On the bank of the River Liffey, as it races out to Dublin Harbor, the Convention Center Dublin played host to the Telecom Infra Project’s FYUZ 2025 conference. The tone was muted, with attendees keen to note it felt a lot quieter than the previous installment, and while the keynote stage had been moved to a less cavernous room, the messaging from the stage made it clear that data and AI were more of the focus for TIP moving forward.
At one point in recent history, TIP felt almost synonymous with Open RAN – having grown from its foundational focus on opening up telco data center and interconnection specifications, all the way out to the RAN itself. But in Dublin, Open RAN was far from the central focus.
Instead, the underlying subtext was data. Here, all talk of new AI-based tools seemed to stress the need for data to be generated in a hygienic fashion, such that these new agentic and generative tools can have the best chance of making use of it. In addition, with AWS joining the board, compute is now a major focus.
The other side of this coin was an uneasy acceptance that 6G conversations need to start happening very soon. Both vendors and operators appear reluctant to admit this, as both have been burned by generally disastrous 5G rollouts – something that the LAN, WiFi, and interconnect factions within TIP are quite happy to highlight.
These roadmaps clash. MNOs are slow movers, and embracing any new technology takes much longer than either the MNO or the supplier expects. This is especially true when changes are being proposed to the underlying operational architectures, such as those that these new data-focused AI-based optimization tools propose, which brings them into the 5G Standalone and soon the 6G window.
5G Standalone should be a clean slate for MNOs. It should enable a truly IP-native network core that can leverage all the necessary cloud-based technologies, supporting the messaging, voice calling, and media subsystems – with ports for any LTE Cat-NB or Cat-M operations. 5G Standalone should be a line in the sand; the chance to ditch all the legacy headaches on the operators’ books.
But we all know it won’t be. 2G will linger. Just look at the farcical emergency calling mandate for vehicles sold in the EU. 3G sunsets have exposed coverage gaps, and while 4G is mostly a known quantity, large portions of the mobile customer base still make calls on 2G and 3G era voice subsystems – never mind VoLTE or the new VoNR.
With direct-to-device offering some potential relief (migrating legacy 2G operations to these new satellite constellations, filling in terrestrial footprint gaps), the number of 5G Standalone deployments is still a fraction of the global total. Of the 700 or so mobile networks globally, just 77 across 43 countries have launched or soft-launched 5G Standalone services, according to the GSA.
Some 173 across 70 countries have begun “investing,” but now six years post-standard finalization, the lack of progress on the 5G front means that by the time the 6G rollout begins in 2030, perhaps only half of the total global community of MNOs will have made it to 5G Standalone. Per Soni’s keynote opening, 5G might have 60% of total connections by 2030.
Currently, around 275 have deployed 5G Non-Standalone, which uses the 4G network core to underpin the New Radio (NR) used in 5G. For most, Non-Standalone will also have to be ripped and replaced, on the journey to Standalone.
The board sit-down
And yet, the TIP is working to speed up these journeys – not just in the Open RAN working group, but in Neutral Host & Infra Sharing, Open Optical & Packet Transport (OOPT), OpenLAN (including OpenWiFi), and in the newer TelcoAI working group.
We met with the OpenLAN and OOPT groups, as well as the board itself at the very end of the show. We are not sure if the OpenRAN group was holding briefings, and TelcoAI made something of a debut with a three-hour breakout session to kick off the final day, which we caught the end of.
These groups will be explored next week, but our meeting with the board is worth documenting here. Over the course of the final hour of FYUZ 2025, just before a major board meeting, we sat down with TIP Executive Director Kristian Toivo, Chairman Rob Soni from AT&T (VP RAN Technology), and Board Members Kaniz Mahdi (AWS – Director Technology of AWS Industries) and Cayetano Carbajo Martin (Telefonica – Director for Core, Transport, and Service Platforms).
There were just 23 journalists and analysts at FYUZ, out of a reported total of 950 registrants. We know most journos headed off at the end of the second day, and suspect that 23 also included photographers and videographers too. This is all to say that it was still surprising to have just two other members of the press and analyst community at this final meeting – with your writer wearing both of those hats.
“A small but very influential show,” was how AT&T’s Soni put it. “This is not the race of MWC, but we’re getting to the bottom of the issues – at the peak of the hype cycle for AI. Perhaps there was too much Open RAN on the agenda last year, with other pieces underrepresented,” Soni noted, in the opening remarks.
A joke that perhaps Open RAN 2.0 could be launched (encompassing AI RAN efforts) was made, before Soni reiterated that there really was a lot of interest in AI in the TIP community, but that the question was how to ensure that the industry moved in harmony. “But, seeing AI on the agenda so many times didn’t make me uncomfortable,” he noted, indicating that perhaps the volume of Open RAN in last year’s keynotes did.
Telefonica’s Carbajo Martin pointed to the IP over DWDM interconnect trend as a popular one, and teased that the TIP was about to begin work on a Quality of Experience (QoE) project.
“We are having lots of very interesting conversations,” he said, “but AI is still mainly in the operations side of things. Still, we would like to see how to use it in the Neutral Host group, as well as explore how to change the network to suit AI. There, the question will be how to access this data for the operations functions.”
AWS’ Mahdi pointed to the AWS-sponsored hackathon that was mentioned several times over the course of the show, as proof of the advances being brought to the telco sector by tools like AWS Bedrock and large language models.
“This is the right time to talk about AI. Two years ago, this was not the case, but I saw the same problem then in Deutsche Telekom as I do now. Sitting in AWS Telco, some of the Open RAN problems are simply scaling problems, and those have been solved by the cloud industry already – which has already gone through the cycle of testing to monetization. Our hackathon was proof of this agentic potential,” said Mahdi.
TIP’s Toivo agreed, calling the video summary of the hackathon “mindblowing.” He added that it takes time to drive a community forward, it “also needs a little bit of luck.”
Telefonica’s Carbajo Martin built on the agentic angle, stressing that “the main challenge is the access to the data. So, the standardization of data, not only in TIP collaboration, is key. In the transport sector, this is all to do with the software-defined network architectures and the use cases that have them already.
We should have many similar gains in AI, and a PoC with AWS is proof of this. It is one of the few cases where we can all work together to solve those problems.”
AT&T’s Soni said there was little to add on this point, “but we are seeing lots of challenges with data and schema. The northbound RAN connections are still not as open as the transport or the core, but now that agentic tools can probably reverse engineer the interface, we are seeing more of those connections being opened up.”
This would be openness via futility, in some ways. If a vendor cannot hope to keep that somewhat lucrative interface under its control, then it is for the collective benefit that you make it a public interface, as you know a rival vendor will eventually be able to throw enough tokens at the problem and force openness via brute force.
This does address one of our main complaints with Open RAN work – that while it was good to standardize things like the RU, DU, and CU, the main problem for the MNO users is the integration of these components together into one cohesive project.
The various standards groups did not have a good single answer for these gaps between the components, and sure enough, agentic tools running on platforms like AWS Bedrock are a fix – but one that was not really on the cards when Open RAN work started.
Startup scaling problems
The other main complaint leveled at the Open RAN space is the classic ‘single-vendor Open RAN’ project, where the likes of Ericsson or Nokia come to dominate project bids, thanks to their scale.
More barbed criticism accuses these vendors of trying to hijack the standardization process to their advantage, while many vendors are keen to criticize the MNOs themselves, for only using Open RAN advocacy to try and drive their contract prices with Ericsson and Nokia down – and not having any altruistic interest in the Open RAN trend.
This dynamic was brought up by AT&T’s Soni, who said “the number-one rule is not to kill your startup partners.” On the show floor, we met former Mavenir SVP Sridhar Rajagopal, who has now founded Ennoia Technologies – a startup looking to use small language models (SLMs) running on Intel NUC machines to provide network monitoring and predictive maintenance agents at the base stations.
Rajagopal said that the biggest problem facing startups like his was the lack of access to live commercial network data to experiment and develop with. So, we asked Soni what the likes of AT&T would do to support startups on this footing.
“Governments need to give us some relief on the Secure Private Information (SPI) front. Legislation can block progress here, but governments often view this as an industry problem that we must solve,” said Soni, with Telefonica’s Carbajo Martin saying that this problem does not vary much by region.
So, for startups, consumer data protection regulations are a real problem. Government funding flows into these sorts of industry efforts, and yet regulations would prevent the likes of AT&T or Telefonica from letting startups actually get to grips with these problems.
Soni said that much the same concern surrounds bringing any customer data to foundation models and LLMs, but that “the boundaries of the problem are starting to be understood. There is an opportunity to move quickly, here.”
Carbajo Martin was also confident that “visibility of customer SPI shouldn’t be such a problem. It’s very far removed in the data models,” while Mahdi stressed that this was a very solvable problem.
We put the imbalance of influence between the major vendors and the startups to Toivo, who stressed that it would be brought up in the board meetings. “What we learned from the hackathon is that we need to bring new people into the sector.”
Soni added that “bringing the larger vendors back into TIP should make sure that they don’t ignore the needs of the smaller vendors. The vision is big and small, on both operator and vendor, and it feels very different to the TIP of five years ago.” Toivo concluded that there is a lot of motivation to do this.
The AI of it all
“AWS’ selfish ambition is to build 6G. But, the big red flag is the fundamental assumption that the network is AI-native,” said Mahdi. “Remember, 5G was meant to be cloud-native. Look at where we are today. So, starting at the basics of that AI-native point, there are two key customer barriers. Data is one, and the other are the seemingly open interfaces that are not actually open.”
Soni interjected, keen to stress that they are actually open, but not mature – viewing Open RAN as perhaps the first problem in a journey towards full network automation.
Mahdi then said that in this approach “instead of begging vendors for each step, we could train SLMs on the interfaces – eventually aiming to automate away the human in the loop.” [Mahdi reached out post-publication to confirm that work on this had not yet started, and that the TIP is working on the Network Language Models – NLMs]
Carbajo Martin said the process might be more bit-by-bit, “like Jack the Ripper,” causing some level of consternation from the board. We thought this was an excellent line, however – reflecting the painfully incremental process of the work.
This circles back to the points raised at the start of this article, and our usual rant about the toxic co-dependence between the operators and the vendors.
Efforts from the likes of TIP aim to address this problem, by opening up various pain points, and now that these telco-focused SLMs and LLMs can be brought to bear (on very strict guardrails, we strongly advise), there is an air of hope in the industry – that the dissection and reconstruction of every interconnection between the myriad components and functions in the telco stack could be done by these machines.
Still, opinions on AI are somewhat a question of personal politics and philosophy. For the beleaguered telco industry, it is more a question of not looking this gift-horse in the mouth. Now, onto the 6G planning questions …