AI and Cloud: How operators are powering the networks

Generative AI (genAI) has moved beyond novelty to become a platform shift, reshaping devices, networks and enterprise strategies. Millions of consumers already rely on AI assistants, smart search and generative tools daily. Telecoms operators can benefit from the expanding AI value chain if they move beyond just providing connectivity. By offering GPU-as-a-Service, supporting AI inference or hosting sovereign AI clouds, operators can access new revenue streams and better use AI-based services.

Cloud and the AI shift

The advancement of AI is closely linked to the infrastructure supporting it. Cloud platforms, once primarily focused on storage and connectivity, are essential in meeting the massive compute and data requirements needed to train and run AI models.

Telecoms networks, long regarded as essential for enabling digital services, are becoming increasingly integrated with cloud infrastructure. In recent years, operator cloud strategies have concentrated on expanding connectivity by delivering secure and reliable links between enterprises and hyperscalers.

Traditionally, the cloud was viewed as an extension of network ability rather than a distinct layer of competition. However, AI has significantly altered the dynamics, particularly as data traffic continues to grow, with a potential inflection from usage of direct AI (think ChatGPT and co) and indirect AI (people consuming more content because of it).

AI workload placement depends on context and operational environment, with on-device, network edge and cloud each fulfilling distinct roles. Operators managing this distributed infrastructure are well-positioned to introduce advanced AI capabilities.

GSMA Intelligence notes while leading operators were early adopters of these technologies, the broader industry is now actively exploring the implementation of AI-enabled networks.

Edge and inference: real-time intelligence

AI runs in two main phases: training and inference. Training, typically housed in large data centres, develops and refines models using massive datasets. Inference, by contrast, is the process of applying these models in real time – making predictions, classifications or decisions. Inference is central to applying AI in operational contexts, it is where networks and services intersect with AI.

For operators, inference is applied to improve mobile and fibre networks as well as to enable new services based on AI technologies. This is where edge computing comes in. Edge computing supports this process; edge nodes, such as those found in metropolitan data centres, on-premises locations or within networks, allow workloads to be processed with reduced latency, lower backhaul costs and enhanced data sovereignty.

For example, consider a retailer deploying thousands of security cameras on-site. Running advanced analytics in a central cloud would be costly, latency-prone and potentially non-compliant. Network edge inference solves all three challenges: it brings compute closer to the source, keeps sensitive data within the area and lowers costs.

GSMA Intelligence estimates keeping 30% of traffic at the edge can cut energy use by around 20%, with further savings amplified by AI workloads. At the same time, AI could add 30-50% to data centre energy demand by 2030, making edge optimisation crucial for sustainability.

Edge AI also delivers deterministic performance for mission critical applications and creates monetisation opportunities for operators. Distributed inference, once a supporting element of 5G, is now central. AI strengthens the business case for edge by adding speed, efficiency, sovereignty and reduced total cost of ownership.

The interplay between AI training and inference along the network

Note: core generally refers to public cloud data centres, while edge refers to compute nodes that move progressively further out from the core.

Source: GSMA Intelligence

On-device AI: intelligence at the consumer edge

Not every workload requires large-scale infrastructure. Innovations in small language models (SLMs) are enabling AI to run directly on devices. At Nvidia’s GTC 2025, Aible demonstrated a fine-tuned Llama 3.1 model with just 8 billion parameters, which was trained for $5, outperforming larger LLMs on targeted tasks. This proves SLMs are often better suited for real time, on-device AI.

For consumers, this translates into private, latency-free AI on smartphones, wearables or IoT devices. In agriculture, on-device AI supports crop monitoring and livestock health detection in low-connectivity areas. In manufacturing, predictive maintenance can run locally without relying on the cloud. Processing locally also accelerates inference compared to cloud-based alternatives.

Yet device-level AI has limits. Enterprises deploying AI for logistics, finance or video analytics require far more compute. Here, network edge and sovereign AI clouds provide the middle ground which are powerful enough for advanced models, yet close enough to users to reduce latency and meet data residency rules. By orchestrating AI across device, edge and cloud, telcos can deliver both responsive consumer experiences and scalable enterprise solutions.

AI’s revenue moment

Edge compute underpins a growing set of services which are now mainstream. These services are regularly used by individuals, businesses and communities, reflecting their established presence in society. For instance, smart city solutions like real-time traffic management and public safety monitoring are integrated into urban infrastructure. Immersive media such as augmented and virtual reality experiences are now common in entertainment and gaming.

AI adoption across consumer and enterprise markets will accelerate mobile data growth even further, with a profound impact on the networks. Even before AI, GSMA Intelligence projected public cloud and enterprise edge would generate a net rise of 700 exabytes of enterprise cellular traffic between 2024 and 2030.

Meeting this surge requires compute closer to customers (on-premises or at the device edge) and networks that are AI-ready.

For operators, the AI revenue moment lies not in replicating hyperscalers but in orchestrating this distributed ecosystem. By enabling inference across device, edge and cloud, telcos can convert AI demand into sustainable revenue streams and redefine their role in the digital economy.