Mr Speaker,
The Prime Minister has set out an ambitious vision for AI. He asks us to harness it as a strategic advantage. I agree with that framing. The question I want to put to this House today is: a strategic advantage for whom? The answer will determine whether Budget 2026 fulfils its promise—that growth must translate into good jobs and rising incomes for Singaporeans.
I want to address three things. How we can own the technology rather than just rent it. How we protect the worker who is being asked to adapt. And how we ensure the gains are shared with the people who helped create them.
Part one: From fast adopter to global exporter—Singapore’s AI industrial strategy
Let me start with the technology itself.
Mr Speaker, I want to push our AI ambition further. Right now the heavy fiscal levers in our AI strategy are largely weighted towards adoption—identifying where AI can improve our existing industries and subsidising our businesses to deploy it faster. That is necessary, but not sufficient. If our industrial strategy remains focused on fast adoption, we are agreeing to a permanent, compounding transfer of value to foreign technology monopolies. Every API call, every subscription, every enterprise licence, is rent paid to a landlord in Silicon Valley, year after year. Singapore has never just accepted that kind of relationship with any industry. We move up the value chain, extracting value at the source. We do not just use semiconductors—we make them. Our ambition for AI should be no different: we need to stake out our own defendable positions, and capture the value that comes with them.
The question is where, specifically, we have a right to win. And I think the answer starts with understanding our constraints.
We cannot win a compute war. We lack the land, the power grid, and the capital to train trillion-parameter frontier models. But Singapore has been here before. When we faced constraints in other industries, we did not try to outspend larger competitors. We identified niches where our specific combination of capabilities gave us an advantage, and we built from there.
We have already done this once in AI. Singapore has produced SEA-LION and MERaLION—large language models built on open-weights foundations, fine-tuned for Southeast Asian languages and our cultural context. We identified a gap in the region, and we filled it. That was and is the right instinct. We must now look further afield, with global ambitions.
The case for Edge AI
I want to make the case for another niche, one with global export potential: Edge AI, and the efficiency research that underpins it.
The rest of the world is running into the same energy wall we face. Data centres are straining electrical grids everywhere. The United States is planning new nuclear plants specifically to power AI infrastructure. My colleague Dennis Tan earlier highlighted the outsize environmental impact of an indiscriminate rush to embrace large-scale models for every minor task—a dilemma every nation is grappling with.
The global problem, in other words, is the same problem our constraints have been forcing us to think about: how do you make AI run on less?
This is the field of model distillation and quantisation—compressing large AI models so they can run locally, with minimal degradation in performance, without an internet connection to the cloud, on a fraction of the energy. The result is what practitioners call Edge AI: intelligence that runs on the device itself, rather than routing data to a distant server.
The Prime Minister announced four National AI Missions in this Budget. I want to look at two of them through the lens of Edge AI.
In advanced manufacturing, the robots on a factory floor cannot depend on a cloud connection. A production line that hesitates because of network latency, or halts because of an outage, is a competitive liability. The AI running that factory needs to operate locally, and be resilient to disruption.
In connectivity, the autonomous systems managing containers at our ports make thousands of routing and sequencing decisions every second. Our ports handle a huge proportion of the world’s transshipment traffic. Those decisions cannot wait for a round trip to a data centre.
The export opportunity follows directly. As the rest of the world scrambles to solve their energy and latency constraints, the software that makes AI run efficiently on local hardware will command a significant premium. We are already being pushed toward solving this problem. We should be solving it deliberately, with an eye on selling the solution.
I am not a deep technical expert, and I am not offering a comprehensive blueprint today. What I am suggesting is a way of thinking about where Singapore has a right to win—and Edge AI is one possible answer to that question—among many others.
The same logic points toward explainable AI: building tools that make AI reasoning transparent enough for regulated industries to deploy. Finance and healthcare, the two other sectors that are the focus of our AI missions, are industries that operate under obligations to explain automated decisions. Singapore’s regulatory institutions and our reputation for rule of law may give us a genuine advantage in that space. These are just ideas. There will be other niches neither side of this House has yet identified.
So what do I ask for?
The ask
The Government has committed $37 billion under the Research, Innovation, and Enterprise 2030 plan (‘RIE2030’), with AI explicitly in scope. I ask that within that envelope, Edge AI efficiency research—model distillation, quantisation, and on-device deployment—be named as an explicit priority, with success measured by commercial export potential, not just domestic adoption.
My more specific ask concerns the National AI Missions. The four Missions are already funded. I ask that each Mission consider edge deployment capability as a key design criterion: whether an efficient, low-latency, sovereign edge model is appropriate for the use case.
And more broadly: I ask that the government establish a formal process for identifying Singapore’s AI export niches—applying the same constraint-aware logic that produced SEA-LION and MERaLION—across our strategic industries. Not just edge AI. Not just explainable AI. A systematic search for where Singapore has a right to win, and a commitment to back those niches with capital.
The goal? To ensure that in years to come, Singapore is a genuine owner of commercial AI capability that we export to the world. Not just a well-compensated tenant, but a landlord of our own.
Part two: The broken rung—protecting the next generation’s career ladder
But owning the technology is only the first part of the answer. The second is protecting the people being asked to adapt to it—including those the market has already started to leave behind.
Mr Speaker, the aggregate economic data looks encouraging. Broad-based wage growth. Lower income inequality. But averages can be dangerous. They can mask a specific problem that I think deserves more attention than it is currently receiving.
Research published in October last year by Stanford and ADP found that since late 2022, when ChatGPT broke cover, entry-level hiring in AI-exposed sectors in the US has fallen by 16%. For software developers aged 22 to 25 specifically, headcount has collapsed by 20%. We do not have such granular data for Singapore, but that alone is part of the problem.
I call this the broken bottom rung.
I want to be clear about what is at stake here. Not in economic terms—in human terms.
The first job is not just income. It is the place where you discover what you are capable of. Where someone more senior takes a chance on you, teaches you something you could not have learned any other way, and passes on a standard of craft or judgement that you will carry for the rest of your career. It is where professional identity is formed.
If that rung is gone—not because graduates are less capable, but simply because it is now cheaper to automate the tasks that used to justify hiring them—we do not just have an unemployment problem. We have a rupture in how knowledge and expertise pass from one generation to the next. The experienced professionals of 2046 are the junior hires of today. [PAUSE] If we do not hire them today, we will not have them in twenty years.
After my parliamentary colleague Eileen and I put out a call for feedback some months ago, we heard from many young Singaporeans who are living this. Graduates who applied for over a hundred roles. Who took unpaid internships hoping it would lead somewhere. Who are talented, willing, and being told by the market that there is simply no place for them yet. They are not asking for sympathy. They are asking for the chance to start.
The broken rung is the most visible symptom. But the disruption does not stop at the entry level. Workers mid-career face a parallel challenge: skills that took years to build, being absorbed by AI faster than retraining programmes can respond.
So what do I ask for?
Data first
We cannot manage what we do not measure. I call on MOM to publish granular employment data for new graduates, broken down by sector, role type, and AI-exposure level. If the broken rung is happening here, we need to see it in our own numbers before we can respond to it.
Make the EIS deduction conditional on the graduate hiring pipeline
The Government’s 400% Enterprise Innovation Scheme (‘EIS’) tax deduction for AI expenditure is a powerful lever. While the details are being firmed up, I ask that the Government require that firms claiming the EIS deduction—above a meaningful threshold—demonstrate a credible plan for maintaining graduate and entry-level pipelines—not by preserving roles that AI has genuinely made redundant, but by investing in reimagining what junior work looks like in an AI-augmented firm. Public money for AI adoption should come with a commitment to developing the workers who will work alongside it.
AI tool subscriptions
The Government’s announcement of six months of free premium AI tool access for workers taking selected courses is a genuinely welcome step. But six months is a trial. Mastery takes longer. AI is not a course you complete—it is a daily practice, the way using a word processor or a spreadsheet is a daily practice. A worker who builds that habit over six months and then has the access removed is being set up to fall behind again.
The Workers’ Party has called for SkillsFuture credits to cover AI tool subscriptions on an ongoing basis. I renew that call today—for young workers starting out, for those mid-career who need to stay current, and for our senior workers looking to embrace new technologies.
Lower the mid-career qualifying age to 35
The SkillsFuture Level-Up Programme currently targets workers aged 40 and above. But the disruption is hitting earlier. Singaporeans in their mid-to-late thirties—the millennial cohort whose careers were built on skills that AI is now absorbing—are facing displacement before they even qualify for mid-career support. Lowering the qualifying age to 35 would be a concrete, targeted step that addresses where the disruption is actually occurring.
Part three: Rewriting the social contract for the AI age
That brings me to the third and final question.
AI is not just changing how we work. It is changing what we owe each other. The assumptions baked into our social contract—about risk, about reward, about the relationship between capital and labour—were written for a different economy. Budget 2026 is an opportunity to update them.
So what do I ask for?
Redundancy insurance
The Workers’ Party has called for Redundancy Insurance across multiple Parliaments. I argued for it myself in my maiden speech. I will not repeat the full argument today, but I want to offer a different frame for it.
Critics have called it a welfare crutch. I want to make the opposite case: Redundancy Insurance is the engine of the economic agility this Budget demands.
A worker with six months of financial breathing room can say yes to retraining. They can take the risk of moving sectors. They can be the bold, adaptable Singaporean the Government is asking them to be. Without that buffer, the rational response to uncertainty is not agility—it is paralysis, or grabbing the first job that comes along, plugging them into a cycle of underemployment. And paralysis is exactly what we cannot afford in an economy undergoing structural transformation.
The Government is rightly optimistic about AI’s potential. But responsible governance requires managing for all outcomes, including the downside. Redundancy Insurance is that hedge. I urge the Government to act on it.
Retraining tax credits—the incentive update
The EIS deduction condition I proposed earlier—on maintaining graduate and entry-level pipelines—addresses new workers entering the workforce. But there is a parallel question about existing workers: what happens to the person already in the job when AI arrives?
Right now, the tax system is neutral on this question. A firm can deploy AI to augment its workforce, or to replace it—and incentives are often available either way. Without a nudge, the default logic of capital is to substitute labour. Labour is expensive. Shareholders benefit when headcount falls. AI makes substitution easier than it has ever been.
I propose a targeted Retraining Tax Credit: available only to firms that can demonstrate they have reskilled a specific worker into an AI-augmented role, rather than retrenching them. Retention, done right, costs less than redundancy. Let us make the fiscal system reflect that.
The AI gains audit
Which brings me, finally, to the distribution question.
Global evidence consistently shows that productivity gains from AI do not automatically translate into wage growth. The gains can flow to capital—to shareholders—while workers on the same payroll see no change in take-home pay.
The Government is investing taxpayer resources and regulatory sandboxes in four National AI Missions: advanced manufacturing, connectivity, finance, and healthcare. These are the right priorities. But they come with a public obligation.
I propose an annual AI Gains Audit, scoped specifically to these four sectors. Not red tape for every business. Not a surveillance regime on private enterprise. A targeted accountability mechanism for sectors receiving extraordinary levels of public support—to answer one straightforward question: are the productivity gains from state-backed AI Missions flowing into the wages of our workers, or are they flowing into shareholder returns?
Parliament and the public have a right to know the answer.
Conclusion
Mr Speaker, this is a Budget with genuine ambition, and the Workers’ Party supports much of its direction.
But ambition must come with accountability. The strategic advantage of AI must be felt in the wages of our workers—not just in the earnings reports of the companies deploying the technology.
Own the technology, not just rent it. Protect the worker who is being asked to adapt. Ensure the gains of AI accrue to workers, not capital owners.
These are not obstacles to Singapore’s AI ambitions. They are the conditions that make those ambitions worth pursuing.
Thank you, Mr Speaker.
