Opening Address by SMS Tan Kiat How at Tech Week Singapore

1. A very good morning, everyone.

2. Welcome to Tech Week, and to friends from overseas – a very warm welcome to Singapore.

3. It's wonderful to see the crowd and energy here at TechWeek. We’ve already seen the event grow from strength to strength every year, and it’s wonderful to see familiar faces as well as new friends coming to join TechWeek here in Singapore.

4. It has been an eventful few weeks for technology and geopolitics.

5. President Trump and President Xi met in Washington last week. Artificial intelligence featured prominently in the discussions of the two world leaders.

6. That tells us something important. AI is no longer simply a technology story; it is an economic story, an infrastructure story, and increasingly, a strategic story.

7. That makes this year's theme, the Infrastructure Era, especially timely.

8. Over the past few years, the world has invested enormously in infrastructure for AI. GPUs. Data centres. Networks. Energy. Cloud infrastructure. All of this matters.

9. But as AI becomes more capable, we need to broaden our understanding of what infrastructure means.

10. Because having enough compute does not mean AI will be adopted. A powerful model does not mean it will work inside a hospital, bank or factory. And an AI agent capable of taking action does not mean an organisation will trust it to do so.

11. The next infrastructure challenge is not simply to build more. It is to build the infrastructure around AI.

12. The systems that enable AI to move from the model into the real world - and from experimentation to scale.

13. I often use an analogy. Think of traffic on the road. You have a very capable car, with a very powerful engine inside the body of the car that can get from zero to 100 in just seconds.

14. But if you think of infrastructure as just building more roads or more lanes on the road, that doesn't mean traffic can go smoothly. It doesn't enable the system to allow different cars and vehicles on the road to get from point A to point B safely and seamlessly, while ensuring that people can trust the whole system.

15. So, infrastructure goes beyond just building more roads and more lanes but thinking about the software aspects that wrap around that hard infrastructure – the rules of the road or who has the right of way.

16. The software and the platforms that optimise traffic, doing traffic monitoring, and controlling the traffic lights.

17. And when incidents happen -- how to remediate, resolve, and get traffic moving smoothly again, so that everyone can safely drive on the road and have confidence in getting from one point to another safely and responsibly.

18. So, infrastructure is not just about building roads but thinking about the whole system.

Infrastructure for AI

19. For infrastructure around AI, I think about this in three layers.

20. The first layer is the infrastructure we already know: compute, data centres, connectivity and energy. Demand for all of these will continue to grow as AI becomes more deeply embedded in businesses and essential services.

21. For Singapore, the challenge is particularly acute. We are small. Resources like land, energy and water are constrained. So, simply building more cannot be our strategy. We must make every unit of infrastructure work harder for us.

22. More efficient compute. More resilient networks. More sustainable data centres. And infrastructure that supports high-value economic activity.

23. This is why Singapore has taken a deliberate approach to digital infrastructure growth. We plan ahead, invest early, and balance growth with the constraints of a small city.

24. The Digital Infrastructure Bill, the new legislation, is part of this forward-looking approach. I will be moving the Bill in Parliament next week. It proposes clearer expectations for the security and resilience of major cloud services and data centres, while raising sustainability standards across the entire data centre sector.

25. We should not wait for infrastructure to become critical before we decide how we want it to operate. We should plan ahead.

26. And this disciplined and deliberate approach gives us the headroom to grow sustainably. So, we're not saying no to growth, but thinking about how to do so in a sustainable, disciplined, and deliberate manner.

27. But physical infrastructure is only the foundation. A GPU creates economic value only when somebody can put the intelligence it produces to work.

Infrastructure to deploy AI

28. That requires a second layer.

29. When an AI application moves from demonstration into production, it needs more than compute. It needs access to data. It needs to connect with existing systems. It needs identity and permissions. It needs cybersecurity. And it needs to work within an actual business process.

30. Increasingly, AI agents will also need to interact with other systems - and with one another. This layer is less visible than a data centre, but it may become just as important. It is what allows intelligence to move from the model into the economy.

31. We should not require every company, hospital or government agency to rebuild the same foundations from scratch.

32. I wear another hat in healthcare, not just at the Ministry of Digital Development and Information. I also have a role at the Ministry of Health.

33. In healthcare, I pay a lot of attention to how technology, especially AI, is being used in our healthcare system to deliver better care for our Singaporeans.

34. For example, shared platforms such as HEALIX and Tandem provide common foundations for public healthcare institutions to work with trusted data, and to develop and deploy AI applications, rather than having each hospital or healthcare institution build these same capabilities from scratch.

35. The principle is broader than healthcare. Where it makes sense, we should have common platforms, common standards and common infrastructure.

36. So, the approach we take here in Singapore is to centralise what should be common and decentralise what should be innovative.

37. That is how we make deployment faster - and make the infrastructure we build more productive.

Infrastructure to trust AI

38. But even that is not enough. As AI becomes more capable, and especially as it becomes more agentic, we need confidence not just in what it can do, but in what it is allowed to do. We must ensure that it is developed and used safely and responsibly.

39. Just last week, Singapore joined countries around the world in calling for stronger safeguards and international cooperation on frontier AI. But these principles must also be translated into practical safeguards for how AI operates in the real world.

40. Imagine an AI system that can plan, access systems, call other tools, make purchases or move information.

41. The question is no longer simply, “Is the model accurate?” We also need to ask: “What can it access? Who authorised it? Can we see what it has done? Can we intervene? And if something goes wrong, who is accountable?” These are infrastructure questions too.

42. Trust allows us to deploy. That means evaluation, testing, cybersecurity, identity and authorisation, provenance, auditability, assurance in real-world applications, and monitoring after deployment.

43. We are already building parts of this in Singapore. Using another example in healthcare, AimSG is a shared national imaging AI platform that allows public hospitals and technology partners to develop, validate, deploy and monitor imaging AI models, with governance built in from the start.

44. That is an example of trust moving from a principle into infrastructure.

45. A hospital will not rely on an AI system for consequential decisions if it cannot establish that the system is safe. A bank will not allow an AI agent to make a transaction if it cannot establish who authorised it.

46. Trust is not the brake on innovation. It is what gives us the confidence to scale.

From model assurance to system assurance

47. This also changes what we need to evaluate.

48. In the earlier phase of AI, much of our attention focused on the model - how accurate it is, how robust it is, how frontier it is, how many billions or trillions of parameters it has, and what its limitations are. Those questions remain important.

49. But increasingly, the model is not the whole system.

50. Consider healthcare again, as it is applicable to many other industries. A model can perform well on a benchmark. But what happens when it meets a clinician, a patient, hospital data, an existing workflow and a real decision that it has to support?

51. The outcome depends on the whole system: human + AI + workflow + organisation.

52. So, assurance needs to move from “Is this a good model?” to “Does this AI-enabled system work safely and effectively in the real world?”

53. This is an area where Singapore can contribute. We have sophisticated users, strong digital foundations, and researchers, companies and regulators who can work closely together.

54. Our opportunity is not merely to test AI here. It is to develop practical ways of evaluating and assuring AI in real-world deployments. If a system works safely and effectively in a demanding real-world situation, that experience can travel with the technology and scale beyond Singapore.

Infrastructure must also connect

55. There is one further challenge when I talk about infrastructure. Realistically, the world may not converge on a single AI ecosystem.

56. We may have different models, different clouds, open and closed systems, countries making different infrastructure choices, different tech stacks and different approaches to data, identity and governance.

57. Different AI ecosystems may compete, but at the end of the day, they will still need to work together. As agents from different companies interact, infrastructure must also be interoperable.

58. We are already seeing this in payments, where the industry is working on ways for payment ecosystems to recognise trusted AI agents acting on behalf of users.

59. There is a sandbox that the Monetary Authority of Singapore, Singapore’s financial regulator, has put in place. Key major international payment players come together on the Know Your Agent interoperability framework.

60. We are not asking for each of the tech ecosystems around those payments to be uniform, or to have one rulebook. It's not practical. But how do we ensure interoperability, even though there are different systems?

61. The underlying questions are simple: Who is this agent? Who does it represent? What is it authorised to do?

62. For decades, our financial systems have built ways for people and institutions to establish trust with one another. The agentic economy will increasingly need similar mechanisms between machines acting on behalf of people and organisations.

63. That is the next frontier for interoperability between infrastructure, and it is where Singapore can contribute too. We are deeply connected to global technology ecosystems but also embedded in Southeast Asia. We can work with different systems where it is useful, practical and trusted.

64. In Singapore, we cannot build every model or every technology stack. But we can become very good at helping different systems connect, deploy, earn trust and scale.

Build for change, not just for today

65. Ladies and gentlemen, I’d like to make a final point. We cannot know exactly what infrastructure we will need five or 10 years from now.

66. The technology will change. The models will change. The way companies use AI will change.

67. Our task is not to predict the future perfectly. It is to build foundations that can adapt as the future changes.

68. That means close partnerships between government, industry and research. Industry needs to tell us where demand is heading. Researchers need to help us understand what is technically possible and where new risks may emerge. Government need to plan ahead and create the conditions for investment and innovation. And businesses need to continue investing in security, resilience and capability as they transform.

69. Singapore’s small size can help here. We can bring people together, test ideas in real environments, and learn quickly. We can also work with leading technology companies from around the world.

70. Our Singapore National AI Missions are one example, bringing companies, researchers and government agencies together around real problems in healthcare, finance, connectivity and advanced manufacturing.

71. Our objective is not to build everything ourselves. It is to bring together the right capabilities, build the right foundations, and create the conditions for them to work together.

72. What we bring to the table is not scale. It is the ability to connect capabilities: strong digital infrastructure, trusted institutions, sophisticated users, research and engineering talent, and close links to regional markets.

73. We can bring these pieces together, test what works, and scale beyond Singapore.

The Infrastructure Era

74. So, this is what I think the Infrastructure Era is really about. It is not simply about building more. It is about building what we need for a world in which AI becomes increasingly embedded in our economy and society.

75. Physical infrastructure gives AI power. Deployment infrastructure puts AI to work. Trust infrastructure gives people confidence to use it. And interoperability allows these systems to work across organisations and across borders.

76. Infrastructure is no longer simply a backdrop for the digital economy. It is part of the capability of the economy itself.

77. We cannot predict exactly what technology will come next. But we can make sure that when it arrives, we are ready to use it, ready to deploy it, ready to trust it, and ready to scale what works.

78. Thank you.