Singapore has no shortage of artificial intelligence strategies, funding schemes or research programmes. The harder question is whether these efforts are changing how companies, public agencies and industries actually work.

A visit to Beijing in July for the 2026 Global Digital Economy Conference put this issue into sharper relief. We had expected a familiar spectacle: bigger models, faster chips and more impressive demonstrations. But at the China National Convention Centre, what caught our attention was not simply what AI could generate. It was where AI was being placed – inside factories, security systems, transport networks, healthcare and city management.

At 360 Digital Security Group, the conversation quickly moved away from the novelty of generative AI towards a more practical concern: how should an organisation monitor, audit and contain an AI system once it is connected to sensitive data?

One company used AI agents as tools for customer service and decision-making. At a humanoid robotics training base, the important feature was not merely that a robot could walk. It was the work around it: training, testing, engineering adjustments and industrial support. During our visit to Xiaomi’s smart manufacturing and electric vehicle facilities, we saw a similar effort to connect hardware, software, data, manufacturing and commercial applications into one system.

What we observed was a clear shift in emphasis. No, China is not abandoning the race to build advanced models. Its focus is on making those models useful in the daily operations of the economy. It is moving from impressive demonstrations to effective deployment.

This is the transition Singapore needs to make.

Singapore has identified advanced manufacturing, financial services, connectivity and healthcare as national AI mission areas. The National AI Impact Programme aims to support 10,000 enterprises over three years, while more than $1 billion has been committed to public AI research and talent development between 2025 and 2030.

The problem is not a lack of ambition. It is whether policy support can carry companies beyond experimentation and into their core operations.

A Ministry of Manpower survey, covering private-sector establishments with at least 10 employees, found that 71.5 per cent of firms had yet to adopt AI. Of those that had started, only 3.8 per cent had integrated AI into core processes. Another 6 per cent were still piloting and 7.4 per cent were at the planning stage.

The broader message is difficult to miss: digitalisation is widespread in Singapore, while deep AI integration remains narrow.

A small and medium-sized enterprise (SME) that uses AI to draft social media posts or summarise documents has not necessarily transformed its business. The real test is whether AI changes inventory management, production scheduling, customer service, risk assessment, product design or decision-making.

Every publicly supported AI project should therefore answer three simple questions: Which core process will change? What result will be measured? When will the project enter regular operations?

Public support should not end when a pilot is completed. It should help companies prepare their data, redesign workflows, integrate systems, strengthen cybersecurity and train workers to use the technology effectively.

Singapore’s Digital Leaders Accelerator Bootcamp is moving in this direction by requiring companies to identify real business problems and develop practical transformation road maps. The national AI missions should push further.

They should not become another collection of themes, conferences and demonstrations. Each mission should have a difficult problem to solve, a real industry operator to work with, a production system to deliver and a clear path for scaling beyond the first organisation.

This is especially important for SMEs, which need technical expertise and integration capacity. Subsidising a software licence will not solve all of these problems. SMEs need specific solutions and support that continues after the first pilot.

Singapore will not win by matching the biggest AI powers dollar for dollar. We do not have their market size, computing capacity or capital resources. Our opportunity lies elsewhere: making AI deployment trusted, auditable and interoperable across borders.

Singapore has already built important foundations, including AI Verify, the Global AI Assurance Sandbox and a governance framework for agentic AI. It is also contributing to international testing standards and leading the ASEAN Working Group on AI Governance. But these initiatives have yet to become a complete regional pathway, from testing to deployment.

A company using AI across several Asian markets still needs to determine which tests regulators will recognise, which governance documents they will accept, how cross-border data should be handled and who will be responsible when an AI system behaves unexpectedly.

Singapore should become the place where these questions can be answered. It should offer a usable service that brings together testing, governance, data protection, cybersecurity and sectoral regulation.

A regional AI testing and assurance platform could help companies assess sector-specific risks, validate governance controls, structure cross-border data arrangements and prepare evidence that can be recognised in different markets.

Regulatory sandboxes also need clearer exit routes. A successful sandbox should not end with a report or a conference presentation. It should lead to actual deployment, procurement approval, monitoring and, where possible, mutual recognition by other jurisdictions.

Trust should reduce uncertainty and transaction costs. If it simply creates another layer of paperwork, it will not become a competitive advantage.

This requires a more integrated approach to regulation. In Beijing, industrial policy, data security, cybersecurity, standards and AI applications often appeared within the same operational conversation.

Singapore must also change how it measures success. Research grants, training places, AI centres of excellence and pilot projects are useful measures of effort. But they do not show whether AI is creating economic value. By the next Budget cycle, the Government should publish an AI deployment scorecard. It should measure the share of firms using AI, the share integrating it into core processes, independently assessed productivity gains and the time taken to move from sandbox testing to actual deployment.

Public services should be included. GovTech’s VICA platform already supports more than 60 government agencies and over 100 chatbots, handling more than 800,000 monthly queries. The next question is not simply how many tools are operating, but whether they reduce response times, improve accuracy and allow public officers to focus on work that requires human judgment.

Training targets are also only inputs. The Government should track how many workers are using AI in redesigned workflows, and whether those changes improve the quality and value of their jobs.

Our visit to Beijing challenged three assumptions: that AI capability will automatically spread from research institutions to ordinary businesses; that governance can be treated as a compliance layer after deployment; and that the number of pilots is a reliable measure of progress.

Singapore does have an AI strategy. What it needs now is a deployment plan. That plan should turn grants into measurable adoption, turn regulation into usable trust infrastructure and turn national missions into production systems.

Singapore’s AI success will ultimately be judged not by how many pilots it funds, but by how much of the economy those pilots have changed.