Fewer soldiers, smarter systems: Singapore's AI defence imperative.

Singapore's total fertility rate hit 0.87 in 2025. To contextualise what that number means for national defence: every child born in Singapore today is a potential national serviceman or servicewoman in 18 years. A country that consistently produces fewer children than needed to replace its population is, in effect, writing a slow constraint into the size of its future military. With national service as a cornerstone of Singapore's defence architecture, the demographic arithmetic is not abstract. It is a pipeline problem with an 18-year lead time.

The geopolitical context makes this more urgent, not less. Defence Minister Chan Chun Sing, speaking at the Committee of Supply Debate in February 2026, was direct: "What once required advanced and costly military platforms can now sometimes be achieved more quickly and at lower cost by integrating commercial off-the-shelf drone systems with Artificial Intelligence." The threat environment is accelerating faster than the population can keep up. The SAF must find a way to do more with less and AI is the most credible force multiplier available.

01The human-machine equation

Minister Chan has been consistent on one point: AI can and should speed up defence decision-making, but humans must remain in the loop. This is not a philosophical preference, it is a strategic and ethical constraint that shapes how Singapore is building its military AI capability. No government wants an uncontrollable AI-enabled weapons system. Singapore has worked with partners including the Netherlands and the Republic of Korea to establish global standards for military AI governance. The principle is human judgement at the decisive moment, AI everywhere else in the chain.

The SAF is already moving in this direction. The Republic of Singapore Air Force has expanded its partnership with Shield AI to co-develop autonomous drone applications using Shield AI's Hivemind SDK, allowing RSAF engineers and aircrew to independently design, test, and deploy mission autonomy. The Victory-class Multi-Role Combat Vessel will serve as a mothership for unmanned systems. Drone swarms, autonomous surveillance, and AI-assisted mission planning are no longer concepts but are current operational priorities.

And in June 2026, the Defence Science and Technology Agency expanded its collaboration with France's Mistral AI to develop agentic AI systems (autonomous agents capable of reasoning, planning, and executing multi-step workflows in complex defence environments) alongside vision-language models for drone navigation. DSTA Chief Executive Ng Chad-Son framed the ambition clearly: agentic AI that can reason, plan, and act reliably in complex settings will be a key focus. The goal is trustworthy, deployable autonomous systems that strengthen Singapore's long-term capabilities.

02The sovereign AI imperative in defence

The DSTA-Mistral partnership is significant not just for what it enables technologically, but for how it is structured. Defence AI is one domain where the arguments for data sovereignty are not merely economic or regulatory — they are existential. Running classified military data through foreign AI infrastructure is not a risk to be managed. It is a vulnerability to be eliminated.

DSO built a Mixture-of-Experts language model trained on Southeast Asian language data sourced via AI Singapore, combined with defence-specific operational data, deployed on compute provisioned entirely within secure, internet-separated, on-premise environments. The model was not built to be good in general. It was built to be good for Singapore's specific strategic context: understanding the threat landscape, the local operating environment, and the terminology and doctrine of the SAF.

That is the right model for defence AI, and it extends a principle that TrustedHub has held for 25 years: the most valuable AI is the AI that knows your specific context, trained on your specific data, running on infrastructure you control. In defence, that principle is not a competitive differentiator. It is a survival requirement.

03Where the data problem lives

Singapore's defence institutions, like all large organisations that have operated for decades, hold an enormous archive of institutional knowledge in formats that AI cannot directly access. Training records. Tactical after-action reports. Intelligence assessments. Decades of institutional knowledge exists in physical or unstrucutred digital formats that were not designed to be AI-readable.

This is the data problem that precedes defence AI deployment. An agent that helps a commander plan a mission cannot do so effectively without access to relevant historical operational data, current personnel capability assessments, equipment readiness states, or threat intelligence that are all in structured, queryable form. A drone that uses AI-vision to navigate autonomously needs to have been trained on verified, high-quality imagery data specific to its operational environment.

It is only with quality training data that enables AI models to accurately segment and make sense of the contents of a report, an image or audio file. This creates a pipeline of information for AI agents to advise soldiers on what to do in real time. This is what differentiates winning and losing.

But none of this works if the underlying data is locked in files, or legacy systems, or digital formats that AI models cannot interpret. The data foundation always comes before the AI capability.

04What sovereign defence AI actually requires and where TrustedHub fits

Building a genuine sovereign defence AI capability requires the same foundational stack that TrustedHub has spent 25 years building for Singapore's enterprises, but with higher stakes at every step.

Digitisation and certification of records. Decades of training records, operational logs, intelligence images, and institutional documentation need to be converted into structured, machine-readable form. This is not just a scanning exercise, but rather a smart knowledge extraction exercise, governed by strict chain of custody and legal admissibility standards. TrustedHub's Evidence Act–certified digitisation process, which processes documents across text, image, and audio modalities using proprietary AI models, is built precisely for this. Records that are digitised without proper certification cannot be relied upon in any high-stakes environment, let alone a defence one.

Vector-embedded knowledge infrastructure. Once records are digitised, they need to be converted into high-dimensional embeddings that AI models can reason over. A knowledge base built on TrustedHub's Trusted KnowledgeVerse can hold decades of operational intelligence, personnel assessments, equipment data, and doctrine that is queryable in natural language, grounded in verified records and hosted entirely within a secured, air-gapped environment if required.

Agentic workflows on sovereign data. Trusted ActionVerse enables the construction of autonomous AI workflows that run on top of that governed knowledge base. For defence applications (training analytics, equipment maintenance scheduling, personnel capability assessment, mission briefing generation, real-time drone telemetry analysis), the agentic layer is where the productivity multiplier is realised. But only if the data underneath it is clean, structured, verified, and contextually complete.

Local inference on sovereign compute. TrustedHub runs its own inference cluster using open-source models on local hardware, with no external API dependency. This is the architecture defence AI demands: models that can be deployed in secured environments, trained on classified operational data, and operated without any connection to foreign infrastructure.

05The conscription window is an advantage if we use it right

Here is what is underappreciated about Singapore's demographic challenge: the 18-year lead time of the national service pipeline is not just a problem. It is a planning gift that most countries do not have.

Singapore knows, with reasonable precision, how many NSmen will be available in 2035, 2040, and 2043. That certainty creates an obligation to use the window wisely. The technology, training systems, and operational doctrines that need to be in place to compensate for smaller cohorts must be built now, not when the cohorts start to thin. An AI-enabled SAF that can multiply the effectiveness of each serviceman or servicewoman, that can operate drone platforms and autonomous systems with fewer personnel, and that can give commanders faster and more accurate decision support, is the hedge against a demographic future that is already written.

But that AI-enabled SAF needs its data foundations built first. The training simulations need to be captured and structured. The after-action intelligence needs to be made queryable. The equipment maintenance and readiness data needs to flow into systems that can reason about it. The personnel capability profiles need to be digitised, governed, and made accessible to the right analytical tools. None of this is a technology problem at its core. It is a data infrastructure problem and it needs to start now, not after the demographic pressure has already arrived.

Singapore has never had the luxury of size. Its security has always depended on being smarter, faster, and better prepared than larger adversaries. That is the national philosophy that built the SAF into what it is today. AI is the next expression of that philosophy, but only if the data it runs on is as secure, verified, and sovereign as the institutions it serves. The demographic clock is ticking. The technology window is open. The foundations need to be built now.