Governments are spending billions on their own 'sovereign' AI technologies (www.theguardian.com)

Governments from Singapore to Malaysia, Switzerland and India are investing in “sovereign” AI — local large language models tailored to national languages, cultures and security needs. Examples include Singapore’s SEA-LION family (conversational in 11 Southeast Asian languages), Malaysia’s ILMUchat (disambiguates local place names) and Switzerland’s Apertus (handles Swiss‑German orthography). These projects aim to fill gaps left by US and Chinese models that often mishandle regional languages, cultural context or raise data‑sovereignty and national‑security concerns. The broader backdrop is a global AI arms race worth hundreds of billions of dollars, while middle powers try to carve out viable, lower‑cost alternatives. Technically and strategically, smaller states face hard tradeoffs: building frontier models requires massive compute, chips and funding, so many aim for compact, region‑optimized models (some targeting sizes comparable to Mistral releases) that complement—not replace—global giants. Advocates stress cultural adaptation, safer data handling and defence use cases; critics warn of wasted public funds and argue governments might get better returns from regulation, safety frameworks or multinational collaboration. Proposals like an “Airbus for AI” would pool resources across middle powers to compete at scale, while countries such as India are combining modest public investment (~$1.25bn via IndiaAI) with local talent to try to close the gap. The debate now centers on whether targeted, interoperable sovereign models or coordinated regulation and cooperation offer the best path forward.