Scale AI Expands APAC Data Labelling Operations to Address Southeast Asian LLM Data Gap

Scale AI has announced an expansion of its APAC data labelling operations, establishing labelling centres in the Philippines, Vietnam, and Indonesia and growing its APAC annotator workforce — funded by a new capital raise targeting the APAC AI data infrastructure market — with the stated goal of closing the data quality gap between English-dominant LLMs and Southeast Asian language models.

The APAC data labelling expansion is Scale AI's response to a measurable gap in the LLM training data market: publicly available high-quality text corpora for Indonesian, Thai, Vietnamese, Filipino, and Burmese lag English by orders of magnitude in volume and quality. The Common Crawl web corpus — the primary pretraining data source for most open-weight and commercial LLMs — has vastly more English than any Southeast Asian language, producing models where reasoning quality in SEA languages is demonstrably weaker than in English even on directly translated equivalents of the same task.

Scale AI's APAC labelling operation focuses on three data types with the highest return on investment for APAC LLM quality: instruction-following datasets in SEA languages (question-answer pairs, task instructions, assistant responses evaluated for quality by native speakers), domain-specific annotation in APAC vertical markets (legal, financial, medical terminology in SEA languages), and RLHF preference data (native speaker judgements on response quality that incorporate cultural context not captured by English annotators evaluating translated content).

For APAC enterprise teams building LLM applications that must perform well in Southeast Asian languages, Scale AI's APAC expansion increases access to APAC-language fine-tuning data services — reducing the barrier to improving SEA language performance through supervised fine-tuning or RLHF on commercially available base models.