Who’s building AI for the way Southeast Asia actually speaks?

Southeast Asia's approach to artificial intelligence (AI) has been shaped by a shift towards native language usage, according to Google's Gemini Southeast Asia report published in July. Nearly 70% of prompts in the region are now submitted in native languages, with 89% in Vietnam, 87% in Thailand, and 84% in Indonesia. The trend is evident across all four countries, with three-quarters of requests coming from mobile devices and over 40% arriving as voice, photos, or video.

However, the tools that people are interacting with have not been built to accommodate this shift. Most frontier AI systems are English-first by construction, which presents a challenge for a region where most people communicate in their native languages. The region's demographic makeup, characterized by a young population that grew up using phones rather than desktop computers, has expedited this change.

People are increasingly using tools that respond in their native language or listen to them, rather than translating their thoughts into English.

Several AI companies are responding to this shift. AI Singapore's SEA-LION covers eleven regional languages, with its latest version being one of the strongest open models available and capable of running on a laptop. AI Singapore has also launched SEALD, a project focused on assembling the Southeast Asian language data necessary for these systems to function effectively.

Indonesia has taken the lead in this effort, with Sahabat-AI, a national project launched in late 2024 by Indosat Ooredoo Hutchison and GoTo. Sahabat-AI supports Bahasa Indonesia, Javanese, Sundanese, Balinese, and Bataknese, aiming to provide digital sovereignty and cultural preservation. Thailand's leading effort, Typhoon, comes from the venture arm of Siam Commercial Bank, SCB 10X, and is open for use in various sectors, including consumer finance, legal services, healthcare, and education.

Vietnam's approach has been less successful, with VinAI's PhoGPT and ViGPT models experiencing a setback after Qualcomm acquired VinAI's research and generative AI teams, causing the public model repository to stop being updated. Meanwhile, the Philippines and Malaysia largely rely on regional models rather than national ones.

The unevenness in AI development across Southeast Asia highlights the challenges and opportunities in creating AI systems that cater to the region's linguistic diversity. While Indonesia and Thailand have made significant strides in building domestic AI models, Vietnam's efforts have stalled due to talent loss, and the Philippines and Malaysia rely more on regional models.

The disparities in AI development could have significant implications for the region's digital future, particularly for languages like Taglish and Manglish, which pose challenges for general-purpose models.