Parsing the recent AI stock volatility: From Hype to Hard Returns

Every tech revolution begins with unbridled excitement about its possibilities – before invariably transitioning into a more measured reckoning over what actually pays off.

By Chez Anbu, Head of Wealth Advisory, OCBC

Singapore, 7 September 2026 – Every tech revolution begins with unbridled excitement about its possibilities – before invariably transitioning into a more measured reckoning over what actually pays off.

And true enough, after intense focus on the AI boom since the start of the year, the recent stock price volatility witnessed among some AI-related companies seems to be pointing to the inevitable reality: the AI powered stock boom is entering the proverbial moment of reckoning.

Yet the wrong response to volatility is to abandon the AI theme entirely. An equally wrong response is to double down with leverage simply because the long-term story feels obvious.

Rather, the better response is purposeful participation.

Payback time

Dig deeper beneath the headlines, and investors will find that the AI story is entering an interesting phase - both in terms of the technology’s maturity and the investment opportunities being presented.

As a crucial part of this transitional phase, the recent volatility is a sign that the story has become a bit crowded, over-leveraged and priced for extraordinary outcomes. When expectations become this high, even strong earnings can disappoint. For instance, a semiconductor company can report record demand and still see its share price dip if margins, guidance or capacity plans fall short of what investors imagined.

Perhaps the most recent prominent example is the turbulence in South Korea, which serves as a useful warning for what happens when (admittedly lofty) investor expectations, coupled with over leverage, do not match up to reality.

Earlier this quarter, the Kospi became one of the most visible expressions of the AI boom because of its heavy exposure to memory and semiconductor names. Conversely, when confidence turned more recently, leverage amplified the move. Margin calls and forced selling have not changed AI’s long-term importance - but they have shown how quickly a good investment theme can become a bad market structure.

The reality that real revolutions do not protect investors from overpaying, overconcentrating or overleveraging is not without precedent. In fact, past lessons show that adoption, earnings and valuation must be treated as three different things.

The personal computer revolution changed the world, but the lasting value did not accrue to every company that sold a computer. Consider how it was IBM that helped to legitimise the personal computer - yet much of the economics that still endure today ultimately shifted toward Microsoft’s operating system and Intel’s processors.

The Internet, similarly, was even more powerful than its early believers expected, but many dot com companies disappeared without a trace. Cisco, the poster child of dotcom, sold real equipment into real demand, but peak valuation was too high to sustain. Conversely, Amazon survived not because the internet story was fashionable, but because it built a business model that improved with scale.

Likewise, the AI boom is real, and its productivity gains may be meaningful - but the buildouts required are also massive.

AI’s first phase has been infrastructure. Hyperscalers are spending enormous sums on chips, servers, networking, data centres, cooling and power. This has lifted companies across semiconductors, high-bandwidth memory, chipmaking equipment, cloud infrastructure and electrical systems.

But capital expenditure is not automatically good news; it is, in fact, a claim on future profits. When hyperscalers fund part of this buildout through bond issuance, the cost of capital becomes part of the AI story. Rising hyperscaler debt supply can widen spreads and push investors to demand more compensation. That can weigh on equity valuations, because shareholders must ask whether all this spending will generate returns quickly enough.

Here, the AI debate is shifting as the market is no longer rewarding spending alone. It is beginning to demand payback.

That makes the next phase important. AI will not remain confined to data centres and chatbots. Looking further ahead, the arguably more interesting evolution is the movement from digital AI and agents to physical AI. This will manifest in the form of robotics, factory automation, warehouse systems, and autonomous inspection - even healthcare devices, logistics and humanoid machines.

This broadens the opportunity. Physical AI needs sensors, motors, precision components, safety systems, machine vision, edge computing, batteries, industrial software and simulation tools. It also raises the difficulty level. A chatbot can make an error and correct itself. The stakes are much higher for a robot moving through a warehouse, factory or hospital; it must be safe, reliable and economically useful.

Adopting a more systematic investment approach

To return to purposeful participation, Singapore investors should adopt a more systematic approach to the AI boom.

Given the extensive discourse around AI, investors should likely be familiar with the first two layers by now.

The core AI infrastructure basket comprises semiconductors, memory, semiconductor equipment, cloud and data centre enablers. This remains essential, but it should be sized with respect for volatility. These are no longer undiscovered companies; many have already been priced for strong execution.

The second layer is the bottleneck basket: power equipment, cooling, grid infrastructure, advanced packaging, networking and selected materials. These companies may not always carry the AI label, but they benefit when the physical system strains under demand.

But moving up to the third and fourth layers reveal where the lasting value for the AI boom may really lie. The “adoption basket” comprises the third layer: Companies using AI to improve margins, productivity and customer experience. Over time, the winners may include banks, insurers, healthcare companies, logistics operators, manufacturers and software firms that embed AI into their operating model.

The fourth layer, finally, is the emerging physical AI basket. Unitree Robotics’ recent blockbuster stock market debut in Shanghai is strong evidence already of this emergence – but again, the broader narrative should not be reduced to one (humanoid robot) name. The opportunities are likely to sit across a wider range of applications: automation, sensors, industrial software, safety systems and edge computing.

For investors, concentration management is especially important. Investors can trim large single stock positions in AI leaders from strength, or manage them through hedging and structured participation strategies where suitable. Yield enhancement structures may be useful only when the investor is comfortable owning the underlying stock through a drawdown.

On the fixed income side, wider spreads from hyperscaler issuance may eventually create opportunities in high quality credit. But this should be approached selectively. AI related debt is still debt and long duration exposure can suffer if inflation stays sticky or the Fed remains cautious.

Volatility isn’t the end of the AI story

AI is moving from infrastructure to adoption and eventually to physical automation. It will create large winners, expose weak business models and punish companies that cannot justify the capital being poured into them. Investors need to understand where scarcity exists, where pricing power is real, where balance sheets can carry the investment cycle and where expectations have already run too far.

They also need to understand that, as mentioned earlier, volatility is far from the end of the AI story. Rather, it is the market’s way of reminding us that even the strongest revolutions must still pass through the discipline of earnings, cash flow and valuation.

This article was first published in The Edge Singapore on 7 September 2026.