Telecom operators are increasingly redrawing their entire stack in an emerging AI-driven era, moving from legacy OSS/BSS and RAN to orchestration layers, as they shift to an agile, modular approach, writes Anna Ribeiro. Traditional monolithic systems cannot scale and move fast enough for Open RAN and AI, nor offer needed capabilities for automation and continuous intelligence. Now the focus shifts to managing the entire lifecycle of machine learning models, real-time data flows and feedback-driven learning systems. Therefore, agile telco stacks must be constructed on CI/CD, continuous training, real-time feedback loops and active model governance to satisfy the instant needs of AI-driven networks.
McKinsey discovered that around 50% of telco providers saw benefits from AI by 2025, up from just 25% the year before. The majority are boasting 10 to 15% EBITDA improvements and point to the data, models and intelligence infused into each layer as the big drivers, pushing adoption of connected, agile operating models. Deloitte expects the global BSS/OSS market to grow to be a US$70 billion opportunity by 2025, driven by adoption of cloud native platforms, API-centric models and unified service-centric architectures, highlighting the need for modern and flexible telco stacks.
Strategic case for agile modernisation
The disaggregated nature of Open RAN makes it possible for operators to mix and match components from different suppliers using open interfaces set by the O‑RAN Alliance. But modularity alone isn’t enough. Operators simply can’t iterate, integrate and optimise without an agile delivery engine of CI/CD pipelines and automation. Staying competitive, they require quick, easy-to-update software on multiple vendor technologies, lower risk of integration and rapid, reliable deployments across a multi-vendor environment.
Telcos are incorporating MLOps, which is a discipline that brings DevOps practices to AI, in their software delivery chain. It treats models, datasets and code as artifacts, with versioning, CI/CD, testing and compliance checks, for traceability, reproducibility and auditability in regulated environments. For instance, models predicting customer churn or detecting anomalies inside the RAN need to be retrained regularly with updated data to remain accurate.
Such changes force operators to rethink how they handle feedback loops. In traditional agile development, questions are answered through user testing and business metrics. In AI-centric systems, feedback is received from real-world model drift, changes in telemetry, and prediction scores.
Continuous intelligence via RIC and data streaming
At the heart of a smart telco stack is the RAN Intelligent Controller (RIC) deployed in non‑real‑time and near‑real‑time modes. These modules run rApps driven by AI/ML models to continuously optimise coverage, energy consumption, capacity and user experience.
Vendors such as Ericsson, Samsung, Mavenir and Parallel Wireless embed automation and AI logic into their OpenRAN suite of products. For example, Ericsson’s Intelligent Automation Platform provides an open software development kit to allow for rApps to be built for CSPs and third-party developers.
In one telco example, Verizon began using Samsung’s AI‑powered Energy Saving Manager on Qualcomm’s Dragonwing RAN Automation Suite. Field-test results identified energy savings of 15%, going up to 35% savings per sector during low‑traffic periods. Such operational efficiency is only achievable when the intelligence is natively built into the control layers of the network, instead of slapped on as an afterthought.
Automation backbone using CI/CD and zero‑touch provisioning
Telcos require automation across the software life-cycle to deploy continuous intelligence at scale. OpenRAN adds complexity with multi‑vendor ecosystem: For each new rApp, network function or radio unit, integration into an end‑to‑end solution must be accomplished and tested and validated on service providers’ platforms. That fuels the application of DevOps practices and CI/CD pipelines that enable frequent, reliable updates without any disruption of service.
Zero-touch provisioning is paramount to Open RAN, automating the installation and operation of various RAN elements and services, minimising manual configuration and optimisation work, and therefore cutting operational expenses,” it added, as well as allowing us to make that gradual shift from proprietary to virtualised, vendor-neutral architectures. It helps network operators to increase operational efficiency, enhance user experience and be agile to rough demand changes for operator-centric telecom applications.
Vendor momentum and early wins
Global carrier trials show strong momentum. NTT DOCOMO and NTT DATA report up to 30% reduction in total cost of ownership, halved network design workload and 50% lower power consumption through automation, virtualisation and GenAI‑enabled orchestration.
Samsung claims it leads the world in live vRAN Open RAN sites and is rapidly embedding AI‑driven RIC and private‑network automation logic into its stack. By the end of 2025, it expects to be the largest AI‑enabled Open RAN ecosystem globally.
ABI Research singles out Ericsson, Mavenir, Nokia, Parallel Wireless and Samsung as top Open RAN vendors thanks to their modular, AI‑oriented orchestration platforms and sustainable automation tools. Mavenir, for example, supports tens of thousands of radios via compliant cloud‑native vRAN and RIC installations across multiple operators.
To achieve continuous AI‑driven intelligence, telecom operators must:
- Break up monolithic stacks: Replace proprietary OSS/BSS and RAN modules with cloud‑native, modular microservices and real‑time data layers.
- Adopt CI/CD pipelines to enable rapid release, validation and rollback of network code, rApps and orchestration modules.
- Deploy RIC‑driven control loops in near‑RT and non‑RT contexts, so AI models can continuously optimise network resources based on live telemetry.
- Invest in telco DataOps stacks built on Kafka/Flink to collect, correlate and act on data across RAN, transport and core.
- Enable zero‑touch provisioning for vendor‑agnostic deployments and fast scaling.
Challenges ahead and how to mitigate
According to a 2025 report by STL Partners, nearly 60% of telecoms AI initiatives fail to move beyond proof-of-concept due to weak feedback mechanisms and disjointed data infrastructure. Agility must consider how models behave under changing traffic patterns, user behavior, or macro events like a network outage or regional spike in device usage. This demands closer teamwork between data science, network engineering and operations, with joint telemetry pipelines and platform observability to guide it.
Operators must also overcome cultural and capability hurdles. Open RAN requires coordination across multiple vendors and a DevOps mindset centered around automation, not manual patching. Skills shortages remain real, as telcos must retrain engineers in cloud, AI/ML and multi‑vendor orchestration tools.
Security is also a consideration, as disaggregated architectures expose larger attack surfaces. It must be baked into the telco stack, with stringent data governance, secure APIs and ongoing compliance checks. Open RAN standards bodies and researchers have started releasing threat models and mitigations to inform safer deployments.
Conclusion
The telco stack needs to be rebuilt on an agile, cloud-native base now, not meandering down the line. It is what is necessary to survive in the AI age. The requirements of speed, automation, and intelligence are beyond the capacity of legacy systems.
Operators need to evolve from classic DevOps to combined DevOps and MLOps with automation and feedback in real-time. Adding AI to agile transformation also requires a culture shift, aligning data scientists, engineers and operations teams around shared organisational goals and telemetry.
Anna Ribeiro