What NVIDIA, Nokia and the operators announced
According to NVIDIA's press release, also published on GlobeNewswire, several operators are installing NVIDIA accelerated computing in their data centers, at metro sites and at the base stations themselves, to run AI models closer to the people, businesses and machines that use them. The most notable announcements were:
- T-Mobile (United States): completed a call on its commercial network in Bellevue, Washington, using NVIDIA Aerial RAN Computer (ARC) servers, Nokia radio software and a commercial phone, connected to its existing 5G core.
- Orange (France): offers open models, such as NVIDIA Nemotron, as a managed service on its Live Intelligence platform, and is deploying ARC at distributed sites for voice services with real-time translation for emergency and field teams.
- Elisa (Finland): carried out a field trial that combined voice, mobile data and video analytics on a single 5G platform, and already has a business voice assistant running at a distributed site. It plans commercial services in 2027.
- Deutsche Telekom and BT: lab validations and live trials planned for 2027; BT is building agents that review photos of installations to avoid repeat technician visits.
NVIDIA also presented two new ARC server configurations, to be manufactured by Dell, Quanta Cloud Technology and Supermicro, and said that nearly 20 operators have announced collaborations of this kind with Nokia. Jensen Huang, NVIDIA's founder and CEO, summed it up this way: networks were built to carry the world's conversations, and now they will also carry its intelligence.
What AI-RAN is and why bring AI to the network edge
The RAN (radio access network) is the part of the mobile network that connects your phone to the antenna. Until now it ran on specialized equipment that did only that. The idea behind AI-RAN is that those same machines, equipped with graphics processors, can handle the radio and, at the same time, run artificial intelligence applications.
The Orange Business case illustrates what it is for. According to Telecom Reseller and The Fast Mode, Orange combines NVIDIA Blackwell servers in its central cloud with ARC equipment at its regional points of presence. In a demonstration with a wildfire scenario, the AI translates in both directions between a caller speaking another language and the emergency operator, transcribes the call and prepares summaries. If a site goes down or becomes overloaded, requests are rerouted to another. Orange has about 50 points of presence in Europe and plans to reach 75 by 2030.
Three advantages are sought: lower delay in responses, resilience against outages or traffic spikes, and control over where data is processed, which is key for banks, insurers or public entities that do not want their customers' voices to leave the country or a trusted environment.
What is still a promise
It's worth reading the announcement with a cool head. TelecomTV points out that most results come from lab tests, short trials or plans for 2027, and that phrases like “world first” are NVIDIA's characterizations, not independent verifications. The same outlet notes that NVIDIA is an investor in Nokia.
Telecom Reseller also notes that Orange has not published results demonstrating response times or resilience under real conditions, that distributing computing does not by itself guarantee regulatory compliance and that operating servers at many sites carries equipment, power and cooling costs. None of the sources publish latency figures. And no Latin American operator appears on the announcement's list, so in the region this technology will first be seen in pilots.
What it means for companies in Colombia and Latin America
Although the deployments are in Europe and the United States, the direction is clear and has consequences for anyone who already uses or plans to use AI in their communications:
- The network is becoming part of the AI solution. The quality of a voice agent depends not only on the model, but on how long the audio takes to travel back and forth. The closer it is processed, the more natural the conversation feels.
- Data location is becoming a buying criterion. Ask your providers where the recordings and transcripts of your calls are processed and stored, and whether that is compatible with data protection law.
- Open models are gaining ground. Operators like Orange offer models that customers can audit and adapt, instead of depending on a single provider.
- There's no need to wait. Voice and chat agents already work today on existing infrastructure; what matters is choosing an architecture that can take advantage of these improvements when they arrive.
At 19 80 TIC Cloud we integrate AI voice agents with your business phone system, so your company can automate customer service today and be ready for what comes next.
