The AI Fabless Model That Could Save Telcos

GSMA can turn fragmented telcos into a fabless AI giant, owning the models that will run the networks of the future.

From Fabs to Fabless: A Lesson from NVIDIA

NVIDIA is today the most valuable semiconductor company in the world. Its GPUs power everything from AI clusters to gaming rigs.

Yet here’s is the thing, NVIDIA does not own a single fabrication plant.

Unlike Intel, which once built chips end-to-end, NVIDIA chose the fabless path. It focused on what mattered, designing the most advanced architectures and outsourced the capital-intensive, high-risk manufacturing to neutral foundries like TSMC. That bet transformed it into a trillion-dollar giant.

Telecom now faces a parallel moment. Artificial Intelligence will be the backbone of the entire industry, not an add-on at the edge. Horizontal AI foundation models, and soon physical AI embedded in networks will underpin everything from network optimization and fraud detection to customer care and automation. Some telcos are already dabbling: a fine-tuned chatbot here, a predictive maintenance pilot there. But these efforts are narrow, vertical, and reliant on external vendors. Worse, they risk binding operators into permanent dependency on generic hyperscaler models, built for internet text rather than telco realities.

The costs of going it alone are prohibitive. Designing and training a frontier-scale AI model takes hundreds of millions of dollars in compute and scarce research talent. No single operator can shoulder it. But just as NVIDIA proved in chips, telcos don’t need their own fabs. What they need is design, focus, and pooled investment.

A GSMA-hosted AI fabless consortium, where members share anonymized data, pool capital, and jointly commission training runs from cloud GPU factories, offers a way out of the trap.

With only 5% of annual telco capex redirected, that’s nearly $15 billion in collective firepower. Enough to build frontier-class telco models that no hyperscaler can replicate, models that become smarter every day they operate across global networks.