- T-Mobile and Nvidia are working with Nokia and a developer ecosystem to deploy physical AI apps — robots, autonomous vehicles, smart city systems — over distributed edge networks
- The initiative reframes the cell site as an AI computing platform
- It also puts to the test the theory that telcos are uniquely positioned to capture new revenue and lead the AI economy
NVIDIA GTC, SAN JOSE, CALIF. — Telcos' physical real estate, networks and technical chops are the holy trinity that will usher in floods of new revenue in the AI economy. Telcos are uniquely positioned to deliver AI applications in edge locations such as radio towers and central offices, fed by data carried on telco networks and tended to by telcos' deep ranks of skilled and educated engineers.
That's the theory, at least.
T-Mobile and Nvidia are putting that theory to the test, as the two companies announced Monday that they are working with Nokia and a growing developer ecosystem to deploy physical AI applications over distributed edge networks.
The initiative reframes the wireless cell site as an AI computing platform capable of powering everything from autonomous vehicles and robots to smart city traffic management and industrial safety systems.
The announcement, made at Nvidia's GTC 2026 conference, marks a significant step in T-Mobile's long-running AI-RAN buildout and signals that the era of inference at the network edge is arriving fast.
"The radio tower used to be [just] a radio tower," Jensen Huang, Nvidia founder and CEO , said during his keynote address Monday. "In the future, it's going to be an AI infrastructure platform — a robotics radio tower that can reason about traffic, figure out how to adjust the beam to save as much energy as possible and increase fidelity."
What's new
T-Mobile is piloting Nvidia's RTX Pro 6000 Blackwell Server Edition AI infrastructure at mobile switching offices as part of its AI-RAN Innovation Center program, demonstrating that cell sites can support high-performance edge AI workloads while continuing to deliver 5G connectivity. The pilot runs on Nokia's anyRAN software and integrates with Nvidia's Metropolis platform.
Physical AI developers including Fogsphere, LinkerVision, Levatas, Vaidio and Siemens Energy are building reasoning and vision AI agents on top of that infrastructure, using the Nvidia Metropolis Blueprint for video search and summarization — now in its third iteration — to deploy use cases across smart cities, utilities, industrial facilities and construction sites.
The City of San Jose is among the first municipalities to assess the technology, specifically a computer vision-based City Operations Agent that integrates digital twin capabilities to optimize traffic light timing, targeting a fivefold improvement in incident response times. Separately, Levatas and Skydio are automating inspection of hundreds of thousands of miles of transmission lines over 5G, detecting anomalies such as leaning power poles, corrosion and thermal hotspots five times faster than current methods.
Why this matters
Huang has been making the case for some time that the telecom industry is on the verge of a fundamental transformation. In his keynote, he described telecommunications as a roughly $2 trillion industry whose infrastructure — the base station — is about to be completely reinvented. The physical AI announcement with T-Mobile is a clear demonstration of what that reinvention looks like in practice.
And Nvidia isn't alone. Leaders like Bell Canada and AT&T are achieving measurable return on investment by reinventing themselves for AI.
The basis for the reinvention is that AI agents that operate in the physical world — autonomous vehicles, robots, surveillance systems, industrial sensors — generate massive amounts of data that needs to be processed with very low latency. Sending all of that to a centralized cloud isn't viable. Putting all the compute on the device is too expensive. The network edge, sitting between device and cloud, is the logical place to offload heavy computation.
T-Mobile CEO Srini Gopalan framed the strategy in terms of network architecture: "Turning networks into distributed AI computing platforms to unlock the full potential of physical AI will require ultra-low latency and space-time coherency at the network edge for billions of endpoints, and that's what we've built at T-Mobile," Gopalan said in a statement.
T-Mobile's position
T-Mobile has been building toward this moment for several years. The carrier was the first in the U.S. to launch a nationwide 5G standalone network — a prerequisite for the kind of guaranteed quality of service that physical AI agents require. It was also the first U.S. carrier to pilot Nvidia's AI-RAN infrastructure with Nokia's anyRAN software, giving it a head start on the operational and integration challenges that competitors will now face.
That early-mover advantage matters. Deploying AI compute across thousands of cell sites and mobile switching offices is a multi-year infrastructure project, and T-Mobile's existing relationship with Nvidia and Nokia gives it both a reference architecture and a developer ecosystem that rivals are still assembling.
Nvidia's telco play
For Nvidia, the telco vertical is one of the clearest paths to distributing AI inference at global scale. The company's Aerial AI-RAN platform is purpose-built for this use case, and the partnership with T-Mobile gives Nvidia a high-profile commercial deployment to point to as it pitches the broader vision to other carriers worldwide.
The Metropolis VSS 3 Blueprint — which can summarize long-form video up to 100 times faster than manual review and search hours of footage for specific events in under five seconds — gives developers a production-ready framework for building physical AI applications without starting from scratch, Nvidia said. That lowers the barrier to entry and accelerates the ecosystem development Nvidia needs to make AI-RAN a platform rather than a point solution.
Challenges and competition
The path forward isn't without friction. Deploying AI servers at cell sites introduces power, cooling and space constraints that don't exist in traditional data center environments. Monetization models for edge AI remain nascent, and enterprise customers are still evaluating whether network-based compute can reliably meet the latency and reliability requirements of mission-critical physical AI applications.
On the competitive front, AT&T and Verizon are both investing in edge computing infrastructure and AI networking. And Ericsson is developing its own AI-native RAN capabilities. Cloud providers including AWS, with its Wavelength service, and Microsoft Azure with its edge computing portfolio are also competing for the workloads that T-Mobile and Nvidia are targeting. What T-Mobile and Nvidia can point to that most rivals cannot is a working pilot, a named municipal customer and a developer ecosystem already building on the platform.