DTW Ignite Is Now the Agentic AI Telecom Event

Every year in Copenhagen, TM Forum brings the telecom world together at DTW Ignite with the goal of improving networks, adopting best practices from IT and enabling more-autonomous operations. This year, Huawei had its 11th OTF event next door. DTW used to be focused on BSS and OSS suppliers and their customers at telecom operators. This is no longer the case. DTW Ignite is now focused on telecom AI throughout the network, not just in the monetization, BSS and OSS domains.

The discussion and innovation at DTW were all about understanding AI, AI agents and developing AI’s use throughout the telecom market. During my three days in Copenhagen this year, I saw briefings on AI in the radio access network (RAN), in data centres and in fixed infrastructure — not just in mobile networks. A few years ago, it would’ve been unusual to see RAN come up in presentations or demos at DTW. Generative AI and agents are changing everything and helping operators to tie together their entire networks.

Footfall at the show looked to have dropped this year — probably because of the clash with MWC Shanghai and the FIFA World Cup — but the show floor and keynotes continued to be vibrant and well attended.

Speakers included chief information officers and chief technology officers from operators, such as Scott Petty (Vodafone), Antonietta Mastroianni (e& UAE), Chen Hong (Singtel), Charles Molapisi (MTN Group) and Kanwardeep Singh Ahluwalia (Deutsche Telekom). We also heard from CEOs, including Claire Gillies (BT), Vivek Sood (Axiata, former CEO), Pietro Labriola (TIM) and Jussi Tolvanen (DNA). Operator sponsors this year were China Mobile, Deutsche Telekom, Jio, Orange and Vodafone. As usual, operators and vendors collaborated on the numerous catalyst demonstrations spread throughout the show floor.

Cross-Domain Agentic AI

The major hyperscale cloud providers, Amazon Web Services (AWS) and Google Cloud, again had a big presence at DTW and announced new collaborations. Google Cloud partnered with Nokia on embedding Google Gemini-based AI agents into Nokia’s network portfolio. AWS also linked up with Nokia to enable Nokia’s Autonomous Networks Fabric on AWS as part of the shift to achieving level 4 autonomy.

With its recent reorganization and new leadership team, Nokia was making a strong case that it’s the ideal supplier that can deliver cross-domain AI in all areas. Given that the large operators typically offer services in fixed as well as mobile, and many are developing sovereign AI data centres, or what Nvidia calls AI factories, there’s clearly a fit with Nokia’s broad portfolio. AI is helping to boost Nokia as the industry looks to build AI-native 6G networks and improve uplink and capacity to handle the changing data traffic patterns caused by customer AI usage.

While Nokia is repositioning itself as an AI player after having had a rough few years — especially on RAN — Ericsson continues to aim to be a leader end-to-end in just mobile. One Ericsson demonstration jumped out. The company has taken the technology and architectural approach from its rApps platform for intelligent apps running on the RAN and extended it into its core network products with cApps. There are now more than 80 rApps published in Ericsson’s market. For now, there are a handful of cApps that tackle tasks such as predictive capacity, self-healing, the security of the digital stack and offering a premium user experience. Ericsson expects the first deployment this year with others following. It has lead operator customers in both Europe and Asia–Pacific for the solution.

Notably, Ericsson is also highlighting the end-to-end cross-domain potential here, with rApps and the new cApps able to communicate and collaborate. AI models are also key with cApps being able to query a data store using model context protocol, or MCP, to provide clear guidance to ensure the cApp delivers a predictable deterministic behaviour, unlike the variability of much generative AI output.

Amdocs further developed its AI positioning following the launch of its new cross-portfolio aOS branding at the start of the year. The company announced new network workflows that included the AI-RAN and announced a deal with PLDT — which uses the brand Smart for its mobile services — to bring Amdocs’s agentic AI-based product and Store Genie to its broadband customers. Amdocs also used Nvidia to support its proposition. The problem is, almost everyone is now partnering with or buying from Nvidia, bar Nvidia’s most-direct competitors.

Nvidia also had a presence this year, although it was smaller than Google Cloud or AWS. It made an announcement about its capabilities in supporting autonomous operations and AI agent deployments. As usual with Nvidia, there were multiple partners referenced, including Amdocs, ServiceNow, Viavi, Keysight, TCS and NTT Data and SoftBank. Nvidia’s communication challenge is in the breadth of its AI offerings, and how to ensure the most important things stand out. At GTC, Nvidia emphasized the agentic OpenClaw and NemoClaw opportunity, but at DTW discussing AI agents isn’t a new thing and it was harder for Nvidia to rise above the noise.

The Rise of China

Huawei and ZTE continue to have strong autonomous network and AI offerings. Their Chinese operator customers have been ambitious in streamlining their networks in recent years. ZTE unveiled a new Network Graph Model to eliminate operational silos. This builds on themes from previous years where ZTE showed a natural language interface generative AI tool to help network teams identify, track and resolve problems in all parts of a network even if the engineer involved was expert in only one network area.

Huawei cited metrics on improvements in complaint resolution and customer satisfaction among operator clients in Hong Kong and Saudi Arabia. As usual, it had demonstrations of advanced autonomous capabilities from its Chinese operator customers. On monetization, Huawei showed demonstrations from Africa, the Middle East and Pakistan.

Both Chinese suppliers benefit from the rather vague communication among Western suppliers in what AI models and AI hardware they’re using. This gives Huawei and ZTE enormous opportunity to continue to use their in-house telecom models, which they have been developing for many years now — longer than their Western rivals Nokia and Ericsson have been actively embracing AI. If Western companies had settled on using one of the major US-based frontier AI model solutions this could create a barrier for ZTE and Huawei, but there’s little sign that this is happening.

Most of the telecom suppliers at DTW continue to argue that they’ll use “the best AI” or “the best agents” for the task at hand. Or, even more vaguely, that they’ll use whatever AI models that their operator customer prefers. This sidesteps the issue. AI models aren’t yet interchangeable. There are significant differences in the quality of AI models and the cost of operating them. I don’t believe that it’s sustainable to maintain products based on several different AI frontier models — for manageability and cost reasons suppliers will need to standardize on one or two key AI model sources. The question is, will the winning telecom models be proprietary or, as Nvidia highlighted at GTC in March, open-weight models?

TM Forum’s AI Trust Gap

TM Forum had a particular snappy sound bite this year that’s highly relevant to the ongoing AI discussion, what it described as the “AI trust gap”. In TM Forum’s survey of communications service providers (CSPs), “72% of surveyed CSPs believe their AI deployments are trustworthy, only 14% can produce objective operational evidence to prove it”.

With the increasing usage of AI agents and generative AI, this data point hits at the heart of the challenge. It’s also why the solutions from ZTE, Huawei, Amdocs, Ericsson, Nokia and others are now focusing on how to coordinate teams of AI agents. The expectation of the industry is that we’ll see AI agents that are tightly focused on key tasks, with supervisory agents coordinating across them, and other agents checking and validating the input — to avoid malicious data injections — and the output to ensure quality.