For decades, telecom operators have sought growth in service layers beyond connectivity. Nokia and NVIDIA are now offering another opportunity through AI-RAN. But has anything fundamentally changed in the operators’ business environment? It is worth asking whether AI-RAN represents the industry’s next great growth opportunity—or simply a new technology supporting an old belief. Is AI-RAN and the Telco Industry’s Never-Ending Promise of Growth?
The Same Growth Story, Again
In his article *Will Nokia’s AI-RAN Strategy Work?*, Tom Nolle highlights an interesting finding: 49 of the 64 operators he interviewed still believe that long-term growth will come from higher-value services. That is a striking number. Operators have pursued the same ambition throughout the 3G, 4G, and 5G eras. Each new generation of mobile technology has promised to move beyond connectivity into applications, digital services, and new business models. Meanwhile, the market has largely moved in the opposite direction.
Traditional voice and SMS have lost much of their importance as growth drivers, while OTT services have captured a significant share of the customer relationship in communications. Cloud providers, software companies and industrial OT vendors have built the digital platforms and application ecosystems that enterprises increasingly depend on.
Operators still have an extremely important role: providing reliable, high-performance connectivity. But increasingly, that is also where their role in the customer’s value chain ends. Why, then, do operators still believe the next generation of technology will change this?
Nokia and NVIDIA Offer a New Answer
The AI-RAN collaboration between Nokia and NVIDIA is technically interesting. AI is being integrated into the 5G radio signal-processing chain, including channel estimation, beamforming, multi-user MIMO pairing, and radio resource scheduling. In preliminary laboratory tests conducted by Nokia and Deutsche Telekom, AI-based channel estimation reportedly improved typical uplink throughput by 10–15%. These are promising results. Radio networks operate in constantly changing environments. User location, mobility, interference, and many other factors affect signal propagation. AI can plausibly improve radio resource utilization and overall network efficiency.
But this leads to two very different business opportunities. The first is improving radio network efficiency: delivering more capacity using the same spectrum, improving user experience, or reducing total costs. The second is far more ambitious. The idea is to turn GPU-based radio infrastructure into a distributed computing platform, enabling operators to offer new AI and edge services for industrial applications, robotics, and other Physical AI use cases.
This is where technological progress once again becomes a familiar telecom growth story. Operators are not simply being sold a more efficient radio network. They are also being sold the opportunity to finally capture value from the application layer.
Why Would an Enterprise Choose an Operator’s Platform?
Earlier in my career, I led the enterprise business of a telecom operator. Even then, I recognized a problem that, in my view, has not disappeared over the years. Large enterprise customers generally don’t want to build business-critical systems around a local operator’s technology, geographic footprint, or operating model when a globally scalable, operator-independent platform is available.
This is not about operators lacking technical expertise. It is about the customer’s business. An internationally expanding company needs a consistent architecture that works across new markets without rebuilding its application platforms, integrations, or operating models around each local operator’s requirements. For an enterprise, the network is essential infrastructure. It should enable business operations, not dictate their technological direction.
Later, as a consultant, I saw the same issue from the customer side. Enterprise network strategies increasingly aim to reduce dependence on individual operators. SD-WAN, cloud services and modern security architectures make it possible to build services across multiple connectivity providers. Enterprises buy connectivity from the operator, while keeping application management, identity, security, and service operations consistent and as independent of the connectivity provider as possible. Why would an enterprise make the opposite strategic choice for AI services?
The Telco Operating Model Is Not the Same as an IT Services Business
The traditional telecom business model has been built around subscriptions, average revenue per user (ARPU), service provisioning, contracts, and billing. This model has been effective and efficient for delivering connectivity services. But enterprise applications and platforms create value through different capabilities: integrations, open APIs, developer ecosystems, global scalability and an understanding of business processes.
Within an operator, a new service may easily be viewed as a way to increase revenue per subscription, strengthen customer lock-in, or expand the existing service portfolio. Meanwhile, the customer asks why they should depend on that particular operator.
These perspectives can be fundamentally at odds. An operator-owned IT services company can certainly succeed just like any other IT company, provided it has genuine freedom to choose its technologies, partners, and business model. But in that case, its competitiveness would not depend primarily on its parent company’s network. Even then, the customer may still consider an independent provider a strategically better choice.
Owning the network is not necessarily a competitive advantage in the application business. From the customer’s perspective, it may also represent a dependency they would rather avoid.
Why Would AI-RAN Change This?
NVIDIA is bringing a new technological opportunity to the telecom market: GPU computing can support both radio network processing and other AI workloads. A shared computing platform is attractive. But the technical ability to share computing resources does not automatically translate into a commercially competitive AI service. Hyperscalers already have global customer bases, developer ecosystems, AI models, platform services, and the ability to aggregate computing demand across a massive customer base.
In an operator’s distributed GPU infrastructure, computing capacity is tied to the radio network’s geographical locations and real-time requirements. If the operator needs that capacity for radio operations, it cannot sell it to other customers without limits. If the operator deploys significant additional GPU capacity to meet demand for AI services, it ends up competing with global cloud providers. Furthermore, many Physical AI applications do not necessarily need operator-provided edge computing.
An industrial robot can perform critical inference locally. A factory can run its own edge infrastructure, while the cloud handles model training and centralized management. Operator edge computing may be justified in certain specialized scenarios, such as mobile applications distributed across wide geographical areas. But a specialized use case does not automatically create a large and profitable growth market.
Does AI-RAN Even Need a GPU?
There is another interesting question. Using AI to optimize radio network performance does not inherently require a GPU-based RAN architecture. In modern WLAN networks, for example, AI-powered analytics and optimization can operate in the cloud while access points continue to handle real-time radio processing using specialized chipsets. Of course, AI models integrated directly into the 5G signal-processing chain have very different real-time computational requirements from WLAN management analytics. Nevertheless, AI inference can, in principle, also run on specialized accelerators. GPU-based processing is therefore an architectural choice, not an unavoidable requirement of AI.
If AI-based channel estimation improves radio performance, operators should ask which implementation delivers that benefit at the lowest total cost and energy consumption. Performance improvement alone does not show that moving RAN processing to a GPU-based platform is the most economical solution.
Three Parties, Three Different Interests
- Nokia, NVIDIA and telecom operators do not have entirely identical objectives.
- NVIDIA wants to expand GPU computing into a new market.
- Nokia wants to deliver a competitive radio network architecture and sell new infrastructure solutions.
- Operators, meanwhile, need a return on their investments.
Improving radio network efficiency can provide operators with a clear business case. However, the idea of operators becoming major AI service platforms involves considerably more uncertainty. This is where the risk lies. For technology vendors, operators’ continued belief in future growth from higher-layer services can be an extremely effective sales argument—even when the operator’s own business case remains unproven.
That does not make the technology bad. It simply means that a technology vendor’s growth strategy should not be confused with the customer’s return on investment. Mobile operators have already widely criticized 5 G’s unfulfilled promises about new use cases and business opportunities. Now, similar promises are being presented again, repackaged around AI-RAN and 6G.
Perhaps Operators Should Ask a Different Question
For decades, operators have been asking how they can move up the value chain. Perhaps it is time to ask why they should. Reliable, secure, high-performance, and cost-efficient connectivity remains an extremely valuable service. AI can make delivering that service even more efficient. Operators can also develop new services in market segments where they have a genuine competitive advantage. But a new generation of network technology does not have to turn an operator into an application company to be a successful investment.
Tom Nolle’s observation about operators’ continued belief in higher-value services illustrates, above all, just how persistent this strategic ambition remains. Nokia and NVIDIA’s AI-RAN may represent a significant technological advance. An operator can provide AI computing, but that doesn’t mean enterprise customers will want to buy AI services from that operator.
Is AI-RAN the Telco Industry’s Never-Ending Promise of Growth? That is the question to answer before the next great growth promise becomes another major investment decision.
5Feet Networks Oy
