For a few years now shift changes have been expected in the telco world, driven mainly by innovative technology being made available, fierce competition and savvier end customers wanting and expecting more for their money. However, operators have been somewhat cautious on their investments, expecting to see more evidence that would reinforce their thinking before deciding to commit to specific innovative approaches and technologies. 2025, though, is expected to be the year where such changes will start taking place, innovative technologies will be supported by more clearly defined standards and examples of implementations in the real world will increase the operators’ confidence.

A few trends and predictions can be identified:

1.1 rApps vs xApps for the RIC OpenRAN

rApps will consolidate as the preferred application type for the RIC OpenRAN architecture, since the vast majority of use cases are not needed to meet the strict time-bound requirements typical of the near real time response of the xApps.

In addition, rApps allow for easier and quicker use case implementations. This is because they use a wider variety of simpler high-level programming languages such as Python instead of C/C++ or Golang, which are more typically used for xApps.

The future rApps (but also the xApps) should be capable to implement sophisticated, high-value use cases by using Artificial Intelligence/Machine Learning (AI/ML) to add pattern forecasting, hence issuing fast close loop correcting actions in the provisioning systems of Communications Service Providers.

Finally, from a business perspective RIC-hosting providers may consolidate as needed middleman players in the OpenRAN applications arena, along with the figure of the xApp/rApp vendor specialists, as a way to reach the CSPs.

1.2 Telco APIs standardisation with CAMARA APIs

After the meta release in Autumn 2024 of 25 APIs, the CAMARA Telecom consortium, backed up by major telcos and hyperscalers along with many other members (+1100 contributors and 396 organisations), will ensure that the new developed applications operate seamlessly and securely in any network. Such APIs have the power to provide opportunities to monetise the investments in 5G, by offering powerful network capabilities to applications implemented in a wide range of use cases such as credit card fraud prevention, low latency video streaming, semi-autonomous vehicles and augmented reality.

Analysts indicate that the API market shows a potential of $100 to $300 billion in the next five to seven years. However, in the past up to two thirds of the value was lost by CSPs to other players. For example, the far more flexible WhatsApp and iMessage applications killed the expensive and outdated SMS service.

Now, CAMARA open-source project has an opportunity to change that picture by allowing developers to deliver new applications capable to exploit the unique 5G network benefits across any network. However, agreement on key APIs, timelines and coordinated efforts among telco players must happen to succeed. For example, APIs addressing location and fraud prevention may be relevant for banking or fintech and should not face regulatory issues. Also, other APIs looking at quality on demand (i.e. applicable to bandwidth-sensitive or delay-sensitive applications) may be of interest for certain customers too. 

1.3 Intent-based network automation and Large Language Models

Along with adoption of 5G-SA and the so-called paradigm shift expected in the form of new benefits, intent-based network automation and large language models (LLMs) also bring their inherent complexity with them. This is in addition to the compounded complication derived from their coexistence with older RAN technologies such as 4G or 3G. In a not too distant future, this will push many CSPs to embrace a journey towards zero-touch autonomous network, as defined by the TM Forum. Intent-based automation will be an important piece of work as early as in Level 3 ‘conditional autonomous network’ of the autonomous journey, and it will definitely be a prerequisite for those few CSPs working towards their Level 4 ‘highly autonomous network’, where human supervision is still required.

Intent-based automation refers to the fact that specific goals at the right level are formulated, leaving to the network an autonomous (and intelligent) course of action. As a consequence, the use of AI/ML technology becomes a must for intent automation. Specifically, the use of LLMs as a subset of Machine Learning (ML) is an interesting proposition, bearing in mind its considerable capacities in natural language processing as well as the significant progress made in reasoning.

However, it is also fair to say that LLMs have encountered difficulties when trying to be applied on the very technical and highly specialised world of telecom network management. Examples of difficulties could be the comprehension of full context on data manipulation and highly complex planning and decision-making, domain-specific challenges such as applying complex technical mobile standards, and hallucinations coming from apparently correct information and critical omissions, which are often difficult to detect. On the other hand, the progress made on several AI fields such as using divide-and-conquer modular network management agents or integrated general LLMs and human-expert grade models, show that those obstacles may be addressed. Further progress will be made in 2025 but we will know only later.

1.4 Predictive maintenance

Telecommunications industry is one of the most complex and rapidly growing industries where key players such as CSPs attend millions of users consuming a wide range of digital services. The complexity of the underlying technologies (multiple RANs such as 5G, 4G, 3G, CORE, Transport, …), and infrastructure that make such services possible is paramount, and ensuring the ‘promised’ service levels to the end customer means that nowadays AI-based predictive maintenance is a must in the near future, if not in the present time. Analysts have estimated that the predictive maintenance market will reach USD 9 billion with and expected growth of CAGR 17% until 2028.

AI models applied to network maintenance make possible for CSPs to evolve from traditionally expensive and labour-intensive reactive maintenance to a preventive approach, where the interventions needed are predicted and, if in close loop mode, self-healing actions are carried out. Already several operators such as AT&T, Telefonica and Vodafone have applied AI/ML models to implement predictive maintenance. The range of applications goes from identifying signs of early equipment degradation (signal degradation, abnormal temperature, etc.) to monitoring of power usage or real-time tracking of environmental factors to anticipate failure prediction.

In general, multiple ML models will coexist since the data to process and monitor comes from several sources (i.e. sensors, weather records, network PM counters, etc.), but the goal remains the same: anticipate equipment failures and reduce service downtime to achieve optimal user experience and operational expense.

The AI-based preventive maintenance is likely to evolve and standardise to support the autonomous network journey, where an intelligent course of action is taken by the network itself. The network is expected to be capable of not just identifying and reporting potential points of failure, but of also implementing a self-healing or failure mitigation action plan.