MAS has stress-tested Singapore’s corporate sector against a severe AI investment shock. For investors, the next question is more practical: which AI beneficiaries have enough financial resilience to withstand a slowdown?
The biggest mistake investors can make with Singapore’s AI trade is to treat all beneficiaries as the same.
They are not.
A semiconductor equipment supplier, an electrification contractor and a data-centre owner may all benefit from AI investment.
But they are exposed to the AI cycle in completely different ways.
One depends on semiconductor equipment orders.
Another depends on project execution and infrastructure spending.
A third is exposed through data-centre demand, rents and financing costs.
Their balance sheets can also look very different.
That matters because the Monetary Authority of Singapore’s latest Financial Stability Review has provided investors with something unusually useful: a severe stress scenario for the corporate sector.
MAS found that 32% of Singapore-listed companies would be classified as at risk under its severe AI-investment downturn scenario, although those firms represented only 16% of corporate debt. The scenario included revenue shocks of up to 30% and interest-rate shocks of up to 400 basis points.
MAS did not identify the individual companies in that 32%.
So investors should not attempt to reverse-engineer the result by declaring that a particular stock is one of the 32%.
Instead, the stress test gives us a framework.
The question becomes:
Which Singapore-listed companies have meaningful AI upside without excessive operating, financial and valuation risk?
The five variables that matter
A useful way to analyse the AI trade is:
AI exposure × operating leverage × financial leverage × liquidity × valuation
The first variable tells us how much a company benefits from AI spending.
The other four determine how much of that benefit shareholders may actually keep — and how much downside exists if the cycle turns.
A company with high AI exposure and a strong balance sheet can be very different from one with high AI exposure, high debt and a valuation dependent on uninterrupted growth.
That distinction is visible in several Singapore-listed companies.
AEM: when AI becomes a major earnings driver
AEM Holdings provides one of the clearest examples.
In its 1H2026 results, AEM reported revenue of S$247.2 million, up 29.9% year on year.
The company said the increase was driven by the ramp-up of its fabless AI/HPC customer, which became its largest revenue contributor for the period.
AEM also reported a sharp improvement in profitability, with PBT margin reaching 15.5%, compared with 2.1% a year earlier.
The company subsequently raised its FY2026 revenue guidance to S$630 million–S$680 million, from S$550 million–S$600 million, and said its AMPS backlog had exceeded S$400 million.
On the surface, this is exactly what an investor wants from an AI beneficiary:
AI demand → customer ramp → higher revenue → operating leverage → higher earnings.
But the same data creates the investment question.
If one customer has become AEM’s largest revenue contributor, then the AI opportunity also becomes an important source of concentration risk.
That does not mean AEM is financially vulnerable.
It means the investor needs to distinguish between:
AI exposure
and
AI dependence.
Those are not necessarily the same thing.
AEM is therefore a useful case study for the upside side of the MAS framework.
The next question for shareholders is how durable the customer ramp is, how much of the current earnings improvement is structural, and how the economics would change if semiconductor investment growth slowed.
UMS: high semiconductor exposure, but balance-sheet strength matters
UMS Integration illustrates a different part of the equation.
UMS’s FY2025 annual report shows that the semiconductor segment accounted for 86% of group revenue, compared with 84% in FY2024.
The semiconductor business is therefore central to the company’s economics rather than simply one growth division among many.
FY2025 revenue increased 4% to S$251.1 million, while net profit attributable to shareholders increased to S$41.6 million from S$40.6 million.
The company also reported S$43.1 million of cash and S$2.1 million of free cash flow for FY2025.
The important point is not that UMS is an “AI stock.”
Its own annual report explicitly links its semiconductor strategy to rising demand for AI-related technologies and advanced packaging.
UMS said it had invested more than S$155 million over four years to expand and upgrade its manufacturing capabilities in preparation for production ramp-ups from key customers.
That creates a classic cyclical-investment question.
The company is spending capital today because customers expect future demand.
If those expectations prove correct, the additional capacity can support growth.
If semiconductor capital expenditure slows materially, however, the same capacity expansion can increase the sensitivity of returns to utilisation and customer demand.
That is precisely why capital intensity belongs in an AI-stock analysis.
CSE Global: AI infrastructure does not automatically mean AI economics
CSE Global provides a different case study.
CSE operates across electrification, communications and automation rather than being a pure semiconductor company.
That makes its AI exposure more indirect.
But it is increasingly relevant to the infrastructure buildout supporting data centres and electrification.
The company’s investor-relations site describes electrification as one of its core solution areas and says around 90% of revenue comes from its recurring Flow business.
More recently, CSE announced two major electrification contracts worth S$190.5 million on 24 September 2026.
This illustrates why simply labelling companies “AI beneficiaries” can be misleading.
CSE can benefit from the broader infrastructure investment associated with electrification and data-centre expansion without having the same direct semiconductor-cycle exposure as AEM or UMS.
That potentially changes the risk profile.
The investor should therefore ask:
How much of the company’s earnings depends on AI-related infrastructure, and how much comes from broader recurring demand?
The distinction matters enormously during a downturn.
The real dividing line is not AI versus non-AI
These three examples already demonstrate why.
AEM:
direct semiconductor testing exposure + major AI/HPC customer
UMS:
high semiconductor revenue concentration + substantial manufacturing investment
CSE:
broader engineering/infrastructure exposure + recurring business model
All can benefit from the AI infrastructure cycle.
But the mechanisms are different.
That means an investor cannot answer the risk question simply by calculating how much “AI exposure” a company has.
The more useful framework is:
| Factor | More resilient characteristics | Greater sensitivity |
|---|---|---|
| AI exposure | Diversified | Highly concentrated |
| Revenue | Recurring/diversified | AI-cycle dependent |
| Margins | Strong and stable | Thin/cyclical |
| Balance sheet | Strong liquidity | High leverage |
| Free cash flow | Consistently positive | Negative/volatile |
| Capex | Moderate | Capital intensive |
| Customers | Diversified | Concentrated |
| Working capital | Modest | Heavy |
| Valuation | Supported by current earnings | Dependent on aggressive growth |
| AI dependence | One growth driver | Core earnings driver |
This is where the MAS stress test becomes useful for equity investors.
It does not tell us which stocks to buy.
It tells us which variables deserve more attention.
The valuation problem may be bigger than the revenue problem
There is another reason AI investors should not focus exclusively on revenue growth.
Suppose an AI-related company is expected to deliver very rapid earnings growth.
If growth subsequently slows — even while earnings continue increasing — investors may decide that a lower valuation multiple is appropriate.
That creates a double sensitivity:
slower earnings growth
plus
lower valuation multiple
The share price can therefore fall much more sharply than the underlying business.
MAS itself highlights this issue at the global level.
It says current equity valuations require sustained strong revenue growth and substantial eventual profitability from large investments in data centres and advanced semiconductors. A material shortfall in earnings or expected returns could trigger a broader reassessment of AI-related valuations.
This is one reason we should resist judging Singapore’s AI beneficiaries purely by revenue growth.
The better question is:
How much future growth is already required to justify the current valuation?
That requires a fresh, date-specific valuation screen for each company rather than simply comparing trailing P/E ratios from different sources.
The bear case does not require an AI collapse
This is perhaps the most important point in the entire series.
An investor does not need to believe that AI demand will disappear to construct a bearish scenario.
A more realistic downside scenario could be:
AI demand remains strong
↓
but hyperscalers become more selective
↓
capital expenditure growth slows
↓
semiconductor equipment orders normalise
↓
supplier growth slows
↓
operating leverage works in reverse
↓
investors reduce valuation multiples
The companies with the strongest balance sheets and most diversified revenue streams would have more room to absorb that adjustment than companies dependent on constant high growth.
Again, this is an analytical scenario, not a forecast from MAS.
What would confirm the bull case?
Investors should look for evidence that AI investment is translating into sustainable economics.
For semiconductor-related companies:
- customer orders continuing to grow;
- backlog converting into revenue;
- margins remaining strong as volumes rise;
- new product introductions generating incremental demand;
- cash flow keeping pace with earnings;
- customers continuing to invest in semiconductor capacity.
For infrastructure companies:
- sustained contract wins;
- recurring revenue growth;
- healthy project margins;
- manageable working-capital requirements;
- disciplined capital expenditure.
For data-centre REITs:
- sustained occupancy;
- rental growth;
- tenant quality;
- manageable refinancing costs;
- disciplined development spending;
- and evidence that AI-driven demand is translating into sustainable rents rather than simply short-term capacity demand.
What would invalidate the bull case?
The warning signs would be different.
Semiconductor suppliers
Watch for:
- order cancellations;
- slowing backlog;
- inventory accumulation;
- margin compression;
- customer concentration becoming more problematic;
- capex commitments rising faster than demand.
Infrastructure companies
Watch:
- falling project margins;
- working-capital deterioration;
- weaker cash conversion;
- increased acquisition or capex requirements.
Data-centre businesses
Watch:
- rising financing costs;
- supply growth outpacing demand;
- weaker rental negotiations;
- tenant concentration;
- increasing capital requirements.
Across all three categories, the most important warning sign may be:
Earnings continue to grow, but cash generation and returns on incremental capital deteriorate.
That would suggest the AI boom is generating activity without generating sufficient economic value.
The investor should not ask “Who wins AI?”
A better question is:
Who can capture AI’s upside without taking disproportionate balance-sheet and valuation risk?
That is a much harder question.
It also produces a much more useful investment framework.
A company with moderate AI exposure, strong free cash flow and a diversified customer base may ultimately deliver a better risk-adjusted investment outcome than a company whose entire earnings trajectory depends on continued hyper-growth in AI spending.
Conversely, a company with high AI exposure may still be attractive if its balance sheet, customer relationships, margins and valuation provide sufficient protection.
The point is not to eliminate AI exposure.
It is to understand what you are being paid for taking it.
What investors should watch now
MAS has already supplied the stress scenario.
The next step is monitoring whether corporate fundamentals move towards or away from it.
For Singapore-listed AI beneficiaries, investors should track:
1. AI-related revenue concentration
Is AI becoming a larger share of the business?
2. Customer concentration
Is growth increasingly dependent on one or two customers?
3. Operating leverage
Are incremental revenues producing proportionately higher profits?
4. Free cash flow
Are reported earnings turning into cash?
5. Capital intensity
How much new capital is required to support growth?
6. Financial leverage
Could a downturn materially affect interest coverage?
7. Liquidity
How much time does the company have if cash flow weakens?
8. Valuation
How much future growth is already embedded in the share price?
These variables can change well before an AI downturn becomes obvious in headline revenue numbers.
Bottom line
MAS’s stress test does not provide investors with a list of stocks to avoid.
It provides something more valuable: a way to think about AI risk.
AEM demonstrates how rapidly AI/HPC demand can translate into semiconductor-testing revenue and operating leverage. UMS demonstrates how AI-driven semiconductor demand can sit alongside substantial manufacturing investment and high segment concentration. CSE demonstrates that exposure to the broader AI infrastructure buildout can take a different, more diversified form.
None of these companies should be described as one of MAS’s 32% of at-risk firms.
MAS did not publish such a company-level classification.
The real investment exercise is therefore more nuanced.
AI exposure creates opportunity. Balance-sheet strength determines resilience. Cash flow determines economic quality. Valuation determines how much disappointment shareholders can absorb.
The Singapore AI trade may continue to deliver substantial earnings growth.
But the next phase of the investment story will be less about identifying companies exposed to AI.
It will be about identifying companies where AI upside is not outweighed by operating leverage, financial leverage, liquidity risk or excessive expectations already embedded in the share price.