How is Singapore’s $120 million national AI for Science (AI4S) programme, announced through eight research projects in June 2026, using artificial intelligence to accelerate scientific discovery and translate laboratory findings into real world technologies—including a $10 millio

Singapore’s AI4S is a S$120 million National Research Foundation programme designed to make AI part of the scientific workflow—from choosing hypotheses and running experiments to verifying software and producing deployable technologies. Its first eight projects, launched on 16 June 2026, pair AI specialists with domain scientists and international partners; their results are prospective research targets, not yet demonstrated commercial outcomes.

• Materials: closing the prediction-to-production gap. The S$10 million NUS–University of Toronto Materials Data Foundry combines AI, robotics and in-situ measurement in an open autonomous lab. Instead of training models only on published materials properties, it will generate experimental data connecting synthesis conditions to real-world performance—targeting tens of thousands of reactions, synthesizable candidates and open software tools. This is intended to yield reproducible “recipes,” not merely AI-predicted compounds, for beyond-silicon/quantum materials, durable electrocatalysts relevant to clean hydrogen, and corrosion-resistant coatings.

• Catalysts: search a vast design space computationally, then test the best options. A*STAR and Imperial’s S$10 million programme will build a Surface Science Foundation Model and cloud platform to predict catalyst properties and synthesis conditions for cleaner fuels, CO₂-derived feedstocks and bio-based chemicals. The primary programme description states a target 500-fold acceleration in computational screening; claims of “nearly five times faster and cheaper” should therefore be treated cautiously unless they refer to a separate downstream development metric.

• Trustworthy AI-generated code. NUS and Imperial are developing program-reasoning tools that automatically analyse, test, formally verify and prove software correctness, while using less-formal reasoning to interpret undocumented code. They will test the tools on consequential software such as network protocols and Linux-kernel components, with the aim of enabling AI agents to audit code written by other AI systems.

• Biomedicine: integrate biological data rather than treat each layer separately. NUS and A*STAR’s MultiOmicsFM is planned as a foundation model that jointly interprets DNA, RNA and gene-activity data from Singapore’s multi-ethnic datasets, supporting disease-risk research and optimisation of mRNA therapies. Separately, the BloodCounts! project will train a DeepCBC model on the full raw parameters of complete blood-count tests plus blood-smear images, to extract more diagnostic value from a common test and support prediction of conditions such as stroke and cancer risk.

• Food and climate resilience. NUS and Illinois ARCS will build knowledge-guided agricultural digital twins—virtual representations of Southeast Asian farmland that combine observed data with crop-science principles. The intended outputs are interpretable forecasts and decision tools for planting, resource use and supply chains, rather than opaque predictions alone.

• Other discovery domains. The eight-project portfolio also includes AI-designed interfaces to improve energy transport, alongside the materials, catalysis, genomics, software, agriculture and blood-analysis efforts. Together, these projects span materials, computing, life science, energy and agriculture rather than treating AI4S as a single-sector initiative.

Strategic logic

• Translation: AI proposes promising materials, molecules, experiments or code; automated labs, high-throughput screening, measurements and formal verification supply the real-world evidence needed to decide what is manufacturable, safe and useful. The Materials Data Foundry is the clearest embodiment of that “closed loop.”

• Talent: AI4S explicitly aims to train “bilingual” scientists—researchers fluent both in AI and in fields such as materials science, life sciences and quantum science—so that domain questions, data quality and experimental constraints shape the models from the outset.

• Global positioning: Singapore is using NRF funding, CREATE-based international collaboration, shared digital/experimental infrastructure, open tools and industrial partners to make itself a convening site for AI-driven research and translation. Imperial’s two programmes, for example, are hosted through Imperial Global Singapore at CREATE and are designed for worldwide researcher access.

The central test will be whether these projects produce validated, scalable outputs—manufacturing protocols, catalysts, clinical decision-support evidence and reliable software—not simply faster model predictions.