Jensen Huang isn’t satisfied owning the data center. Now he wants the airwaves.

At Mobile World Congress 2026 in Barcelona, Nvidia unveiled a coalition of 13 major companies — Nokia, T-Mobile, Deutsche Telekom, Cisco, Ericsson, SoftBank, SK Telecom, BT Group, and more — all committed to building sixth-generation wireless networks on open, AI-native platforms. Not AI-assisted. Not AI-enhanced. AI as the foundation.

This is Nvidia doing what Nvidia does best: establishing the platform layer that everyone else builds on, then collecting rent for the next decade.

What “AI-Native” Actually Means

There’s a real distinction here, and it matters. Today’s 5G networks were designed with traditional architectures. AI got bolted on afterward for optimization — a turbocharger on a horse-drawn carriage.

AI-native 6G flips the entire approach. The network architecture assumes AI from the ground up. Radio traffic management, energy optimization, security monitoring, resource allocation — all running on models that continuously learn and adapt. The network becomes a distributed computing platform that also happens to handle phone calls.

The technical backbone is AI-RAN: artificial intelligence radio access networking. It’s a software-defined approach where GPU-accelerated platforms handle traditional radio processing and AI workloads simultaneously on the same hardware.

This Isn’t a Slide Deck

The demos at MWC were the real story. T-Mobile ran concurrent AI and RAN processing on Nvidia’s platform using Nokia’s CUDA-accelerated software — supporting commercial devices streaming video and running generative AI alongside standard 5G service.

SoftBank pulled off an industry first: 16-layer massive MIMO using fully software-defined 5G on Nvidia’s AI-RAN platform. Massive MIMO is the backbone of high-capacity wireless, and doing it entirely in software opens the door to rapid iteration that hardware-defined systems can’t match.

SynaXG demonstrated software-defined AI-RAN running 4G, 5G in both sub-6GHz and millimeter wave bands, plus agentic AI workloads — all on a single Nvidia GH200 server. They hit 36 Gbps throughput with under 10 milliseconds latency across 20 component carriers. Those are carrier-grade numbers from a software stack.

And in Indonesia, Indosat Ooredoo Hutchison completed Southeast Asia’s first AI-powered 5G call, including remote control of a robotic dog across borders over a live network. Because of course there was a robot dog.

The Scale Is Staggering

MWC 2026 saw triple the AI-RAN demonstrations compared to last year — 33 total, with 26 built on Nvidia platforms. The AI-RAN Alliance now includes over 130 companies.

DeepSig showed an AI-native air interface that jointly learns optimal signal encoding and decoding, achieving roughly 2x higher throughput from the same spectrum. That’s not incremental improvement. That’s transformative.

Nvidia also unveiled a Nemotron-based large telco model designed for autonomous network operations, plus blueprints for multi-agent orchestration covering energy savings, network configuration, and advanced autonomy. The Nemotron framework is open, giving operators transparency into training data — a notable move given growing concerns about AI opacity.

Why You Should Care (Even If You’re Not in Telecom)

The physical AI revolution — autonomous vehicles, delivery drones, industrial robots, smart cities — needs networks that can make intelligent decisions in microseconds. Traditional network management can’t deliver that at scale. AI-native 6G is the missing infrastructure layer.

Capgemini demonstrated this at MWC through Project ULTIMO: autonomous shuttles equipped with Nvidia Jetson Orin modules processing sensor data locally while streaming video and telemetry over 5G to AI applications running on AI-RAN servers. The network itself prioritizes mission-critical traffic while simultaneously running AI workloads for scene understanding and safety detection.

That’s the template for how autonomous systems will actually work in the real world. Not standalone AI on a device, not pure cloud AI, but a hybrid where the network itself is an intelligent participant.

Nvidia’s Real Play

Zoom out and the strategy is textbook Jensen Huang. Just as CUDA made Nvidia GPUs indispensable for AI training by owning the software layer, the AI Aerial stack, AI-RAN platform, and Nemotron telco models create the same ecosystem lock-in for telecommunications.

If 6G networks are built on Nvidia’s platform from the ground up, every telecom operator becomes a long-term Nvidia customer. Training, inference, edge computing, robotics, automotive, and now telecommunications — Nvidia isn’t riding the AI wave. It’s building the ocean.

T-Mobile CEO Srini Gopalan summed it up: “As 6G becomes the backbone of the AI era, telecom will serve as the nervous system of the digital economy.”

The Timeline

Commercial 6G deployment is expected around 2030, with trials starting as early as 2028. That sounds far off, but the architectural decisions being made right now will determine how wireless infrastructure works for decades.

After this week in Barcelona, one debate is settled: 6G will be AI-native. The only question left is whether the open, interoperable vision this alliance is pushing will survive contact with commercial reality — or whether it fragments into the same proprietary silos that plagued previous generations.

Given Nvidia’s track record of turning “open platforms” into ecosystems where it holds all the leverage, smart money says Jensen Huang already knows the answer.