Ericsson Unveils Agentic rApp as a Service on AWS Marketplace
On February 23, 2026, Ericsson announced the launch of a new Agentic rApp as a Service available through the AWS Marketplace. The solution marries Agentic and Generative AI capabilities to accelerate the optimisation of communication networks. By providing a natural‑language interface that translates user requests into concrete optimisation actions, the platform claims to elevate network autonomy for service providers. Ericsson already reports that the solution is undergoing field‑testing with major operators, including Vivo Brazil, and that the deployment is designed to reduce operational complexity and improve scalability.
The Technical Fabric of the New Offering
At its core, the Agentic rApp leverages a two‑tier architecture:
- Agentic Core – An AI planner that interprets high‑level requests (e.g., “increase throughput in Zone 12”) and translates them into a series of low‑level network configuration commands. It utilizes reinforcement learning to iteratively refine its policies based on real‑time telemetry.
- Generative Interface – A natural‑language processing engine trained on a corpus of network operation manuals and operator‑generated queries. It converts conversational input into formal JSON payloads that the Agentic Core can consume.
The integration with AWS Marketplace offers a familiar consumption model for operators, allowing them to spin up the rApp on demand and pay per usage. However, this also raises questions about data residency and compliance, especially for operators in jurisdictions with strict data‑protection laws.
Field‑Testing Insights: The Vivo Brazil Case
Vivo Brazil, one of the world’s largest mobile operators, is reportedly running a pilot that focuses on ultra‑dense urban environments. Preliminary results suggest a 12 % reduction in hand‑over errors and a 5 % improvement in packet delivery latency during peak hours. While the numbers are promising, the pilot also exposed latency bottlenecks when the rApp had to fetch large sets of network state from on‑premises controllers, hinting at the need for edge‑computing enhancements.
Beyond Software: Ericsson’s Silicon‑Level AI Push
The previous day, Ericsson highlighted a broader shift in its core business. The company is embedding neural‑network accelerators into its own silicon, collaborating with major technology firms such as NVIDIA and Intel. Coupled with new radio modules and AI‑optimised antennas, the goal is to:
- Enhance uplink performance by enabling on‑device inference for real‑time interference mitigation.
- Lower operating costs through more efficient radio‑resource management.
- Reduce energy consumption by allowing the network to adapt dynamically to traffic patterns.
A prototype chipset, announced at a closed‑door session, integrates a 16‑core GPU‑accelerated inference engine capable of running models with up to 1 B parameters at 5 Gbps throughput. Yet, the practical deployment of such high‑performance silicon in the field raises concerns about supply chain resilience and the environmental impact of semiconductor manufacturing.
Market Implications and Competitive Landscape
Ericsson’s dual strategy—offering AI‑enabled software services while deepening hardware integration—positions the company against key competitors:
- Huawei: Already offers AI‑driven network optimisation in its Kunpeng line, albeit under the shadow of export restrictions in several markets.
- Qualcomm: Focuses on edge AI for 5G, but its ecosystem is less mature for large‑scale network management.
- Nokia: Provides AI‑augmented OSS, yet its hardware portfolio is comparatively limited.
By aligning with AWS, Ericsson taps into a massive developer community, potentially accelerating adoption. However, the reliance on a public cloud provider also exposes operators to vendor lock‑in and cost‑predictability challenges.
Societal, Privacy, and Security Considerations
The introduction of generative AI into network management raises several non‑technical issues:
- Privacy: Natural‑language interfaces may inadvertently capture sensitive operator data. Ensuring that the AI’s training data and inference processes comply with GDPR, CCPA, and other privacy regimes is imperative.
- Security: An AI that autonomously reconfigures network elements can become a target for manipulation. Robust adversarial‑training pipelines and rigorous threat modelling are needed to safeguard against malicious exploitation.
- Equity: If AI‑driven optimisations favor high‑traffic urban zones, rural or underserved areas may experience service degradation. Transparent governance frameworks should be established to balance efficiency with universal coverage obligations.
Looking Ahead: Mobile World Congress and Enterprise Adoption
Ericsson plans to showcase the commercial potential of these technologies at the upcoming Mobile World Congress (MWC) later this year. The event will likely feature live demos of the Agentic rApp and the silicon‑accelerated radio modules. Success at MWC could catalyse broader adoption among enterprise customers who are increasingly demanding network agility to support IoT, autonomous vehicles, and edge analytics.
Conclusion
Ericsson’s expansion into AI‑enabled services and silicon innovation reflects a strategic pivot toward network autonomy. While the technical benefits—reduced hand‑over errors, improved latency, and energy savings—are tangible, the broader implications for privacy, security, and societal equity cannot be overlooked. The next few months will be decisive: will the industry embrace these advances responsibly, or will unforeseen risks undermine their promise? The forthcoming MWC and the outcomes of ongoing pilots will provide critical evidence to answer that question.
