Country Landscape of Large Language Models Development: A Review
Wan Chong Choi, Chi In Chang, Iek Chong Choi, Lai Chu Lam · Preprints.org · 2025
This paper provided a comprehensive comparative analysis of national strategies in developing large language models (LLMs), drawing on literature from 2020 to April 2025. The study examined how different countries approached the design, training, and deployment of LLMs, revealing distinct trajectories shaped by computational infrastructure, policy orientation, and linguistic objectives. The United States maintained a leadership position through models such as ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), and LLaMA (Meta), supported by robust private-sector innovation and integrated cloud ecosystems. China, despite hardware constraints, advanced rapidly through efficient and open models including ERNIE (Baidu), Tongyi Qianwen (Alibaba), PanGu (Huawei), and DeepSeek, emphasizing cost-effective scaling and bilingual capabilities. The European Union pursued openness and multilingual inclusivity with models such as BLOOM (France), Luminous (Germany), and OpenGPT-X, aligned with emerging AI regulatory standards. Other regions demonstrated targeted advancements: India introduced BharatGPT and Airavata to support its linguistic diversity; the UAE developed Falcon and Jais to lead Arabic AI; South Korea produced HyperCLOVA; and Japan contributed models like Nekomata and EvoLLM-JP. These developments illustrated that LLMs have become strategic assets reflecting broader national priorities, from technological sovereignty to regulatory alignment. The study concluded that while model architecture remains central, the evolution of LLMs is increasingly determined by geopolitical, infrastructural, and socio-linguistic factors that shape their integration into national digital ecosystems.