AI Creators
Mistral AI
Building custom code models for Ericsson proprietary silicon | AI Now Summit 2026
Jul 2, 2026
Key Points
- 1.Ericsson collaborates with Mistral AI to develop custom AI models for proprietary silicon.
- 2.The challenge lies in creating sufficient training data for effective model fine-tuning.
- 3.Key objectives include knowledge transfer, strategic autonomy, and enhanced R&D for network development.
Summary
Collaboration Objectives
The collaboration between Ericsson and Mistral aims to produce a state-of-the-art language model that comprehensively understands Ericsson's proprietary silicon. With this model, they aim to facilitate product development within 5G and future 6G networks.
Challenges in Model Training
One major challenge the teams faced was the creation of adequate training data, as the domain of telecommunications often lacks abundant data compared to other industries. Fine-tuning models in this context requires substantial effort and expertise.
Knowledge Transfer
The partnership emphasizes mutual knowledge transfer, where Mistral brings AI training expertise while Ericsson contributes domain knowledge from years of experience. This exchange is critical for developing tailored models and workflows specific to Ericsson's needs.
Strategic Autonomy
A significant goal for Mistral is to support Ericsson in achieving strategic autonomy, reducing reliance on external providers. This focus helps ensure that Ericsson can leverage its proprietary data effectively and securely.
Creation of Custom Tools
The intent is not to develop a general-purpose AI model, but rather a specialized tool that optimally integrates knowledge about Ericsson's ASICs. By institutionalizing expert knowledge within an LLM, they intend to accelerate development and training for their engineers.
Worth watching for
This video is aimed at industry professionals and researchers interested in AI applications in telecommunications, specifically within network infrastructure and custom model training.