Cancellations up, pre-IPO targets missed: the AI case retreats to the plumbing

Raymond Ang, Standard Chartered's global head of private, SME and affluent clients and head of wealth and retail banking for Greater China and North Asia, told the South China Morning Post that AI is "not something investors choose to opt into or out of" and called it a structural shift across industries and asset classes ^[1]. The same article concedes the setting for that confidence: cancellations of data-centre developments have increased since late 2025, and several high-profile AI firms missed revenue targets ahead of planned initial public offerings ^[2][3].

That is the interesting part. Two years after what the article calls the AI boom of 2023 and 2024 ^[4], the pitch has moved from naming the winners to owning the whole chain. The article says public markets are wavering while underlying data shows corporate spending on AI going the other way ^[5], though the material supplied carries no figure to size that gap, which is where a reader would want one. It also cites a report finding that executives plan to spend on customer engagement while expanding into innovation functions such as R&D, a shift from efficiency to discovery ^[6]; the excerpt does not name the report.

The layer argument comes from an analyst identified in the piece only as Wang, who says infrastructure investment remains strong but faster growth is now in the application layer, particularly enterprise adoption ^[7]. He describes current adoption as "wide but shallow", what he would expect early in the diffusion of a general-purpose technology ^[8], and draws the mobile-phone parallel: chips, networks and devices first, software and applications capturing more incremental value later ^[9]. His forward claim is that AI application spend as a share of enterprise revenue is still small and should rise quickly once systems replace parts of workflows rather than augment workers ^[10]. That is a testable line, and it is the one to hold him to.

The supply-side defence rests on a KKR report from November 2025, which argued this buildout is unlike the 1990s fibre overbuild because projects sit behind pre-emptive hyperscaler offtake agreements, while power constraints and permitting bottlenecks limit oversupply risk ^[11]. Its conclusion was that investors focused on execution, unit economics and scarce inputs such as power, land and grid access are best placed as the sector matures ^[12]. Note the timing: that report was published in the same period in which the cancellations began rising ^[13]. The source does not say whether those cancellations reflect power and permitting constraints, which would support KKR's thesis, or softening demand, which would not.

Ang says high-net-worth investors are looking beyond public markets and either seeking access to private assets or raising their allocation to them ^[14]. Subberwal, quoted in the article, is blunter about the limits of stock-picking here: it is "still too early to call clear application-level winners in AI", so clients are encouraged to think across the entire value chain, from enablers and suppliers to end users ^[15]. He expects the next phase to reward resilience over speed, with products embedded deep into client workflows, reliable infrastructure and recurring demand rather than rapid experimentation, and says the market will favour better network effects and return-on-investment discipline ^[16]. Passive exposure, he adds, is not enough ^[17].

Read that as an admission. Diversification across a value chain is what you sell when the terminal value of individual names has stopped being legible.