Recent headlines attest to a flurry of grand AI investments: Meta is pledging US$65 billion for AI infrastructure in 2025; AWS is committing over $75 billion; Microsoft is investing $80 billion; and a there is the monumental US$500 billion plan by President Trump, backed by OpenAI, Softbank, Oracle and the UAE. Even smaller-scale efforts, like the EU’s €20 billion AI Gigafactories initiative, reflect significant global ambition. These headline-grabbing investments demonstrate hyperscalers’ near zealous belief in the inevitability of AI.
Dell’Oro estimates that global data centre capex surged 51% to $455 billion in the past year and forecasts another 30% rise this year. McKinsey projects a staggering US$5 trillion in capital expenditures for AI data centres by 2030, surpassing the GDP of all but two nations. IDC predicts these AI investments could cumulatively add US$22 trillion to global GDP by 2030, reflecting the transformative economic potential of AI technologies.
By any measure, the numbers in AI are vast, almost otherworldly.
In stark contrast, telcos face persistent financial pressures. Global telecoms service revenues are growing at only 2.9% CAGR, barely keeping pace with inflation. Worldwide telecoms capex dropped 8% in 2024, with a forecast annual decline of 2% over the next three years. Telcos, heavily invested in 5G and fibre, urgently need returns on existing investments. Yet, the allure of potential revenues from AI-powered services like virtual assistants, personalised media, IoT, AR/VR, analytics and cybersecurity is compelling and perhaps worthy of more investment.
Analyst Omdia projects AI traffic will surpass non-AI traffic on Europe’s cellular networks by 2031, driven primarily by AI-powered cameras, sensors and emerging XR media. Current 5G networks, though underutilised today, face the risk of becoming overwhelmed as AI workloads scale rapidly in the coming years. The industry is at an inflection point where timely investment decisions can determine long-term market positions.
“Telcos have a huge potential opportunity to carry and process distributed AI workloads, but they need to be strategic in how they prioritise their network investments because it is impossible to predict how these workloads will spread out across the network,” says David Martin, a senior analyst at STL Partners “AI is a monster of an opportunity – but it will take human intelligence and judgment if telcos are to capture it and not be consumed by it.”
Philip Ottley, the managing partner of Telco & Infrastructure at HTEC, an AI software developer working with telcos, sees greater sophistication emerging. “The investment case is evolving from merely offering faster speeds to enabling next-generation services. Of course, this investment needs to happen against the backdrop of investors thinking they were finally coming to the end of a major capital investment cycle. The cycle looks to be starting again before fully achieving envisaged returns from previous cycles.”
Strategic areas for telco investment
Telcos face critical decisions in digital infrastructure investment across several key areas: cloud interconnects, first-mile and last-mile networks, intelligent traffic management and ramping up 5G investment. The question is, should they be speeding up deployment, investment and roll-out, or stick with their current plans?
1) Cloud interconnects and optical transport
Hyperscalers, colocation companies and telcos have announced more than 2,600 new data centres, a quarter of which will be in cities with no data centres today. By the 2030s, this could exceed 11,000 data centres worldwide. As well as owning, running and offering GPUs in some of these data centres, the US$30-US$50 billion opportunity for telcos will be in laying the fibre and interconnecting these data centres with sufficient capacity to avoid bottlenecks and reduce latency. It will require expanding and densifying fibre in metro areas and upgrading optical interconnects and transport to 400G/800G.
2) Access networks
First mile or access networks are not currently experiencing the strain of GenAI, but they will. GenAI-enabled applications using video and image recognition can result in significantly increased uplink traffic, while applications with inference workloads may require roundtrip delays of under 50ms. This will be most acutely felt in enterprise and cellular networks.
According to Zach Bennet, a principal architect at LoopUp, a cloud voice service provider: “GenAI alters network traffic significantly in terms of symmetrical traffic, upstream demand and peak usage time. Traditionally, networks have been downstream heavy, but with GenAI, the upstream demand will grow. This is due to large prompts, voice, video and all other data that must be sent to models for processing. Real-time input will also increase upstream demand.”
3) Utilise the edge
Edge deployments are essential for GenAI use cases requiring real-time processing, like AR/VR or AI video chat. STL Partners says the edge AI opportunity could be worth US$157 billion by 2030, with telcos in a good position to capitalise on it.
According to Renuka Nadkarni, chief product officer at Aryaka Networks, a SASE service provider, a priority should be “moving compute and AI inference closer to users using distributed edge nodes to reduce latency and offload core network traffic.”
Inference is key. Analyst Telegeography says that while AI training can be computed in distant data centres, inference tasks demand low-latency compute close to end users, ideally in network-dense locations. Telcos investing strategically in edge infrastructure can deliver superior user experiences and create differentiated service offerings.
4) Intelligent traffic management
Managing capacity effectively, and therefore user experience, depends on managing peaks and troughs. Due to productivity-embedded GenAI tools, peak hours might shift from evening streaming to daytime business hours.
According to Santiago Bouzas, Director of Product Management at Enea, a traffic management vendor, “sharp peaks are evolving into longer sustained periods of high demand, sometimes lasting six to eight hours. AI can help networks understand and manage these extended traffic patterns effectively … and prioritise accurate traffic classification and measurement to maintain network performance.”
Indeed, there may be no predictable peaks: LoopUp’s Bennet says: “With AI agents there are often no in or out of hours so usage could be consistent and continuous throughout the day and night.”
Using AI to manage AI will be crucial. Tools for predictive capacity planning will involve predicted peak usage, site-by-site utilisation, inventory management and modelling capex ROI. Collectively, they will help telcos make investments where and when necessary and avoid overcapacity.
5) Speed up 5G SA rollout
The deployment of 5G standlone (SA) has been slower than expected due to the huge upfront costs and technical challenges . Many operators blamed the 5G specification themselves, citing overwhelming complexity and excessive optionality in 5G standards, specifically the architecture options of non-standalone/standalone options.
According to the GSA, 64 operators in 35 countries had launched 5G SA networks in specific locations by the end of 2024, with another 89 operators planning to by 2026. China and India lead, with Europe lagging far behind. 5G SA is essential to support greater use of GenAI applications: time-division duplexing (TDD) and massive MIMO antennas allow for more bandwidth and power to be directed to uplinking; the cloud-native core reduces latency, and the specific use of URLLC reduces it further still (under 5ms). Combined with the ability to offer network slicing for specific use cases and 5G private networks, accelerating 5G SA will be essential for supporting widespread use of GenAI in business and consumer applications.
Monetising through new services and SLAs
AI workloads demand sophisticated service-level agreements (SLAs). HTEC’s Ottley suggests telcos must evolve from simple bandwidth metrics to tiered latency SLAs, burst bandwidth allocations, network slicing and local edge fallbacks to ensure service continuity.
These enhanced SLAs enable telcos to create premium service categories, targeting high-margin sectors like financial services, healthcare and logistics. “Monetisation models may need to evolve from flat rate plans to tiered, offerings or usage-based pricing based on application intensity, creating new revenue opportunities,” adds Ottley.
Cautious investment criteria
A TelecomTV survey found that just over half of telcos were confident that AI offers major new revenue-generating opportunities. The other half were split by being unsure (28%) or doubting whether AI could add new revenues (20%). So given that the extensive capital expenditures of recent years are taking time to recoup, telcos will be forgiven for treading cautiously with investment in digital infrastructure to support AI, despite the widespread boosterism.
Omdia believes that there is generally enough fibre, cell density and spectrum available to support AI growth in the short term, but widespread availability of AI-enabled “future media and XR devices” by the end of the decade could take 5G networks to their limit, but 6G will arrive just in time. Just in time for another hype cycle.
Stewart Baines