August 29, 2025 - AI hardware startup Cerebras Systems, in partnership with UAE technology group Core42, has achieved a remarkable milestone in large-scale language model training, successfully developing a 180-billion-parameter Arabic language model in under 14 days. The project utilised an unprecedented array of 4,096 Cerebras CS-3 wafer-scale accelerators operating in parallel, demonstrating new possibilities for rapid, large-scale AI development.
The training achievement represents a significant leap forward in AI infrastructure capabilities and training efficiency. Historically, language models exceeding 100 billion parameters required weeks of training time on cutting-edge supercomputers, making such projects accessible only to the largest technology companies. The resulting Arabic language model is designed to handle both Arabic and multilingual tasks, addressing a growing demand for AI systems that serve non-English speaking populations. According to Solutions Review, this project showcases "an unprecedented level of scaling" that could fundamentally change how quickly organisations can develop specialised AI models.
The breakthrough highlights the growing global interest in developing region-specific AI capabilities, particularly as governments and organisations seek AI systems that understand local languages, cultures, and contexts. This trend towards AI sovereignty reflects broader geopolitical considerations around technological independence and the recognition that English-centric models may not adequately serve diverse global populations. The rapid training timeline also suggests that specialized AI development could become more accessible to nations and organisations with specific linguistic or cultural requirements.
Our view: Cerebras and Core42's achievement demonstrates how specialised hardware can democratise large-scale AI development beyond the traditional technology giants. The focus on Arabic language capabilities reflects a necessary shift towards more inclusive AI that serves global populations rather than reinforcing existing linguistic biases. This development could catalyse similar projects worldwide, potentially leading to a more diverse and representative AI ecosystem where different cultures and languages receive adequate technological representation.
beFirstComment