Nokia’s AI-RAN Field Trials Go Global: Eight Operators, One Platform, and the Long Road from Lab to Live

On a Monday morning in late September, Nokia’s press office pushed out a list of operator names that, read together, spans four continents: A1 Group in Austria, Chunghwa Telecom in Taiwan, du and e& in the UAE, Mobily and stc and Zain Saudi in Saudi Arabia, and TPG Telecom in Australia. The headline was that AI-RAN has moved from early evaluation to lab and live field trials. The subtext was more interesting: the industry’s most consequential radio architecture bet is no longer a single-vendor science project. It is a multi-region, multi-operator, multi-workload campaign—and it is being run on a commercial platform that Nokia only recently started shipping.

For anyone who has watched AI-RAN oscillate between keynote slide and press release over the past three years, the temptation is to file this under “more of the same.” That would be a mistake. The composition of this cohort, the timing relative to Nokia’s platform launch, and the explicit link to NVIDIA’s software toolchain together tell a story about how AI-RAN is actually being productized—and about where the friction will show up next.

The scene: a press release that reads like a deployment map

Nokia’s announcement framed the news as “growing global momentum for AI-RAN,” with operators “advancing AI-RAN from early evaluation to lab and live field trials.” That phrase—“lab and live field trials”—is doing a lot of work. It distinguishes this cohort from the evaluation-stage engagements Nokia has been publicizing for the better part of a year. Lab trials are controlled, instrumented, and reversible. Field trials are not. Once you move AI-RAN workloads onto live spectrum with real subscribers, you inherit real interference, real mobility patterns, and real operational risk.

The operator list is geographically deliberate. A1 Group gives Nokia a Central European anchor with a multi-country footprint. Chunghwa Telecom is a sophisticated Taiwanese operator that has already been publicly linked to Nokia’s AI-RAN push. du and e& anchor the Gulf, where sovereign AI ambitions and 5G-Advanced rollouts are converging faster than almost anywhere else. Mobily, stc, and Zain Saudi deepen Nokia’s Saudi presence at a moment when the Kingdom is pouring capital into AI infrastructure. TPG Telecom provides an APAC challenger with a history of aggressive network modernization.

What is notable is not any single name. It is the diversity—different regulatory regimes, different spectrum positions, different backhaul realities, different organizational appetites for risk. Nokia’s own framing acknowledged this: the breadth of engagement “shows that AI-RAN is evolving from a research topic into a strategic priority for a growing number of service providers.” That is a vendor talking, but the underlying observation is defensible. When operators across the Gulf, Europe, and Asia-Pacific independently decide to spend engineering cycles on the same architectural idea, something has changed.

What Nokia is actually shipping

The announcement builds on a sequence of milestones that Nokia has been stacking since mid-2026. The most important is the launch of what Nokia calls the industry’s first commercial AI-native RAN platform. That claim is contested territory—every major vendor now has an AI-RAN story—but “commercial” is the operative word. A commercial platform means SKUs, support contracts, and upgrade paths. It means operators can plan capex against it rather than treating it as a co-innovation experiment.

Nokia also referenced evaluations with T-Mobile, SoftBank, and Indosat Ooredoo Hutchison, and noted that previously announced trials with NTT DOCOMO “continue to progress steadily.” DOCOMO matters disproportionately here. It has been the most public multi-vendor Open RAN operator in the world, and its willingness to keep Nokia in the loop on AI-RAN suggests the vendor’s platform is interoperating with at least some of the multi-vendor plumbing DOCOMO has built. Nokia also cited a growing number of OEM partners and expanded collaboration with NVIDIA on AI-RAN platform and software development tools.

That NVIDIA thread is the technical spine of the story. Nokia’s AI-RAN architecture puts GPU-accelerated compute alongside—or in place of—traditional RAN acceleration, allowing the same infrastructure to serve both connectivity workloads and AI inference. The pitch is that radio networks become “programmable platforms supporting both connectivity and AI workloads.” Whether that pitch survives contact with real traffic is precisely what field trials are for.

Why this cohort matters more than the last one

Nokia has announced AI-RAN engagements before. In July, it expanded trials to eight operators; the current announcement overlaps heavily with that cohort but adds field-trial language and a clearer commercial framing. The risk is that this looks like a re-announcement. The defense is that the substance has shifted: earlier engagements were evaluations, this one includes live field trials, and it sits on top of a shipping platform rather than a reference design.

There is also a competitive dimension. Ericsson has been pushing its own AI-in-RAN subscription model and landed a sole global role in SK Telecom’s AI-RAN pilot. Samsung and KDDI have demonstrated AI-driven RAN gains on live 5G in Tokyo. Huawei continues to market SingleRAN as a mobile AI foundation, though its addressable market in Europe and North America remains constrained. NVIDIA has been building its own AI Aerial story, including 3x spectral efficiency claims from SoftBank field trials. In that context, Nokia’s announcement is less about novelty than about breadth: no other vendor has publicly assembled this many named operators across this many regions on a single commercial AI-RAN platform.

That breadth is also a hedge. If AI-RAN economics disappoint—and the flat-RAN-economics critique is not going away—Nokia will have learned it across a portfolio of operators rather than betting on one flagship. If it works, Nokia has reference customers on four continents.

The technical questions the field trials will have to answer

The press release is deliberately light on technical specifics, which is normal at this stage. But the questions that matter are well understood, and operators in this cohort will be answering them in different ways.

First, there is the compute-sharing question. AI-RAN’s central promise is that GPU capacity can be dynamically allocated between RAN processing and AI inference. In a lab, that is a scheduling problem. In the field, it is a capacity-planning problem: how much headroom do you reserve for RAN, how much do you monetize as AI, and what happens to your cell-edge throughput when an inference job spikes? Nokia’s platform claims to handle this; the operators will find out.

Second, there is the energy question. GPU-accelerated RAN is not obviously more power-efficient than purpose-built silicon at low load. The efficiency case depends on utilization—running AI workloads on the same hardware that would otherwise idle. Operators in the Gulf, where energy costs are low but sovereign AI mandates are high, may weight this differently than operators in Europe, where energy costs are a board-level issue. That divergence will produce different trial designs and different conclusions.

Third, there is the integration question. Nokia’s platform is proprietary at the platform layer even as it plugs into Open RAN interfaces. Operators like DOCOMO, which have invested heavily in multi-vendor interoperability, will test whether AI-RAN can coexist with the disaggregated architectures they have already built. Operators like Zain Saudi, with a more concentrated vendor footprint, may care less. Both are valid paths, but they lead to different long-term lock-in profiles.

The Open RAN angle

It is tempting to treat AI-RAN as a separate track from Open RAN. Nokia’s messaging encourages this—AI-RAN is framed around AI-native 6G, not around disaggregation. But the two are converging whether the industry says so or not. The same cloud-native, software-upgradeable, API-exposed architecture that makes Open RAN attractive is what makes AI-RAN deployable. You cannot dynamically schedule AI workloads onto RAN infrastructure you cannot program.

That convergence cuts both ways. For Open RAN advocates, AI-RAN is a demand driver: it gives operators a revenue-side reason to invest in programmable infrastructure, not just a cost-side one. For skeptics, AI-RAN is a risk: it could become a proprietary layer that sits on top of open interfaces and captures the value, leaving Open RAN as undifferentiated plumbing. The operators in Nokia’s cohort will, collectively, determine which outcome prevails. If they demand open interfaces at the AI-RAN layer—scheduling APIs, model portability, multi-vendor GPU support—the ecosystem stays open. If they accept vendor-specific toolchains because they work, the layer consolidates.

Nokia’s expanded NVIDIA collaboration is the pivot point. NVIDIA’s tools are powerful and increasingly standardized, but they are not neutral. Operators that build AI-RAN on NVIDIA’s stack inherit NVIDIA’s roadmap. That is not necessarily bad—NVIDIA has been a constructive force in Open RAN—but it is a dependency worth naming.

What to watch next

The most useful thing about this announcement is that it gives the industry a scoreboard. Over the next two quarters, watch for three signals.

First, which operators move from field trial to commercial deployment, and on what timeline. Nokia’s language—“accelerating the evolution toward AI-native 6G networks”—sets an expectation. If A1 or TPG or Chunghwa converts to a commercial contract in 2027, the platform claim is validated. If trials quietly extend, that is informative too.

Second, whether the operators publish measured results. SoftBank and NVIDIA published spectral efficiency numbers; Samsung and KDDI published live-network gains. Nokia’s cohort has not yet. Independent, operator-published data would do more for AI-RAN credibility than any vendor benchmark.

Third, whether the Open RAN community engages with the AI-RAN layer or cedes it. The AI-RAN Alliance, now well past 100 members, is the natural venue. If the Alliance’s working groups start defining interfaces between AI-RAN platforms and multi-vendor RAN, the convergence story holds. If AI-RAN develops its own parallel standards track, the industry will have recreated the fragmentation it spent a decade trying to undo.

For now, the takeaway is simpler. Eight operators across four regions have agreed to put AI-RAN on live networks. That is not proof the technology works at scale. It is proof that enough serious operators believe it might. In telecom, that is usually how the next architecture gets built.

Sources

• Nokia Advances AI-RAN Adoption with Global Operator Trials on NVIDIA Platforms