What is AISG: Singapore's AI Execution Engine

AISG’s full name is AI Singapore, established in 2017 by the Singapore government with funding from the National Research Foundation (NRF). The initial plan was up to S$150 million over five years; it was later extended to 2027, with cumulative funding exceeding S$500 million. This site’s AI Singapore archive has a complete timeline; here I’ll only cover the essentials.

Its work is divided into four domains: research, talent, products, and governance. Most national AI programmes undertake only one of these, whereas AISG pursues all four — this is what sets it apart from most comparable institutions.

Research relies on SEA-LION. SEA-LION is a large language model for 11 Southeast Asian languages, with v3 reaching up to 70B parameters and v4 covering 4B–33B. The approach is continued pre-training: from v3 onwards, it continues training based on Llama 3 and Gemma, adapting Western foundation models to achieve semantic fidelity for Southeast Asian languages. The accompanying SEA-HELM evaluation suite validates this approach: on minority-language tasks, it outperforms same-sized Llama 3, Gemma, and Qwen.

Talent development relies on the AIAP apprenticeship programme. Twenty-two cohorts have graduated approximately 500–600 people, with each cohort comprising about 60 apprentices; roughly one-third come from NUS. The youth pathway is the NOAI National AI Olympiad. Singapore’s team won 2 golds at IOAI 2024 in Bulgaria, 2 golds and 5 silvers at the 2025 Beijing edition, and will co-host IOAI in 2027 with NTU.

Product lines include TagUI (open-source RPA), PeekingDuck (computer vision), and SGNLP (natural language tools). The governance line is AI Verify, the world’s first open-source and testable AI governance toolkit, with the framework from IMDA and AISG deeply involved in engineering.

Two names are often incorrectly attributed to AISG. AI Trailblazers is a project of MCI, DISG, SNDGO, and Google Cloud, launched in July 2023 and expanded to version 2.0 in January 2024. Kampong AI is a precinct by JTC at the one-north LaunchPad, piloted in March 2026 and completing in 2028, spanning 14,500 square metres and accommodating approximately 70 companies. The former is a government-cloud-provider partnership; the latter is real estate development — both are unrelated to AISG.

100E (100 Experiments) was AISG’s largest deployment experiment, running 100+ AI projects from 2018 to 2025, now archived. Archival signals that AISG’s phase-level mission has been accomplished, while also signalling that the next deployment pathway has yet to be named.

Among the seven policy levers on this site, AISG operates on three: Lever 3 (Talent), Lever 5 (Government Self-Use), and Lever 6 (Diplomacy). Government self-use relies on SEA-LION deployment for localized services. Diplomacy leverages SEA-LION and AI Verify as technical calling cards, and AISG has been present at all three AI Safety Summits in Bletchley, Seoul, and Paris. For the current state of the talent pipeline, Talent Dimension contains complete data.

SEA-LION’s value is not in benchmark scores. It demonstrates to Southeast Asian nations that they can have their own LLM foundation model and need not rely solely on OpenAI. Subsequent initiatives by Indonesia, Malaysia, and Vietnam have all followed this pathway.

The bottlenecks are equally clear. AIAP takes 60 people per cohort, capacity is limited, and expansion is constrained by project-based budgeting. Top AI PhDs are lost to the United States and industry. 100E is now archived, and the next-generation deployment experiment is still in progress. Participation from universities beyond NUS remains low; AISG essentially revolves around NUS.

Most of the Singapore AI strategy narrative is written into policy documents. AISG is the hand that writes it into products and talent.