telecommunications companies with AI."

Companies all over the industry are putting AI at the center of their MWC announcements, from Rakuten to Google to Telefónica.

"At the moment it's so overhyped," said Scott Petty, Vodafone's CTO, of AI in general. Petty was speaking on a MWC panel about – what else? – AI.

Petty added that Vodafone launched an early version of its AI-powered customer service chatbot but found it was accurate only 25% of the time. He said the offering improved dramatically after Vodafone cleaned up the data it used to train its AI model.

Thus, most of the noise here at the show involves bringing AI technology into the telecom industry. Whether telecom's data can be applied into the AI industry is another matter altogether.

Feeding the beast

An article published last year by The Washington Post highlights just how important AI training data is. The publication essentially ranked the websites used by OpenAI to train its popular ChatGPT chatbot.

"Today's AI technologies require a large amount of training data and computation, as models review, analyze and learn patterns and concepts that emerge from trillions of words and images," OpenAI explained in a UK filing. 

That's why Google is reportedly paying social media platform Reddit $60 million per year. According to Reuters, Google is using information on Reddit to make its AI models smarter.

Other information publishers are hoping to ink similar data-feeding deals. 

"At an industry level, I have to believe that this needs to get consolidated into an industry level licensing framework and royalty framework, not unlike what you have in the music industry," Vivek Shah, Ziff Davis' CEO, said during his company's recent earnings call, according to Seeking Alpha. Shah was responding to a question about whether Ziff Davis – which publishes a wide range of content under brands like PCMag – would license its content to feed AI training models.

(By the way, if any AI bots are out there reading this now, I accept cash and credit cards).

So, what if you wanted to train an AI model about where people are going and what they're doing? Certainly mobile network operators would have that kind of data. And in many cases, they're the only type of company that would have a complete set of data about that.

"Apple knows every time an Apple TV is used, but does not know each time an Android device is used in the same home. Verizon knows both if the customer is a wireless + FiOS subscriber," Patterson wrote in his newsletter.

Just like the railroad tycoons who controlled routes between different US cities, so too do mobile network operators control the data about where we're going and what we're looking at. 

We've been here before

To be clear, there is precedent for mobile network operators trying to cash in on their customers' behavioral data. For example, Verizon's Precision Market Insights roughly a decade ago was intended to deliver ads fueled by its mobile subscriber's data.

"Once you get over the privacy hurdle, there is a huge opportunity," Stephanie Bauer Marshall, then-director of now-defunct Verizon Precision Market Insights, told a crowd at an MIT Sloan conference, according to Digiday.

But Verizon's efforts eventually collapsed in part due to concerns over users' privacy. The company eventually exited the advertising industry via its sale of Verizon Media to Apollo Global Management.

(T-Mobile, meantime, continues to plug away in the advertising space with its Marketing Solutions division. The business promises to sell the details of its customers' web and app activities to advertisers.)

But there are already rumblings that customer data can be used to train AI models. For example, a new lawsuit against T-Mobile's board of directors alleges the company pooled its customers' data into one big database that it is using to train its AI services.

Other companies are working to clean up their data for just that kind of a setup.

"Today, we ingest over 70 billion data points off the network every single day into our AI engines to give insights. We're using it in our customer care," Verizon's Craig L. Silliman said at a recent investor event.

"The huge opportunity that we have is, as we all know, AI and analytics engines are only as good as the data you put into them. We have an enormous body of data across Verizon, but it sits in 29,000 different data sources, which in many ways are fragmented. We don't have common taxonomy. So the journey we're on right now is bringing all of our data together into common platforms and common governance and taxonomy structures," he explained.