Here’s how Telenor is turning itself into an AI factory

It’s no secret that telcos are keen to cash in on the artificial intelligence (AI) boom. But the question many have run up against is how exactly to tackle the problem. One operator, though — Telenor — thinks it has figured out an answer.

Telenor Director of Research and Innovation Frank Elter told Fierce at the recent TM Forum event in Copenhagen the key is adopting an “ambidextrous” approach that allows it to continue operating its mature telco business while also creating a new unit focused on AI innovation.

In Telenor’s case, that has meant setting up a separate legal entity – Skygard – through which it is working to create an AI factory of sorts. Basically, it’s become a telco with an AI startup business on the side.

In March, Telenor teamed with Hafslund and HitecVision to invest $225 million (NOK 2.4 billion) in Skygard and the construction of the venture’s first data center in Oslo. Last month, Telenor announced plans to roll out AWS’ sovereign cloud technology in Skygard’s data center.

“We organize new business opportunities on the side of our core business to innovate. This a well-established practice to achieve radical innovation," Elter explained. “I think the challenge is that we need to do many different things simultaneously that require different skillsets and different subcultures.”

Wait, you might say. Why not just leave the startup scene to private equity? Well, Elter said the key is for telcos to “as soon as possible to identify how we can utilize the capabilities in the existing operations. That’s how we differentiate.”

He added that the ambidextrous approach also allows telcos to more easily adapt to the much faster innovation cycles AI adheres to. Whereas telcos are used to thinking in years-long timelines, he noted AI and cloud computing innovation cycles are much shorter – often on the scale of months.

“This is truly a learning journey for telcos. We are not used to working in this way. In the past it was much more sequential work…now we need to move much more in parallel” to speed development along, he said.

Elter also pointed to the importance of partnerships as part of the innovation journey and said these too will be different from what telcos are used to. Whereas in the past, telcos signed contracts with partners expecting a set outcome, it needs to be understood going forward that “if you innovate the outcome will be uncertain upfront. This puts new requirements on how partner contracts are developed,” he noted.

“You need a vision for where you want to go but the challenge is you can’t succeed only by the grand master plan anymore,” he said. “If you just try to plan for the big solutions, customers might not want it. You need to be able to pivot.”

So, what do analysts make of all this? AvidThink Founder Roy Chua said of Telenor’s approach “we have seen this many, many times.” By way of example, Chua pointed to Deutsche Telekom’s venture MobiledgeX, which was founded in 2018. The operator eventually sold the edge startup to Google in 2022.

Telenor itself has done something similar before. The operator spun up Working Group 2 in 2017 to build a cloud-native mobile core. It later sold the business to Cisco in 2023 for $150 million.

“It is feasible, but a lot of these in the past haven’t quite succeeded,” Chua said. Basically, he said, it comes down to whether AI talent would want to join a telco offering likely lower pay, less freedom and less prestige than other AI startups might.

Chua said the only reason for a telco to launch its own venture rather than having its investment arm pour money into a third party startup is to ensure the business achieves some sort of telco outcome.

“When you influence that heavily, then it becomes difficult to recruit because then the people don’t want to work on that because you just built a model for telco,” he said.

Chua concluded the best model for telcos is one akin to what SK Telecom is doing with Anthropic where the operator is paying $100 million to have Anthropic’s team build a large language model for telcos. That sort of arrangement gives the AI engineers at Anthropic the flexibility to go work on projects for other verticals in the future.

“That model could work,” he said.