Some of this reframing reflects fundamental shifts in demand. Some, though, could be narrative running ahead of numbers. Investors should know the difference. Either way, this is by no means an argument against AI.
AI is real, demand for it is accelerating, and the capital expenditure driving it is staggering. Hyperscalers — the handful of tech giants that operate the world’s largest cloud computing networks — are collectively spending hundreds of billions of dollars a year on data centres, chips and the infrastructure to power and cool them. Amazon alone spent US$132 billion ($168 billion) last year. Its capex for this year is even higher: US$200 billion.
Some Singapore-listed companies are genuinely positioned within this buildout. But investors should ask harder questions before paying up for the AI label. Does AI actually help companies charge more, earn more on each dollar of revenue, or fend off competitors better than before? Or are companies simply riding a capex boom that may not last?
The distinction matters because the Singapore stock market, with its thinner liquidity and sparser research coverage, is particularly prone to narrative-driven mispricing. When a stock gets tagged as an AI beneficiary in a broker note, the re-rating can be swift and sharp. Whether it is deserved is another matter entirely.
See also: Southeast Asia’s AI push faces its scaling challenge
Driving demand but not margins
To be fair, some Singapore-listed companies deserve the AI label. Late last year, CSE Global, a systems integrator, signed a five-year agreement with Amazon to supply up to US$1.5 billion worth of electrical power distribution systems for its data centres. CSE’s expertise in this area was originally built serving petrochemical plants and refineries.
As part of the deal, CSE issued warrants to Amazon that would give the US tech giant an 8% equity stake if the order target of US$1.5 billion is met. CSE itself will raise about $48 million if the warrants are fully exercised. Its share price responded accordingly, more than tripling over the course of 2025.
See also: Singtel Singapore’s new AI programme to help push SMEs beyond pilots
Still, the economics tell a more nuanced story. CSE’s revenue for 2025 rose 12.5% to $969 million, but profit margins declined across all three business segments. The lower margins were due not just to the Amazon ramp-up but also to project write-offs, job delays and asset impairments in unrelated parts of the business.
So, while CSE’s revenue grew, the margin expansion that would validate a permanent re-rating has yet to materialise. In other words, CSE’s current share price reflects where margins should be, not where they are. It’s also worth noting that the company burned $67 million in operating cash flow last year, while net debt more than doubled to $163 million.
AEM Holdings presents a similar picture. The semiconductor test solutions provider’s share price has more than quadrupled since the start of 2026, driven by its shift into AI-fuelled demand, a broadening customer base, and a more strategic position in next-generation chip testing.
AEM’s thermal control technology is critical for testing the hotter, more complex chips that AI workloads demand. A new AI and high-performance computing customer is ramping up aggressively, and AEM expects the customer to become its largest revenue contributor by the end of 2026, displacing Intel.
But look at the margins. AEM’s 2025 net profit margin was 4.3%, a fraction of the 18.8% it achieved in 2020, when revenue was higher at $519 million. The company is testing more AI chips, but it’s making far less money per dollar of revenue than it did when testing conventional chips.
The market is pricing the trajectory — AEM is expecting revenue of $460 million to $510 million for 2026 — but trajectory without margin expansion points to volume growth, not necessarily transformation. If margins don’t improve as revenue scales, it would suggest that the structural economics of AEM’s business have not changed, even if the label on its customer base has.
Along for the ride
Sink your teeth into in-depth insights from our contributors, and dive into financial and economic trends
Then there is Micro-Mechanics (Holdings). The company designs and builds high-precision tools that reduce defects and improve yields in advanced semiconductor packaging, assembly and testing processes.
The more AI chips the industry produces, and the more complex they become, the more Micro-Mechanics’ products get used. The company’s revenue for the quarter ended Dec 31, 2025, was the highest in over three years. Its balance sheet is shipshape: $27 million in cash, zero debt.
Even so, investors should be asking some pointed questions. Can Micro-Mechanics charge more for tools used in AI chip packaging? Are fewer competitors able to make them? Or is the company simply selling more of the same tools at the same margins? If it is just more volume at the same margins, then AI is dressing up a cyclical uptick as something more.
The same logic applies to UMS Integration, which makes precision parts for semiconductor equipment manufacturers. UMS itself has embraced the AI narrative, telling investors that the semiconductor industry is entering a super-cycle.
That may well be true, but the company’s business is making components for companies that build chipmaking machines. Will the shift to more complex AI chips enable UMS to charge more for its parts and earn better margins? Or is UMS simply filling more orders in a rising cycle?
Even Singtel, which owns and operates data centres through its Nxera subsidiary, deserves scrutiny. Nxera’s newest facility, DC Tuas, delivers 58 megawatts of capacity designed for the heavier power demands of AI computing. The Singapore-based facility was more than 90% pre-committed before its launch in February this year.
That was also the month when Singtel and private-equity giant KKR jointly announced the acquisition of ST Telemedia Global Data Centres, in a massive $13.8 billion bet on digital infrastructure amid soaring AI demand.
None of this changes the fact that Singtel remains a diversified telco conglomerate. Data centres are one division within a sprawling portfolio that spans consumer telco operations in Singapore and Australia, IT services across the region, and varying stakes in mobile operators from India to the Philippines.
How much of Singtel’s recent re-rating reflects data centre fundamentals versus a broader narrative uplift? Rebranding a diversified telco conglomerate as an AI play because one of its divisions runs data centres sounds less like analysis and more like relabelling.
Merely cosmetic?
Then there are some companies where the AI connection, while real, has little bearing on how they actually make money.
Consider Soon Lian Holdings, a distributor of aluminium alloys. Aluminium is used in data centre cooling systems, server racks, and power infrastructure, all of which are expanding because of AI.
Still, the real driver of Soon Lian’s earnings goes beyond AI. What matters is how well it manages inventory, how effectively it navigates aluminium price cycles, and how efficiently it turns over working capital. If aluminium prices move sharply in either direction and Soon Lian is caught on the wrong side, no amount of AI narrative will cushion the blow.
The AI label has also spread beyond the industrial space. On April 23, CapitaLand Investment’s lodging arm, The Ascott, announced a partnership with three other companies to build AI-powered systems that would let software agents — rather than human customers — search, compare, and book hotel stays.
Positioning itself at the forefront of agentic commerce was how Ascott framed the move. But strip away the terminology, and what it is describing is a technology upgrade to its reservations and customer engagement systems.
This sounds like the kind of investment any large hospitality operator would be making, regardless of whether it’s called AI, digital transformation, or simply IT modernisation. If agentic commerce becomes the new way to describe an improved hotel booking engine, it’s worth asking what AI exposure even means anymore.
The question that matters
None of this is to say that AI is irrelevant to Singapore-listed companies. AI is clearly driving real capex, real orders, and real revenue growth in certain segments. But the market’s enthusiasm over AI appears to have conflated three very different types of exposure.
One is structural, where AI genuinely changes a company’s addressable market and competitive position. The second is cyclical. This is where AI increases demand for existing products but hardly changes how the company makes money. The third is cosmetic, where AI is invoked in corporate presentations and broker reports but has no meaningful impact on the income statement.
The first category deserves a premium, though that should be calibrated to actual margin improvement and not just revenue growth. The second category deserves to be valued for what it is: a good business benefitting from a strong cycle. But cycles turn. When AI capex slows, these companies will feel it in their order books. The third category deserves nothing at all.
In a market starved of growth narratives, AI has become a convenient one. But convenience is not the same as value creation. For investors, whether a company touches AI somewhere in its supply chain should not be the focus. Almost every industrial company does, just as many industrial companies claimed exposure to the electric-vehicle boom a few years ago.
The more appropriate question is whether AI changes a company’s pricing power, its margin structure, or its competitive edge. For most Singapore-listed companies claiming AI exposure, the honest answer appears to be: not yet.
