At the recent Qvantel Digital Leaders Summit in Rome, in her presentation called Navigating Reinvention – Deciding Telecoms Next Best Action, Camille Mendler of Omdia discussed, among other strategic topics, the opportunities that providing AI services offers telcos.
To me there were two main take-aways from Camille’s presentation. The first one was trust. Camille mentioned that companies who value network sovereignty say that telcos are their most trusted AI partner. With data sovereignly becoming increasingly important this was a telling statement. The next point for me concerned monetization with Camille stating that 15% of telcos already bill for AI tokens and 48% expect to do so in the next 18 months. Both figures came from surveys carried out by Omdia in 2026.
Looking beyond the Digital Leaders Summit we can see a significant amount of AI sales activity by telcos. Telenor opened Norway’s first sovereign AI factory. Deutsche Telekom and NVIDIA launched an Industrial AI Cloud in Germany, Swisscom is building a Swiss AI platform aimed at regulated sectors, Singtel offers its RE:AI cloud service to businesses and public bodies and in the UK BT Business has launched AI from BT.
AI services handle sensitive material. Prompts, documents and outputs often contain customer records, contracts or internal knowledge. Enterprises and public bodies want to know where that data is processed, who can access it, and whether it is used to train models for others.
Regulations provide further rules that telcos must follow. GDPR, the EU AI Act, sector rules in finance and healthcare, and national data-residency requirements all make location and control a compliance matter. For some organisations, this limits which AI providers they can use. It is why many of the AI services provided by telcos stress that data stays in-country.
Telcos already operate under licence and national regulation. They run infrastructure in-country, from core networks to edge sites and data centres, and they have long experience protecting subscriber data and running mission critical systems. Many also have established enterprise relationships built around SLAs and accountability.
They also provide the connectivity. That makes it practical to offer AI services close to where the data sits, hosted locally, and combined with secure network access.
Location matters for performance as well as compliance. Some AI use cases, such as live video analytics for public safety, emergency services or remote healthcare, need round-trip response times of around 10 milliseconds or less. A round trip to a hyperscaler’s data centre, even a regional one, uses up much of that budget before the application has done any work. Faster fibre doesn’t change this, because it comes down to the speed of light.
A 5G AI slice addresses this by combining a dedicated, SLA-backed 5G network slice with GPU capacity at the network edge. AI inference runs close to where the service interaction happens, and the data can stay within the telco’s network and within national borders. This is something a centralised cloud cannot offer from a distance, and it extends the telco’s role from carrying AI traffic to hosting the AI workload itself.
It is also a different product to sell. Pricing is based on compute cycles, tokens, API calls or outcomes alongside network performance. TM Forum’s Innovate Asia event in November included a catalyst project, with Qvantel, Nokia, Netradar and telcos including NTT, du, Ooredoo, VNPT and Telin showed how a 5G AI slice can be created, sold, monetized and assured, enabling telcos to play a central role in the AI value chain.
An AI service still has to be sold and monetized. AI usage doesn’t fit the flat fees and per-minute or per-gigabyte charges that most telco billing was designed for. A BSS supporting these services needs to handle:
Token-based billing. AI processing (LLMs or local models) is measured in input and output tokens, often at different rates by model, context size or priority.
Outcome-based pricing. Some customers prefer to pay per result, such as a resolved case or a processed document, rather than per unit of consumption.
Algorithmic pricing. GPU capacity is limited and costs vary with demand and model choice, so prices may need to adjust to reflect cost-to-serve.
SLA-backed offers. Products like 5G AI slices carry commitments on latency, availability and compute capacity. The BSS needs to check that an offer can be delivered before the SLA is agreed, link the commitments to the product and the customer, and track performance against them so that credits or penalties apply when they are missed.
Flexible commercial models. Examples include committed spend with burst capacity, prepaid credits, shared budgets across teams, and bundles with connectivity.
Partner settlement. Many offers combine models, platforms and third-party applications, so revenue sharing needs to be automated.
Real-time spend controls. AI costs can rise quickly, so customers want caps, alerts and usage visibility.
If product changes take months, a telco will struggle to keep up with how AI pricing is evolving. A catalog-driven, real-time BSS makes it possible to launch and adjust offers much faster.
The BSS holds customer identities, contracts, usage records and billing data. Usage records for AI services can also show how a customer is using AI, which is sensitive information. If a telco’s position is that it offers a secure, sovereign service, its BSS vendor needs to meet the same standard.
In practice, that means security built into the architecture and operations need to have recognised quality management practises, clear data-handling practices, and flexible deployment options such as on-premises, private cloud or sovereign cloud. Experience operating in regulated environments also matters. A gap at the vendor level weakens the assurances the telco gives its customers.
Omdia’s presentation at the Digital Leaders Summit highlighted the importance of trust and a new approach to monetization. Telcos have the infrastructure, regulatory standing and customer relationships to offer sovereign AI services, including edge-based services that a distant data centre cannot match on latency. They also have trust, and to ensure that they can sell AI services effectively, they need a BSS that can price AI the way customers want to buy it and support the SLAs behind it, supplied by a trusted vendor they can rely on to protect customer data.
Jukka Heiska
CMO, Qvantel
