The telecom industry has officially declared that today's most powerful AI models are not good enough — at least not for running networks. At MWC Barcelona, the GSMA launched Open Telco AI, a collaborative initiative designed to build AI systems purpose-built for the telecommunications sector.
The Problem With General-Purpose AI
Despite the rapid advances in frontier models from OpenAI, Google, and Anthropic, the GSMA found that these systems consistently underperform on telecom-specific tasks. Only 16% of generative AI deployments in telecom are currently applied to network operations — the area where carriers need the most help.
The core issue is domain specificity. General-purpose LLMs were trained primarily on internet text, code, and conversational data. They lack deep understanding of 3GPP standards documents, radio frequency engineering, network topology, and the operational telemetry that telecom engineers work with daily.
What Open Telco AI Delivers
The initiative is organized around four pillars:
Telco-Specific Models
AT&T has released a family of open-weight models trained on publicly available telecom data. Khalifa University contributed RFGPT, a radio-frequency language model, while AdaptKey AI built a Large Telco Model on NVIDIA's Nemotron architecture. These models range in size and are optimized for tasks from network troubleshooting to standards interpretation.
Open Data
A curated library of knowledge graphs, embeddings, and fine-tuning datasets has been assembled by contributors including Huawei, Purdue University, Yale University, and the University of Leeds. NVIDIA is providing pipelines for generating synthetic training data.
Compute Access
AMD and cloud partner TensorWave are providing GPU capacity for model training, fine-tuning, and inference, along with open toolchains to lower the barrier to entry for smaller operators and research institutions.
Benchmarks and Leaderboard
A public leaderboard evaluates models across seven telecom-specific benchmarks, giving the industry a transparent way to measure progress.
Early Traction
The AI Telco Troubleshooting Challenge, launched alongside the initiative, attracted over 1,000 registrations ahead of MWC26. Winners will be announced during the conference.
Why This Matters Beyond Telecom
The GSMA's move highlights a broader trend: as AI adoption matures, industries are discovering that general-purpose models hit a ceiling on specialized tasks. Healthcare, legal, and manufacturing sectors have reached similar conclusions. Open Telco AI could serve as a template for how entire industries organize around domain-specific AI development — pooling data, compute, and expertise rather than waiting for foundation model providers to solve their problems for them.
