Opening the trial booth

The scene marked the start of a two‑week trial of the RFGPT model, a radio‑frequency foundation model that translates signal data into everyday words.

What the model does

It identifies modulation schemes and can analyse crowded spectrum where several signals overlap. The model also estimates interference strength and reports user density for Wi‑Fi and 5G NR, giving a view of network utilisation.

It can answer natural‑language queries about 5G NR parameters such as sub‑carrier spacing, synchronization signal block patterns, and channel state information reference signals.

Why it matters now

Interpreting RF data normally requires specialist engineers and custom deep‑learning models for each task. By letting operators ask plain‑language questions, RFGPT could speed up network planning, spectrum management and interference mitigation.

Faster insight means lower labour costs and quicker rollout of new services. The trial also lets AMD showcase its AI compute infrastructure, which recently saw data‑centre revenue rise 97% in the third quarter of 2026.

Which businesses benefit most?
Network owners and operators that spend time analysing spectrum can cut costs and improve speed of decision‑making.
What risks remain?
The model is still being evaluated; inaccurate answers could lead to mis‑configurations if not double‑checked by engineers.

Tomorrow’s outlook

If the trial shows usable performance, RFGPT could become a standard tool for radio access network (RAN) optimisation.

Operators may integrate the API into existing OSS/BSS (operational support and business support) platforms to automate routine spectral analysis.

The partnership with GSMA also hints at future benchmark suites that could set industry standards for telecom‑specific language models.

  1. November 2025GSMA Foundry and Khalifa University began a strategic partnership.
  2. 2026In its third quarter, AMD reported record revenue of $13.4 billion; its data centre revenue rose 97%, driven by record sales of its Instinct AI chips and EPYC processors.

What to do in your business tomorrow

  1. Identify one routine RF analysis task in your network and draft a natural‑language query you could test with an AI model.
  2. Map the current cost and time spent on manual spectrum interpretation to spot potential savings from automation.
  3. Set up a review process for any AI‑generated RF insights to ensure engineers validate results before action.

Sources

  • middleeastainews.com: Khalifa University opens new RF AI model to telecom operators
  • Zawya: Khalifa University, GSMA and AMD collaborate to bring radio frequency foundation models to telecom operators
  • Mobile Europe: Khalifa University develops AI language model for radio frequency
  • www.wam.ae: Khalifa University launches radio-frequency AI language model ‘RF-GPT’ | Emirates News Agency