Generative AI in Telecom: Cutting OPEX, Boosting ARPU, and Powering Zero-Touch 5G

John Gieske

Director of Technology

Telecom companies are under pressure. Costs are rising, networks are more complex. Customers even want answers right now. Earlier chatbots often frustrated customers. Modern Gen AI LLMs now resolve up to 80 % of tier-1 inquiries without escalating. Generative AI in telecom is already helping telcos trim costs, improve service, and rethink how they operate.

Vodafone’s chatbot TOBi has already taken on over 13 million chats, with 65-80% fully automated. Telefónica’s marketing with GenAI increased sales by 20%, and even customer conversations jumped by 30% for personalized campaigns. Unlike GPU-hungry solutions from hyperscalers, Nybsys delivers Gen AI inference on Intel Xeon at the network edge.

One big European player cut service problem times almost in half (40%) after using AI tools. Customers said things felt much smoother ¹. Meanwhile, nearly 90% of telecom companies already use AI, with 48% piloting and 41% actively deploying it ².

Most of the telecoms are testing or running AI today. Nearly every telecom company is in the game. The market is catching on. The global generative AI in telecom was valued at around $302 million in 2024 worldwide. It is projected to reach nearly $9.8 billion by 2034, growing at a compound annual growth rate of 41.59% ¹.

Still, telcos are investing less than other industries. In 2023, they spent just 0.16% of revenue on AI and GenAI, compared to a global average of 0.49%. On a broader scale, global GenAI spending is expected to reach $644 billion this year, up to 76.4% from 2024 ³.

	2024 Spending	2024 Growth (%)	2025 Spending	2025 Growth (%)
Services	10,569	177.0	27,760	162.6
Software	19,164	255.1	37,157	93.9
Devices	199,595	845.5	398,323	99.5
Servers	135,636	154.7	180,620	33.1
Overall GenAI	364,964	336.7	643,860	76.4

Table 01: Worldwide GenAI Spending Forecast (Millions of U.S. Dollars). Gartner (March 2025).

Let’s look closer at how GenAI changes daily work at telcos. This covers everything from the customer’s first chat to the nuts and bolts behind the service. We’ll sort what’s working and spotlight what to watch before your business decides where to dive in.

Table of Contents

Generative AI is Reshaping Telco Operations

GenAI is no longer experimental in telecom. It’s being used across multiple functions, and the results are measurable. According to McKinsey, 65% of organizations now use GenAI in at least one business function, nearly double from the previous year ⁴. The most common areas are marketing and sales, product development, and IT—functions where GenAI delivers the most value.

1. Smarter Customer Support

Vodafone’s TOBi chatbot has handled over 13.5 million conversations, 65 to 80% fully automated ⁵. European telcos report a 40% drop in resolution time and a 35% improvement in customer effort scores ². GenAI enables real-time, context-aware support that scales without increasing headcount.

2. Personalized Sales and Marketing

Telefónica saw a 20% increase in sales and 30% conversion rates using GenAI for personalized campaigns ⁵. McKinsey confirms marketing and sales as the top value-generating functions for GenAI, with many companies reporting direct revenue growth ⁴.

3. Predictive Network Maintenance

GenAI helps detect faults before they happen. Maintenance costs drop by 25% to 30%, and outages decrease by 10% ⁵. Capgemini reports that GenAI tools help technicians access incident histories instantly, speeding up diagnostics ⁶.

4. Automated Network Design and Optimization

A U.S. telecom operator saw a 30% boost in network efficiency using AI-driven traffic management ⁵. These systems also help reduce energy consumption by up to 15% through smarter resource allocation ⁵.

5. Fraud Detection and Risk Management

AI-driven fraud detection systems now reach 88-92% accuracy ⁷. One major telco saved over €50 million in potential losses. GenAI can analyze transaction patterns, flag anomalies, and generate compliance-ready reports.

6. Internal Task Automation and Developer Productivity

GenAI can automate up to 70% of repetitive tasks in telecom operations ⁷. Developers using GenAI tools complete tasks twice as fast. McKinsey’s 2024 report shows that IT is now one of the top three functions where GenAI is actively deployed.

7. Field Operations and Remote Support

Technicians in the field often need quick answers. GenAI-powered assistants guide them through diagnostics and repairs. Capgemini reports that conversational bots help troubleshoot issues on-site, reducing repeat visits ⁶. Some operators are combining these tools with AR for real-time visual support.

Essential GenAI Considerations Across the Three Layers

Generative AI doesn’t work in isolation. For telcos to get real value, they need to think in layers. IBM breaks this down into three: data, model, and application. Each one matters. If one fails, the whole system underperforms.

1. Data Layer

This is the foundation. Without clean, telecom-specific data, GenAI models hallucinate or miss the mark. Telcos deal with massive volumes of structured and unstructured data—network logs, call records, CRM entries, etc. IBM emphasizes the need for curated, labeled datasets that reflect real telecom operations ². McKinsey’s 2024 report also highlights that data quality and governance are among the top barriers to scaling GenAI ⁴.

2. Model Layer

This is where intelligence lives. Telcos must choose between open-source, proprietary, or fine-tuned foundation models. IBM’s watsonx.ai platform supports telecom-specific fine-tuning, which is critical for accuracy and relevance ². McKinsey notes that 55% of high-performing AI companies now use custom or fine-tuned models rather than generic ones ⁴. These models must be evaluated for accuracy, explainability, latency, and cost.

3. Application Layer

This is what users see. It includes chatbots, dashboards, and internal tools. IBM stresses integrating GenAI into existing workflows, not building standalone tools ². Capgemini adds that conversational interfaces are now embedded into field service apps, customer portals, and network management consoles. The goal is to make GenAI invisible but impactful, working behind the scenes to improve speed, accuracy, and decision-making.

Workflow for Retrieval-Augmented Generation (RAG) with Telecom-Adapted Models

Generative AI models are powerful, but they have limits. They can’t access real-time data or proprietary telecom knowledge unless you give them a way in. That’s where Retrieval-Augmented Generation (RAG) comes in. It connects GenAI models to external knowledge bases—like network logs, manuals, and customer records—to generate more accurate, relevant answers.

RAG works in two phases: retrieval and generation.

First, the system converts telecom documents into vector embeddings. These mathematical representations make it easy to search for meaning, not just keywords ⁸. When a user asks a question, the system retrieves the most relevant chunks of data from the vector database. Then, it passes that context to the GenAI model, which uses it to generate a response. This process reduces hallucinations and improves factual accuracy ⁹.

IBM’s Watsonx platform uses RAG to avoid costly retraining. Instead of fine-tuning a model every time something changes, you update the knowledge base. That means faster deployment and lower computing costs ⁹. McKinsey adds that RAG is instrumental in telecom for customer service, enterprise knowledge management, and drafting assistants ⁸. It allows chatbots to pull from internal policies, account data, and service histories—giving users specific and trustworthy answers.

RAG also supports citations. When a GenAI model answers, it can show where the information came from. That builds trust and makes it easier for users to verify the response ¹⁰. In telecom, this is critical. Whether it’s a technician asking about a network fault or a customer checking billing policies, accuracy matters.

The Crucial Step: Evaluating Generative AI for Reliability

Deploying GenAI is one thing. Trusting it is another. Telecom operations rely on accuracy, speed, and consistency. If a model gives wrong answers or fails to explain its logic, it can cause real damage—lost revenue, frustrated customers, and compliance risks. Generic models don’t understand telecom jargon or workflows. Ericsson points out that telecom-specific evaluation is essential. You need models trained on domain data and tested against real use cases like fault isolation, ticket triage, and configuration support ¹¹. IBM adds that evaluation must go beyond simple metrics. Accuracy alone isn’t enough. You must also measure relevance, coherence, latency, and safety ¹². McKinsey’s 2024 report shows that only 23% of companies have fully defined GenAI risk protocols ⁴. That’s a problem. Telcos need structured evaluation pipelines. Ericsson recommends using multiple metrics and even “oracle models” to benchmark GenAI outputs ¹¹. Think of it like testing a new car. You don’t just check speed. You look at safety, fuel efficiency, and comfort. Human-in-the-loop systems are also key. GenAI should assist, not replace, experts in high-stakes decisions. Whether it’s network troubleshooting or billing disputes, final judgment should come from trained staff. That’s how you keep GenAI reliable, accountable, and valuable.

Challenges and the Path Forward

Generative AI is promising. However, telecom companies face real challenges when trying to scale GenAI. Some are technical. Others are organizational. Ignoring them leads to wasted investment and poor results.

Data quality is the first hurdle. GenAI needs clean, labeled, and domain-specific data. Most telcos still struggle with fragmented systems and inconsistent formats. McKinsey’s 2024 report shows that data governance is one of the top barriers to GenAI adoption ⁴.

Legacy infrastructure is another issue. Many telcos still run on outdated systems that don’t integrate easily with modern AI tools. IBM notes that successful GenAI deployment requires cloud-native platforms and API-ready environments ².

Talent gaps are slowing progress. According to BCG’s TeBIT 2024 report, only a small fraction of telcos have in-house GenAI expertise. Upskilling is essential for data scientists, network engineers, product teams, and customer service leads.

Cost and compliance are also concerns. GenAI models require significant computing resources. And telecoms operate in highly regulated environments. Gartner estimates global GenAI spending will reach $644 billion in 2025 but warns that ROI depends on careful planning ³.

So, what’s the path forward? Start small. Focus on high-impact, well-defined use cases. Invest in foundational data architecture. Upskill your teams. Build partnerships with AI vendors and cloud providers. And most importantly, shift the mindset—from being a connectivity provider to becoming a tech-driven service company.

Strategic Value Creation with GenAI – Why Generative AI in Telecom

Generative AI isn’t just a tool for automation. It’s a lever for growth. When used strategically, it can reshape how telecom companies create value—across revenue, cost, and customer experience.

Start with profitability. McKinsey reports that GenAI can reduce customer support costs by up to 40% and improve telecom source sales conversion rates by 30% ⁷. AI-powered personalization, dynamic pricing, and proactive service recommendations contribute to higher ARPU (average revenue per user).

Then there’s speed. GenAI shortens time-to-market for new services. IBM notes that telcos can use GenAI to generate product documentation, marketing content, and even code for internal tools ². That means faster launches and quicker feedback loops.

It also opens the door to new business models. Some telcos are exploring AI-as-a-Service, offering GenAI-powered tools to enterprise clients. Others are embedding GenAI into B2B platforms for logistics, energy, and smart cities. According to BCG, telcos that adopt AI-first strategies are more likely to diversify revenue streams and move beyond connectivity ¹.

GenAI helps telcos compete in a market where price wars occur daily, and differentiation matters. A telco that offers smarter support, faster service, and personalized experiences stands out. That’s good for customers and long-term value.

Nybsys & Generative AI – Combined for Telecom

Nybsys is positioning itself as a serious player in the GenAI-powered telecom space. At MWC 2025, the company showcased a suite of AI-driven solutions built specifically for telecom operators, focusing on automation, diagnostics, and edge performance ¹³.

One of the standout innovations is Nybsys’s GenAI-powered network operations platform. It enables zero-touch deployment and real-time diagnostics using AI-powered visual recognition. Technicians can simply snap a photo of any equipment, and the system instantly returns specifications, fault insights, and troubleshooting steps. This reduces downtime, speeds up field operations, and lowers support costs ¹³.

Nybsys also introduced Sentra Gen-AI, a GPU-free edge AI solution optimized for Intel Xeon and Dell XR8000 platforms. It’s designed for telecom and industrial environments where power efficiency and compact deployment matter. Sentra supports containerized GenAI workloads at the edge, making it ideal for smart cities, logistics, and public safety networks ¹⁴.

In partnership with Lanner and Advantech, Nybsys launched Private 5G Express (P5GE)—a plug-and-play 5G-in-a-box solution. It supports GenAI workloads, boots in under five minutes, and is optimized for low power and high performance. This makes it a strong fit for telcos deploying private networks in manufacturing, warehousing, and campus environments ¹⁵.

Nybsys’s approach is straightforward: combine GenAI with edge-native infrastructure to help telcos scale faster, automate smarter, and reduce operational complexity. It’s about AI and making networks more intelligent, responsive, and cost-effective.

Competitive Positioning: Nybsys vs. Nokia & Ericsson

Nybsys is carving out a distinct advantage in the race to deploy generative AI across telecom operations. They have kept focus on edge-native, low-power, fast-deployment infrastructure—an area where legacy players like Nokia and Ericsson still rely on heavier, cloud-centric setups.

Edge-Native AI vs. Legacy GenAI Architectures

Nybsys delivers Sentra Gen-AI and Private 5G Express (P5GE) solutions that natively execute generative AI workloads right at the edge—on compact appliances like Lanner’s ECA6050B/6051, with no dependency on bulky cloud GPU farms¹⁶.

Nokia and Ericsson primarily adopt centralized AI models—training in the cloud before distribution—introducing higher latency, bandwidth costs, and greater power consumption.

Speed & Simplicity: “Boot in 5 Minutes”

Nybsys’ P5GE is a fully containerized, plug-and-play private 5G network-in-a-box that can boot in under five minutes. It enables lightning-fast deployments in industrial, retail, or campus environments ¹⁷.

Meanwhile, Ericsson and Nokia build intricate cloud-native ecosystems. These ecosystems often involve multi-hour initialization, third-party clouds, and complex orchestration which lengthens time-to-value.

Power & Cost Efficiency

Nybsys patents its edge GenAI to run on low-power Intel/Xeon platforms (e.g., Dell XR8000, Lanner ECA‑7230) with < 180 W per node. They aim to deliver significant OPEX savings through deployments ¹³. Alternatively, the infrastructure used by Nokia and Ericsson is typically heavier—consuming 350–450 W—and incurs higher licensing and integration costs.

Open, Containerized Ecosystem

Nybsys builds systems on Kubernetes and Intel OpenVINO to support containerized multi-workloads and open O-RAN integrations via SMO and xApp frameworks ¹⁷. This empowers customization and avoids vendor lock-in.

Nokia and Ericsson offer more monolithic, vertically integrated stacks. These stacks can inhibit flexibility and slow down adaptation to new telco APIs and workflows.

Tangible ROI & Industry Tailoring

Nybsys targets clearly measurable operational KPIs: image-based fault diagnostics that reduce mean-time-to-repair by > 40%, GenAI–powered voice assistants for field & NOC personnel, and plug-and-play private 5G for manufacturing and logistics.

Nokia and Ericsson promote GenAI as part of broader digital transformation suites, but with fewer documented ROI wins at scale in edge-heavy verticals.

Lightweight Security Integration

Nybsys includes its Aegis Security Gateway, capable of handling carrier-grade IPsec at up to 200 Gbps, delivered inline with its RAN and cloud modules. Nokia and Ericsson rely on layered security offerings, which may still require separate component installs and third-party partners, adding both cost and rollout complexity.

Dimension	Nybsys	Nokia / Ericsson
AI Execution	On-site GenAI on compact, GPU or GPU-free edge systems	Cloud-first; edge inference often secondary
Deployment Speed	“Box-to-beam” in under 5 minutes	Multi-hour setup tied to cloud orchestration
Power Efficiency	<180 W per node with Xeon-based appliances	350–450 W+ GPU-focused edge servers
Architecture	Kubernetes + OpenVINO, containerized, O RAN friendly	Monolithic, vertically integrated stacks
ROI Focus	Proof-backed benefits (40% faster MTTR, voice assistants, private 5G)	Broad AI platform messaging with fewer edge- ROI case studies
Security Approach	Built-in Aegis gateway with inline carrier-grade IPsec	Modular, across layers with potential vendor and integration bloat

Table 2 Competitive Positioning: Nybsys vs. Nokia & Ericsson

Frequently Asked Questions

1. What is Generative AI in telecom?

Generative AI in telecom refers to AI models that can generate content—like text, code, or responses—based on telecom-specific data. It automates customer support, optimizes networks, and improves internal operations.

2. How is GenAI different from traditional AI used in telecom?

Traditional AI focuses on pattern recognition and automation. GenAI generates new outputs, such as personalized messages, troubleshooting guides, or network configurations, often in natural language.

3. Which companies are leading GenAI innovation in telecom?

Vodafone, Telefónica, IBM, Ericsson, and Nybsys are actively deploying GenAI solutions. Many are seeing measurable improvements in efficiency, customer satisfaction, and revenue.

4. How does generative AI transform telecom operations?

Gen AI transforms telecom operations by introducing intelligent automation across key functions. It enhances customer support, predicts faults before they occur, assists in designing and optimizing networks, and boosts productivity by automating repetitive tasks.

5. How do you choose the right LLM for telecom tasks?

The most effective LLMs are those that have been fine-tuned on industry-specific data. They should understand the terminology, workflows, and context unique to telecom environments. It’s essential to evaluate models based on their accuracy, response time, explainability, and integration ability with existing systems.

Conclusion: Generative AI in Telecom

Generative AI is becoming a core part of telecom strategy. It’s improving customer support, streamlining network operations, and helping teams work faster. The results are already visible in reduced costs, better service metrics, and faster product rollouts.

Telcos that succeed with GenAI are doing more than experimenting. They invest in clean data, choose the right models, and integrate AI into everyday workflows. They’re also building teams that understand telecom and AI, which is key to scaling results.

The technology is moving fast. Spending is rising, competition is increasing, and expectations are growing. Telecom companies that act now will be better positioned to lead in efficiency, innovation, and customer experience.

This shift is not optional. It’s a strategic decision. The next phase of telecom growth will be shaped by how well companies use GenAI to solve real problems and deliver measurable outcomes.

Ready to pilot Gen AI at the network edge? Book a 30-minute workshop with Nybsys solution architects or request a Private 5G Express demo kit.