Telenor picks F5 and Nvidia BlueField-3 for sovereign AI service plane
Telenor AI Factory will run F5 BIG-IP Next for Kubernetes on Nvidia BlueField-3 DPUs as the service plane for sovereign AI, with token-aware governance planned as commercial services scale.

Why it matters
- Telenor AI Factory selected F5 BIG-IP Next for Kubernetes on Nvidia BlueField-3 DPUs for its sovereign AI infrastructure.
- F5 says telco accounted for 13% of its product bookings in the most recent quarter.
- Guetari predicts mobile subscriptions will include token allotments within two to three years.
- Telenor plans token-aware governance, circuit breaking, and inference-aware traffic optimization as services scale.
- Telco AI factory activity has accelerated sharply over the past three to six months, per F5.
Telenor AI Factory has selected F5's BIG-IP Next for Kubernetes, running on Nvidia BlueField-3 data processing units, as the AI service plane for its Nvidia-powered sovereign AI infrastructure. The platform targets Norwegian and European organizations that want high-performance AI services while controlling where data is processed, stored, and accessed.
The stakes for the telecom industry are larger than one deal. Operators have spent two decades trying to avoid becoming a "dumb pipe." Every year at MWC and other telco shows, a new technology wave — cloud, virtualization, software — promises new revenue-generating services. Every year, most operators use the technology to cut costs and keep selling connectivity. F5 and Telenor's arrangement offers a glimpse of a different path: selling governed, metered, auditable AI rather than raw capacity.
F5 supplies the platform's foundation from day one: multitenancy and isolation, policy-based traffic control, and real-time telemetry for auditing and compliance. Telenor plans to add token-aware governance, circuit breaking, and inference-aware traffic optimization as its commercial AI services scale.
Ahmed Guetari, senior vice president and general manager of F5's service provider business, described the deal's logic in an interview. His central claim: sovereignty is a control problem, not a geography problem.
What does sovereignty actually require?
Most of the sovereign AI conversation centers on GPUs located on national soil. Guetari called that table stakes. A rack of GPUs in Oslo does not make an AI service sovereign if the operator cannot prove which tenant accessed what data, which model was called, and where the traffic went.
Guetari was candid about the term itself. "Sovereignty is very vague and not always clear," he said. "It's about control, auditability, observability."
He separated data locality — data staying in-country — from sovereignty, which he described as "more about control, control of what happens to your data and your workload." He then positioned F5's credentials: "This is what we have been doing for two decades."
F5 has always sat in the data path: deciding what an application can communicate with, enforcing security rules, and logging what happened. For a regulated bank or a government agency, the ability to reconstruct exactly what happened after an incident is worth more than a press release about a data center's location.
Why is the token the new unit of economics?
The most consequential part of Guetari's argument concerns business models, not security. He argued that every successful technology business has found the right unit to monetize.
"The cloud is selling the unit of the cloud, which is the CPU core, very well. SaaS is the seat," he said. "Telcos have been very successful early on using the voice minute, and then they moved to the gigabyte. Sooner or later, that unit will be the token, and most telcos will have to move to the token. And the ones who will not go to token, unfortunately, will be just pipes."
Guetari pointed to Chinese operators that have begun adding tokens to their plans. He expects that within two to three years, a mobile subscription will include a token allotment, much like it includes a data allotment today.
If that happens, telcos need to meter, govern, and bill token consumption with the same rigor they apply to minutes and megabytes. That is why token-aware governance on Telenor's roadmap matters more than it might seem. F5 does not provide the billing systems, but Guetari noted that F5 feeds them the granular records they need, as it does today in mobile networks.
Enterprises want the same discipline. Guetari recalled CIOs telling him they had been "hit by the public cloud bill" and wanted to avoid a repeat with AI. Allocating tokens as a budget by user, model, or tenant turns AI from an open-ended expense into something a CFO can manage.
Why does the data path change with AI?
There is also a technical rationale for an AI-aware service plane. Traditional application delivery was built around a predictable unit of work: the HTTP request.
"Now, when you have an HTTP request, it can carry 10,000 tokens. These tokens could be part of a bot or an agent, and the behavior is not the same at all," Guetari explained. "If you don't have that atomic understanding, you're wasting the entire GPU cluster."
GPUs are the most expensive resource in the building, and a load balancer that treats every request the same will leave capacity idle. Running F5's data plane on BlueField DPUs offloads networking and security from host CPUs, whose real job is keeping the GPUs fed. Guetari said F5 has improved tokens per second, time to first token, and tokens per watt — the metrics that determine whether an AI factory makes money.
Can telcos escape their track record?
Guetari did not soften his assessment of the industry. "Two years ago, the telcos did not believe in the opportunity, did not see it," he said.
He credited Indosat Ooredoo Hutchison, another F5 customer, as one of the few operators to move early. Its early AI factory revenue came largely from selling bare-metal GPU capacity, while the company quickly recognized the need to build the software stack to offer sovereign AI services to banks and governments. The path to token-as-a-service may run through capacity deals first.
Telcos have the raw ingredients for this market: power, space, distributed facilities, national trust, and enterprise relationships. What they lack is speed. Twenty years ago, a new service meant building a new network, so every initiative faced scrutiny to death. Today everything is software, but the approval culture has not changed. Hyperscalers experiment and kill what does not work. AI moves too fast for 18-month telco business cases.
What does the deal mean for F5?
The deal also serves F5's own growth problem. Guetari said telco accounted for 13% of product bookings in F5's most recent quarter, and that mobile core spending tends to come in waves, with a likely lull before 6G.
"I don't want to see my telco business stay stagnant or grow at one digit," he said. AI factories give F5 a way to grow independently of the mobile cycle, and Guetari said telco AI factory activity has accelerated sharply over the past three to six months.
What should buyers take from this?
The Telenor deal is small today, and both companies describe much of the token governance as planned rather than deployed. But the direction is right. Guetari captured the shift: "Each application is consulting an LLM, so the AI is part of the workflow. And if you don't embed yourself in that workflow, you're going to miss the entire market."
That applies to telcos as much as to F5. Operators who treat AI factories as a real estate play, renting out GPUs by the rack, will end up as pipes again — only with more expensive pipes. Those who build a governed, metered, auditable service layer on top can finally sell something enterprises value more than bandwidth. For IT leaders evaluating sovereign AI providers, the test is simple: do not ask where the GPUs are. Ask how the provider controls, isolates, meters, and audits the traffic flowing to them.
Source: networkworld.com
More from Rebecca Stone
Show full bio
Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
802 articles