陳杰豪高階政務部長在 NVIDIA AI Day Singapore 上的演講

Good morning.

We are living through an extraordinary period in technology.

AI models are becoming more capable at remarkable speed. At the same time, companies are investing billions of dollars in GPUs, data centres, networks, energy and the infrastructure needed to support the next generation of AI.

NVIDIA has been at the centre of this build-out. At the same time, I think the next phase raises a different question: Can we turn all this AI capacity into AI value?

The first bottlenecks were technological: Can we build the models, chips and computing infrastructure needed to make advanced AI possible?

Increasingly, another bottleneck is emerging – commercialisation. Technology companies need sustained enterprise demand to support continued investment in AI infrastructure. Enterprises and the government, in turn, need to see meaningful returns before they deploy AI at scale.

The next constraint on AI growth may not be what AI can do. It may be about how quickly organisations can turn what AI can do into economic value.

There is a simple distinction that I find useful.

AI infrastructure gives us capacity .

But capacity is not the same as capability .

And capability is not yet competitive advantage .

To create value, AI has to work with data systems, domain expertise, people and business processes. That means enterprises have to change too.

The question is no longer simply: “Where can we use AI?” A better question is: “What should we redesign, now that AI exists?”

There is a progression here: from tools, to tasks, to workflows, to jobs, and eventually to the organisation itself. The biggest gains may come not from making an individual task 20 per cent faster, but from rethinking how an entire process works when people and AI can divide the work differently.

I understand that Kuangli from Sea (Note: Li Kuangli, Senior Director of Partnership and Strategy, Sea) spoke earlier about Sea’s journey from being digitally natives to becoming AI-natives. Sea is already applying AI across its business lines – Shopee, Monee and Garena. In Shopee, for example, AI is helping sellers create better product listings, improve product discovery and make better-informed decisions — bringing enterprise-grade AI capabilities to SMEs across the region.

What matters is not simply that AI is being used. It is that the way work is done is changing. That is the shift from adoption to transformation.

I have seen so many companies build impressive AI pilots. The harder part comes afterwards. Can it connect to the company’s data? Can it work with existing systems? Is it secure? Is it reliable? Can people trust it? Does the workflow around it need to change? And ultimately, does it create enough value to justify scaling it?

This is why a successful pilot alone is not a successful transformation.

I think there are three transitions to get right.

First: From using technology to solving a real problem. Start with an important business or societal problem, not simply a desire to use AI.

Second: From a promising application to a redesigned workflow. A smarter model inserted into an unchanged workflow may produce only incremental gains. The larger opportunity is to redesign the process around what people and AI can do together.

Third, importantly, where I often see many projects fail: From deployment to demonstrable value. We need a line that connects AI investment to workflow change to measurable outcomes — productivity, revenue, quality, safety, and better services.

Then, we need to learn from deployment and improve again. Because models will change. Costs will change. Customer expectations will change. Lasting advantage will come from the ability to keep learning, changing and adapting.

This commercialisation challenge will become even more important as AI moves beyond the screen.

We are seeing growing investment in robotics, embodied AI and the infrastructure needed to connect intelligence with the physical world.

For Singapore, this matters because many of our hardest constraints are physical – land, labour, energy and an ageing workforce. At the same time, we have sophisticated real-world environments – factories, ports, airports, logistics networks, hospitals and urban systems – where AI can make a tremendous difference.

The first wave of generative AI transformed the interface between humans and information. The next may transform the interface between intelligence and the physical world.

The test remains the same: Does the technology improve productivity, quality, safety or outcomes enough to justify scaling it?

This brings me to Singapore. We cannot outspend the largest economies, and we do not need to.

Our opportunity is to become exceptionally good at moving technologies from possibility, to proof, to scale. We can bring frontier technology together with demanding users, researchers, engineers, regulators, capital and real-world operating environments.

One example is our work with NVIDIA on SEA-LION, Singapore’s open-model family designed for the languages and cultures of Southeast Asia. AI Singapore has incorporated NVIDIA’s models and tools to develop NEMOTRON-SEA-LION-V4.8, improving performance across seven Southeast Asian languages.

This is what a lead market can do. We can adapt technology to the languages, contexts and needs of our region, test it in real settings, learn from use, and build something that can travel beyond Singapore.

This is the model we want to encourage – not technology in isolation, but an ecosystem that helps companies move from experimentation to deployment, and from Singapore into the region.

Our National AI Missions apply the same principle in healthcare, finance, advanced manufacturing and connectivity. The objective is not to create hundreds of demonstrations. It is to discover what matters, develop it with users, deploy it safely, measure the value, learn, and scale what works.

Discover here. Develop here. Deploy here.

Demonstrate value here. Scale from here.

There is also a workforce dimension. As AI takes on increasingly complex tasks, the value people bring shifts towards problem framing, domain expertise, judgement, systems thinking and the ability to work across boundaries.

AI can increasingly generate and debug code and take on other parts of the software-development lifecycle. The value of a technology professional cannot simply be the ability to produce more code. The value is increasingly in applying technology to solve the right problem.

Let me share an example. Ryan, a student at Temasek Polytechnic studying Applied AI, interned with Singapore Airlines’ Base Maintenance Division. He worked on a project to reduce the time employees spent manually checking records in the billing process. Using Natural Language Processing (NLP) and machine learning, together with an understanding of the existing workflow, he developed a pre-vetting tool to streamline some of those checks.

What Ryan learnt was as important as what he built. Technical knowledge was not enough. He had to listen to experienced colleagues, understand operational needs, ask the right questions and adapt his technical knowledge to make something useful. The tool he developed is now being used in the division’s daily operations.

That is why programmes like TIP Alliance+ matter. We want our young people not simply to learn about technology, but to learn by building with it – on real problems, in real organisations, alongside people with deeper experience.

Capability is built by doing. This matters commercially, because organisations that learn faster can capture more value from the same underlying technology.

The AI investment cycle will continue to be enormous. But to sustain it, technological capability has to translate into economic value.

That is the commercialisation challenge we need to unlock.

NVIDIA and companies across this ecosystem are helping to build the infrastructure that makes today’s AI possible. The next challenge is to unlock the applications, workflows and business models that make those investments productive.

Singapore wants to help with that challenge, by creating a place where frontier technologies meet demanding users, where difficult problems can be tackled with researchers and engineers, where deployment is supported by trusted institutions, and where outcomes can be demonstrated in the real world.

Because the advantage in the AI era will not belong simply to those with access to the most powerful intelligence. It will belong to those best able to put that intelligence to work.

I wish all of you a fruitful session. I'd like to thank all of you for being part of this ecosystem. Thank you very much.