In November 2025 the Monetary Authority of Singapore (MAS) consulted on Guidelines on Artificial Intelligence Risk Management for financial institutions. The consultation closed at the end of January 2026, and in August 2026 the MAS Chairman told Parliament the guidelines would be finalised soon and would apply to all AI use by financial institutions, including agentic AI.

The draft marks a change of register. MAS has long published principles for AI in finance, notably the Fairness, Ethics, Accountability and Transparency (FEAT) principles. The new guidelines set supervisory expectations across the whole AI life cycle: governance at board and senior management level, an inventory of AI in use, risk materiality assessment, controls at each stage from data and model development to deployment, monitoring and retirement, and the capabilities needed to run all of that. They cover traditional machine learning, generative AI and autonomous agents, and they apply in proportion to an institution's size and risk.

Three points that matter for wealth

First, AI used in customer-facing or regulated activity, such as advice, credit decisions or underwriting, faces the most exacting expectations on explainability, fairness, human oversight and testing. A robo-advisory model is squarely in that category.

Second, third-party AI does not shift accountability. A platform that licenses a model, or builds on a foundation model from a cloud provider, is still answerable for how it behaves. That reaches vendors too: an institution will ask its suppliers to demonstrate the same discipline.

Third, the guidelines expect monitoring in production, not only validation before launch. A model whose performance drifts against real market behaviour has to be seen drifting, and someone has to own the response.

What this means

For family offices and wealth platforms in Singapore, the practical question is no longer whether to adopt AI but whether the adoption can be evidenced. An AI inventory, a risk classification per use case, named owners, monitoring records and a review cycle are the artefacts a supervisor will look for. Industry participants in MAS's Project MindForge published an operationalisation handbook in April 2026 that turns the draft into working practice, which is a useful starting point.

Our view is that the institutions that treat these expectations as design inputs, rather than as a compliance layer added after the fact, will find them lighter to carry. Model risk control built into the delivery pipeline costs less than model risk control retrofitted to it.

General information only, not advice. See Terms of use.

As at the end of December 2025, more than 2,000 single family offices (SFOs) in Singapore received tax incentives under sections 13O and 13U of the Income Tax Act, according to a parliamentary reply from the MAS Chairman in August 2026. The figure was about 400 at the end of 2020.

The growth has come with tighter conditions. Since January 2025 the assets under management test is measured on designated investments rather than net asset value and applies at every financial year end, local business spending is tiered by fund size, and investment professional headcount and residency requirements apply to both schemes. Both schemes have been extended to the end of 2029. A further MAS circular on 31 July 2026 refined the economic conditions for SFO funds, with some changes applying from August 2026 and others retroactively from January 2025.

The allocation conversation has changed

Increasingly, the families setting up in Singapore are asking less whether to hold digital assets or use AI, and more how to hold digital assets under the same custody, reporting and compliance discipline as everything else, and how to use AI in the office without creating a governance gap that a bank, an auditor or MAS would notice.

On digital assets, the Singapore framework is now clear enough to plan around: licensed custody by default, documented key control where self custody is warranted, and anti-money laundering (AML) expectations for digital payment token service providers that MAS updated again in July 2026. Cross-border holdings add jurisdiction-specific structuring, particularly where a beneficiary sits in Europe or Hong Kong.

On AI, the office's own use of models for research, monitoring or reporting sits outside financial-institution regulation in most cases, but the platforms and banks the office relies on are inside it. The forthcoming MAS AI risk management guidelines will shape what those platforms can offer and how they document it.

What this means

Two habits serve families well here. The first is to write the allocation to newer asset classes down as a policy, with limits, before the first position is taken. The second is to treat AI tooling in the office as a supplier relationship with the same due diligence as a custodian or an administrator: what data leaves the office, where it is processed, and who is accountable for the output.

General information only, not advice. See Terms of use.

Hong Kong's Stablecoins Ordinance took effect on 1 August 2025. The Hong Kong Monetary Authority (HKMA) received 36 applications in the first window and granted the first two issuer licences on 10 April 2026, to HSBC and to Anchorpoint, a joint venture of Standard Chartered Bank (Hong Kong), HKT and Animoca Brands. Both intend to launch Hong Kong dollar stablecoins in the second half of 2026.

The September 2026 Policy Address went further. Regulated stablecoins are to be tradeable on licensed virtual asset trading platforms and accepted as a settlement asset for tokenised money market funds, the Securities and Futures Commission's framework is to extend to tokenised gold and other real-world assets on licensed platforms, and a digital asset custody monitoring system is due in the second half of 2026.

Singapore's perimeter

Singapore has taken a different line. From 30 June 2025, any digital token service provider operating from Singapore, even one serving only overseas customers, needs a licence under the Financial Services and Markets Act 2022. MAS said it would set the bar high and would generally not grant such licences, and allowed no transitional period. In July 2026 MAS published supervisory expectations on AML and countering the financing of terrorism for digital payment token service providers.

Meanwhile the institutional side keeps moving: Project Guardian's tokenised fund work, licensed custodians, and tokenised money market funds used as collateral by Singapore-based funds.

What this means

For private capital the two hubs are complementary rather than competing. Hong Kong is the place to watch for regulated stablecoin settlement and tokenised real-world assets reaching a licensed retail and professional market, with mainland connectivity as the draw. Singapore is the place for institutional-grade custody and fund tokenisation under a regulator that prefers a small number of well-supervised participants.

A family office with beneficiaries or counterparties in both cities may need to consider both, and the structuring question is which activity sits where. That is a legal and tax question first and a technology question second, and it rewards getting the framework settled before the platform is chosen.

General information only, not advice. See Terms of use.

MAS launched Project Guardian in 2022 to test tokenisation with financial institutions. By late 2024 it involved more than 40 institutions across several jurisdictions, had completed over 15 industry trials, and had produced two industry frameworks: one for tokenised fixed income and one for tokenised funds. Singapore and UK asset management associations joined in 2025.

In November 2025 the project published "Operationalising Tokenised Funds". The report's most useful contribution is that it stops treating tokenisation as a switch. It describes a spectrum. At one end is a "digital mirror", where the ledger reflects an off-chain register that remains the legal record. In the middle is a "digital twin", a hybrid in which both records exist and are reconciled. At the far end is a fully native tokenised fund. Each model has its own operating, legal and control implications, and most institutions will move along the spectrum rather than jump to the end.

Why the plumbing matters more than the token

A ledger can move value in seconds. The subscription and redemption process, the transfer agent's role, the custodian's assurance, the cyber controls and the claw-back procedure for a token moved in error cannot be assumed away. The report sets these out fund by fund, which is what a board, an auditor or a family office's own adviser needs to see before allocating.

Settlement is the other half. MAS is developing a Singapore dollar wholesale settlement facility for participating institutions, and pilots have shown tokenised fund subscriptions and redemptions settling over existing bank messaging rails. Hong Kong is pursuing a parallel track through its own tokenisation programme, with an end-2026 target for central bank digital currency settlement.

What this means

For an allocator, the questions to ask of any tokenised fund are practical: where is the legal record, who reconciles it, how are subscriptions and redemptions settled, who holds the keys, and what happens when something goes wrong. If the answers are written down, the product is investable on its merits. If they are not, the token is a distraction from the fund.

General information only, not advice. See Terms of use.

The banks that serve private wealth in Singapore and Hong Kong spent 2024 and 2025 running AI pilots. In 2026 the pattern is different. UBS Global Wealth Management runs an agentic AI programme out of Hong Kong and Singapore, with a delivery hub in both cities and staff co-creating tools for their own workflows. DBS, which began investing in AI and machine learning a decade ago, describes AI as embedded across the bank rather than confined to pilots. Standard Chartered has put a generative AI assistant in front of most of its workforce.

The governance side has moved with it. Twenty-four institutions in MAS's Project MindForge, including the Singapore banks, global private banks, sovereign investors and large asset managers, co-developed an AI risk management handbook alongside the major cloud providers. That is unusual: the industry writing its own operating manual under the regulator's eye, ahead of the regulator's final rules.

Where the work is moving

Practitioners at industry sessions in Singapore this year describe the same pattern: AI is strongest at gathering, structuring, summarising and comparing. That covers investment research, content, know-your-customer and source-of-wealth review, and internal policy analysis. Advisory judgement in complex situations remains with people, but the adviser now arrives at the conversation with far more of the preparation done.

The consequence for a wealth platform is capacity. The same team can serve more relationships to a more consistent standard, and the bottleneck moves from analysis to decision.

What this means

A family office or wealth platform should be asking its banks and providers three questions. Which of your processes that touch my affairs now involve a model, and what does a person still check? Where is my data processed, and under whose controls? And when a model gets something wrong, how will I know?

None of these are hostile questions. The institutions doing this well will have ready answers, because the MAS guidelines now in preparation will require them to.

General information only, not advice. See Terms of use.

The Asia Pacific data centre pipeline reached a record 26.5 gigawatts in the first half of 2026, and Southeast Asia accounts for roughly half of the capacity under construction. Malaysia leads with just over 1,000 megawatts being built, most of it in Johor, followed by Thailand at around 860 megawatts. Indonesia's activity is concentrated around Jakarta, with a large AI campus announced in Batam in mid-2026. Vietnam and the Philippines are earlier in the cycle and growing from a low base.

Singapore sits at the centre of this without being able to host most of it. Its own constraints on power and land have pushed hyperscale capacity across the causeway, where Johor offers Singapore-grade connectivity at Malaysian cost. The Johor-Singapore Special Economic Zone formalises that corridor.

The constraints have changed

In February 2026 Malaysia stopped accepting applications for data centres not linked to AI, and the state has begun to hear community objections over water and pollution. Grid capacity, water supply and social licence are now part of the cost base for every project in the region. The projects that clear those hurdles will be worth more for having done so.

Infrastructure or application

For an investor in early technology adopters, the region offers two distinct layers. The infrastructure layer, data centres and the power and connectivity around them, is capital intensive, increasingly consolidated and priced by large operators and funds. The application layer, the companies putting AI, machine learning and distributed ledgers to work in finance, manufacturing, logistics and services across these markets, is where earlier positions and hands-on expertise still make a difference.

What this means

Our interest is in the second layer, and in the companies within it that adopt ahead of their market and can show the discipline to run what they build. The infrastructure boom is the context, and it is real. It is not, by itself, the opportunity.

General information only, not advice. See Terms of use.