Post by : Anis Farhan

Language as Infrastructure

Generative AI has exploded onto the world stage — from ChatGPT and Claude to Gemini and open-source models. But in Southeast Asia, one truth is becoming clear: if your language isn't part of the model, your people aren’t part of the future.

In 2025, countries like Indonesia, Malaysia, Thailand, and Vietnam are rapidly building, fine-tuning, and funding localized large language models (LLMs) — trained in Bahasa, Thai, Vietnamese, and Malay — to ensure AI serves their citizens, not just English-speaking elites.

The rise of regional AI models marks a pivotal shift: language is no longer just cultural identity — it’s digital infrastructure.

The Inclusion Imperative

Southeast Asia is one of the world’s most linguistically diverse regions, with over 1,000 spoken languages. While English remains the digital default, the reality is that over 70% of the region’s population prefers or only understands local languages.

This creates an AI accessibility gap. If LLMs can’t comprehend or generate content in native languages, they fail at tasks from:

• Government service automation

• Education personalization

• Legal translation

• Health information delivery

• Small business e-commerce support

To close this gap, regional tech players, universities, and governments are investing in LLMs trained on local languages, dialects, and cultural context — creating AI that’s not just smart, but culturally fluent.

Indonesia’s Nusantara AI: A National Priority

Indonesia has taken a bold lead with Nusantara AI, a government-backed initiative launched in 2024 to develop a Bahasa Indonesia LLM for national use.

In early 2025, Nusantara AI reached its third phase:

• Trained on over 20 billion tokens of Bahasa text

• Fine-tuned for legal, educational, and public service use cases

• Deployed in ministries for auto-translation, citizen chatbots, and policy drafting

Built in partnership with local universities and cloud providers, Nusantara AI is now being tested in rural Java and Sumatra for healthcare delivery — translating diagnostic content into local dialects using a combination of NLU (natural language understanding) and text-to-speech AI.

It’s not just a tech project — it’s digital nation-building.

Thailand and Vietnam: Cultural AI in Action

Thailand has launched SiamGPT, an open-source Thai language LLM developed by researchers at Chulalongkorn University. It’s trained on a combination of government records, media archives, and Buddhist texts, making it uniquely attuned to Thai syntax, honorifics, and social cues.

SiamGPT is already being piloted in:

• Court document translation

• Tourism chatbots for rural destinations

• AI-powered Thai history education platforms

In Vietnam, the Ministry of Information and Communications is backing VietLM, a bilingual LLM (Vietnamese–English) to support digital governance and AI-assisted SME tools.

Local voice tech startup Zalo AI has layered its own voice synthesis model on VietLM to create hyperlocal voice assistants for farmers and small traders in the Mekong Delta.

These projects are about more than efficiency — they aim to preserve language, improve digital equity, and make AI reflect Southeast Asian realities.

Malaysia’s Multilingual Mandate

Malaysia, a tri-lingual nation (Malay, English, Mandarin), faces a different challenge: code-switching AI. Its LLM strategy, led by national AI agency MyAI, focuses on training models that understand mixed-language queries — where a user switches between languages mid-sentence.

In 2025, MyAI partnered with Google Cloud and Universiti Malaya to launch MalayaGPT, a multilingual LLM now used in:

• Government digital services

• Islamic finance advisories

• Cultural preservation archives

• Content moderation for social platforms

Malaysia’s approach could serve as a template for other multilingual societies navigating the complexities of natural language processing in mixed-language environments.

Beyond Language: Local Values and Norms

Localization isn’t just about words. It’s about values, metaphors, taboos, and inference patterns. Southeast Asian AI developers are now embedding cultural layers into models:

• Politeness and honorifics

• Family and community structures

• Religious sensitivity

• Traditional beliefs and folk medicine references

Without these, AI outputs may seem “correct” linguistically but still alienate users or produce culturally inappropriate responses. Local grounding is not a luxury — it’s the cost of trust.

The Regional Race: Who Will Lead?

ASEAN’s decentralized structure means countries are taking different paths:

• Indonesia is going national-first, focusing on government deployment

• Thailand is pushing open-source and education

• Vietnam is embedding AI into trade and agriculture

• Malaysia is aiming for cross-sector adaptability

There’s growing momentum to create a shared ASEAN AI framework, with joint funding, ethical guidelines, and cross-border model sharing. Talks are underway for an ASEAN AI Supercomputing Center, backed by Singapore, to pool compute resources for smaller nations.

Challenges Ahead

Localizing generative AI in Southeast Asia is not without obstacles:

• Data scarcity for minority languages and dialects

• Talent shortages in deep learning and NLP

• Bias and misinformation risk from culturally sensitive topics

• Regulatory gaps on AI governance and content control

But the commitment is strong, and the stakes are high. Without localization, Southeast Asia risks becoming a passive consumer of Western-trained AI — locked out of its own digital future.

Conclusion: Speak the Language of the People

In 2025, the AI conversation is finally shifting from global dominance to regional relevance. For Southeast Asia, that means building LLMs that don’t just understand English — but understand context, nuance, and the lived experience of millions.

Generative AI may be the brain of tomorrow’s internet. But for Southeast Asia, language is its soul.

Disclaimer

This article is intended for editorial and informational purposes only. It does not constitute technical advice or policy guidance. Readers should consult with language technology and AI governance experts for implementation strategies and ethical compliance.