Mr Speaker, as with past years, the Finance (Income Taxes) Bill before this House seeks to implement the tax changes announced in Budget 2026, by amending the Income Tax Act and related legislation. My speech today will be largely domestic in focus, while my Sengkang Colleague Jamus Lim will focus on international issues such as those relating to BEPS2.0.
My speech today is in two parts. In part one, I will touch on key specifics of the Bill, in particular the new deduction for qualifying AI expenditure. In part two, I will set out what remains my wish list for tax changes in Singapore, which hopefully can come as soon as Budget 2027. While the Minister and the team from the Ministry of Finance will likely find some of them familiar, these issues are, if anything, more pressing than when I last raised them in this House during the Budget 2026 debate and on last year’s edition of the Finance (Income Taxes) Bill in 2025.
Part 1: Supporting companies in their AI journey
Let me first touch on how we can support our companies in their AI journey.
Clause 13 inserts a new section 14ZK, allowing a deduction for qualifying AI expenditure for YA2027 and YA2028. Up to $50,000 of such expenditure a year qualifies, with a multiplier of 300% where the cost is already deductible under section 14, and 400% where it is not. I support the intent of this new measure.
Now, qualifying AI expenditure includes expenditure on AI systems or a qualifying AI business service, as defined by section 14ZK(2). However, especially after the “SaaS-pocalypse” fears in early 2026, almost every self-respecting developer of business software will call itself AI-powered, and the definitions in section 14ZK(2), which include the "provision of an online platform", are understandably broad. That creates two opposite risks. The first is over-claiming. A firm could claim its entire CRM or accounting subscription because an AI assistant is built in. The second is the reverse: an overly narrow reading could penalise the firm that pays a premium for software to adopt AI where its staff actually work, inside everyday software, rather than buying a standalone chatbot. In my view, the guidance from IRAS should address both errors explicitly.
Consider what small businesses actually buy. Looking at the major software providers today, Xero's AI agent, JAX, is currently included in existing subscriptions at no extra cost, so there is no AI-specific price at all. Is the qualifying amount zero, the whole subscription, or something in between? Intuit, the developer of the widely used accounting QuickBooks, has woven its AI features into QuickBooks Online and changed its subscription prices from August this year. Is the price increase the AI component? Salesforce, arguably the largest CRM player today, sells some of its Agentforce AI as separately priced add-ons and credits, which are easy to identify, but also in bundled editions where the CRM licence and the AI come at a single per-user price. Then there is the additional complication of per-seat vs. usage-based pricing models, which may evolve quickly alongside the rapid developments in AI.
Section 14ZK(5) requires a person who cannot, "despite reasonable efforts", obtain evidence of the qualifying portion to apportion "in a manner that is reasonable in the circumstances", and if the Comptroller is not satisfied that it is reasonable, section 14ZK(6) allows the Comptroller to apportion according to his or her best judgment. With respect, a sole-proprietor or five-person firm in Sengkang cannot negotiate a cost breakdown from a global software vendor. Hence when AI is bundled into ordinary software and the vendor discloses no AI-specific price, what evidence does the Government expect a small business to obtain? And what allocation will IRAS accept?
I suggest that IRAS provides additional guidance with worked examples on bundled software soon, after the Bill’s passing. As with other forms of claims such as the deemed business expenses for commission agents and taxi drivers, or employment work-from-home expenses, we can also adopt a simple default percentage for smaller claims where no such pricing transparency exists. Third, a list of common software products with accepted apportionments, much as the Productivity Solutions Grant lists pre-approved solutions that qualify. And fourth, no penalties for a good-faith apportionment made in line with the guidance, even if later adjusted.
On a related note, I can understand the urgency to encourage AI adoption among firms, especially as an MOM report released on 30 April 2026 found that 71.5% of firms have yet to adopt AI, and only 3.8% are integrating AI into their core processes. Looking forward, however, if the data breach at Bee Cheng Hiang, the first AI-related breach notified to the PDPC, is anything to go by, the scheme should be expanded to place greater emphasis on qualifying expenditure in the areas of AI governance, risk management and security, so that controls and responsible use can be built up alongside adoption.
Part 2: Supporting individuals in their AI journey
Next, let me turn to how we can better support an individual's AI journey.
Course Fees Relief ended from YA2026. In April, I asked whether it would be reinstated from YA2027, or replaced with another personal income tax relief for skills upgrading. The Senior Minister of State replied that the Government had introduced "more targeted, direct support for lifelong learning", and that there are "no plans" to do either. I think this is a lost opportunity. As I said in November 2025 and again in February 2026, a broad-based Skills Investment Relief, a capped deduction for out-of-pocket training expenses, would reward workers who retrain on their own initiative, including for a career unrelated to their current job.
The contrast with corporates is stark. A company that spends $50,000 on AI subscriptions can deduct up to $200,000. An individual gets a single six-month subscription to a premium AI tool, and only if enrolled in a selected AI course.
Why does the Government give businesses an enhanced deduction for AI subscriptions, but give individuals only a time-limited, course-linked one? AI fluency, like any skill, comes from daily practice, not from a course certificate. How will the Government support continued access after the free six months, particularly for workers who need to practise independently, or who are between jobs with no employer to pay? For these groups of workers, they inadvertently also lose out on opportunities to upskill themselves, especially if they have already exhausted their SkillsFuture credits and do not have the ongoing training provided by firms seeking to build up their human capital.
If the concern is that a flat relief favours higher earners, there are alternatives. A targeted credit, or a subsidised renewal for lower-income learners would reach those who pay little or no income tax more directly. As I shared in my Budget 2026 speech, PM Wong spoke in length about AI and in an era where AI is recognised as a critical driver of economic transformation, it is incongruous that personal expenditure on AI-related tools and education does not receive the same tax relief consideration as corporate AI investments.
Lasting structural changes to personal income tax
Staying on the personal income tax front, for YA2024 and YA2025, resident taxpayers received a personal income tax rebate capped at $200. For YA2026, there is none. As I said last November, the rebate looked temporary, and so it has proven to be the case.
As I stated in my Budget 2026 speech, the lower and middle personal income tax brackets, including the $20,000 tax-free threshold, have not been updated since 2001, some 25 years ago. Since then, nominal wages and the cost of living have risen significantly, leading to "bracket creep," where taxpayers are pushed into higher marginal tax brackets even if their real purchasing power has stagnated.
In his Budget round up speech, PM Lawrence Wong said that about 1 in 3 resident workers pay no personal income tax, and that about 8 in 10 of those who do have an effective tax rate below 6%. In a November 2020 parliamentary reply, then-DPM Heng Swee Keat said about half of workers paid no personal income tax, and 80% of taxpayers had effective rates below 4%. In about five years, the share outside the tax net has declined from about half to about one in three, while the effective tax rate the Government cites for 8 in 10 taxpayers has moved from below 4% to below 6%. Some of that is real wage growth, which we welcome. But part of it is what bracket creep looks like when the $20,000 threshold has not moved since 2001.
With the earned income relief at $1,000 for workers below 55 for as long as I can recall, I hope that we can institute a mechanism to regularly review both the personal income tax brackets and basic relief quanta, so the system keeps pace with wages and prices.
Lasting structural changes to corporate income tax
Sticking to tax rates but this time on the corporate income tax front. In our motion on "An economy of the future that works for all" in August, the Workers' Party argued that we must better support local SMEs, a point that I believe saw broad agreement in this House.
According to the OECD's Corporate Tax Statistics 2024 report, MNCs were responsible for 73% of Singapore's corporate income tax revenue in 2021. That makes us the fourth most dependent economy in the world on MNCs for corporate income tax, behind Chile at 76%, Hong Kong at 79% and Ireland at 87%. In his round-up, the Prime Minister himself attributed recent elevated collections partly to "better-than-expected contributions from multinational enterprises in a few key sectors".
The flip side is that we can better support SMEs without sacrificing much revenue. In August, MTI upgraded this year's growth forecast to between 4.5% and 5.5%, citing global AI-related capital expenditure: growth led by large, export-oriented firms, not the neighbourhood SME. Nor are SMEs lightly taxed. As I noted in 2021, based on figures from my Parliamentary Question, SMEs accounted for a mere 9% of total profits before tax in YA2019, yet contributed an outsized 29% of corporate income tax paid.
As I have often shared, the current system of fluctuating corporate income tax rebates, varying from 20% to 50% over the last decade, lacks the certainty and predictability needed for long-term planning. I hope that the Government can introduce more structural progressivity into our corporate income tax regime, and better support our SMEs, such as through reversing the cut to partial tax exemption limits that were introduced in Budget 2018, where the maximum exemption was reduced sharply from $152,500 up to YA2019 to $102,500 from YA2020 onwards, or introduce a two-tiered profits tax similar to Hong Kong's model.
Tax changes to address our TFR
Finally, on measures to address our Total Fertility Rate or TFR. I welcome the measures announced at the National Day Rally, including the SG Child Support Package of up to $62,000 per Singapore Citizen child regardless of birth order, and more childcare leave. Our tax policies now need to follow suit.
In my Budget 2026 speech in February this year, I spoke about Hungary. To recap and update: since 1 October 2025, Hungary has exempted qualifying mothers of three children, regardless of age, from personal income tax on their earnings from work, and its government estimated that about 250,000 women were eligible. From January 2026, it removed the income cap on the existing relief for qualifying mothers under 30 and introduced a similar exemption for qualifying mothers of two born in 1986 or later. The two-child measure is scheduled to extend to older mothers in 2027 and 2028, and to all qualifying mothers of two from 2029. I asked then whether we are willing to be this bold.
If we do not wish to go that far, there is a more modest step. I am comforted that this Government is not afraid of U-turns. The Large Families Scheme was launched in Budget 2025 for the third and subsequent Singapore Citizen child, and the Prime Minister cited it again in his round-up earlier this February. Six months later, it is being absorbed into the SG Child Support Package, just 18 months after its launch. I welcome the move away from birth-order arithmetic. But if the Large Families Scheme can be revisited in 18 months, surely the Working Mother's Child Relief can be too.
Let me recap the Workers' Party manifesto. The adjustment of the WMCR to a fixed-dollar relief from YA2025 leaves a large majority of working mothers either unaffected or worse off, while offering minimal additional tax savings of about $40 a year or less to a small minority. The WMCR should revert to a percentage of income. And because working mothers earning less than $22,000 a year pay no income tax and so gain nothing from the WMCR, we propose a $1,000 Working Mother's Tax Credit for all mothers earning under $25,000 a year, including single mothers.
Conclusion
Allow me to conclude Mr Speaker, with five asks. First, AI for businesses. IRAS should publish clear guidance on claiming tax deductions for bundled AI software, and commit to no penalties for good-faith apportionment. Second, AI for workers. Introduce a Skills Investment Relief, so that Singaporeans who upskill and build AI literacy at their own expense are recognised and supported. Third, end bracket creep. In about five years, the share of workers who do not pay income taxes has fallen from about half to about one in three, while the effective tax rate for 8 in 10 taxpayers has moved from below 4% to below 6%. Fourth, a fairer deal for SMEs. Since YA2020, the maximum partial tax exemption has been slashed by almost a third, from $152,500 to $102,500. Reverse the cut, or adopt stronger progressivity. Fifth, working mothers. If the Large Families Scheme can be revisited in 18 months, surely the Working Mother's Child Relief can be too. The fixed-dollar WMCR leaves a large majority of working mothers unaffected or worse off. Restore the percentage-based relief to send the right signal to our working mothers.
