[DigitalToday reporter Jinju Hong] An assessment said the next competitive edge for large language models (LLMs) depends more on multilingual design and reflecting local context than on model size or computing resources.

On April 9 (local time), IT outlet TechRadar reported that today's base-model structure, built around English, is showing limits as AI spreads globally, and that multilingual understanding is emerging as a core condition for sovereign AI.

In the early generative AI market, an English-first structure effectively became the standard. Public training data was concentrated on the English-language internet, and early model development also took place in regions where English is central to digital communication. But as companies and governments move to fully adopt AI across the economy and administration, this structural bias is being cited as a problem.

The key point is that simple multilingual support is different from true multilingual understanding. Many widely used LLMs can technically process multiple languages, but in many cases they remain at the level of translating English knowledge. The outlet said the difference matters, noting that language is not just a means of communication but carries culture and context, social nuance and local knowledge systems.

The limitation is more evident in global markets. Customer service, finance, healthcare and public services rely heavily on understanding local language variations and context. If AI fails to interpret that properly, accuracy can fall, adoption can be constrained and trust can also decline. As a result, demands are growing for AI to go beyond translation and be able to reason within each language structure.

As a result, demands are growing for next-generation base models to go beyond a translation-centered approach and be able to reason within each language structure. That would require changing the design philosophy itself, beyond simply increasing the number of supported languages. It also called for training data to cover local languages and dialects, and for academia, government and industry to work together to build high-quality datasets.

Some also argue that model architecture must evolve to handle multiple language systems efficiently through approaches such as mixture-of-experts structures, specialized tokenization strategies and language-specific reasoning pathways. They also say evaluation standards need to be redesigned beyond English-centered tasks to measure reasoning, contextual understanding and cultural suitability in multilingual environments.

The trend is also intertwined with discussions of sovereign AI. Sovereign AI refers to a country's capability to develop, deploy and control AI suited to its own language, culture and regulatory environment. It includes control of data infrastructure, alignment with national regulatory frameworks and fostering domestic innovation ecosystems. In sectors handling sensitive data such as finance, healthcare and public services, demands are growing for data storage location and governance.

It was also presented as a backdrop that governments are starting to view AI as a strategic asset that affects economic competitiveness, technological sovereignty and national security. Language representation was cited in this process as a factor that will determine the spread of inclusive AI. Countries with linguistic diversity must design systems so citizens can use AI services in their native languages.

India's digital public infrastructure and AI ecosystem were presented as an example of this trend. India has built a foundation that covers large-scale users based on a digital identity management system, an open financial network and interoperable public platforms. The case shows that open standards, multilingual design and a collaborative ecosystem can become key conditions for AI adoption.

As Europe and Britain also pursue their own sovereign AI strategies, open infrastructure, multilingual capabilities and a collaborative ecosystem are likely to become more important factors. The outlet said, "The future of AI is not determined only by model size or the amount of training data," adding, "Organizations and countries that can design systems that work across diverse languages, cultures and regulatory environments will hold the advantage."