SINGAPORE – Technology stocks may have recovered from their worst monthly performance in years, but most analysts believe the July retreat – amid concerns about the mismatch between earnings growth and capital spending – signalled that investors are no longer betting that tech stocks will keep rallying.

In the last week of July, large institutional investors – including top US hedge funds that had led the rally so far – unwound their exposure to tech stocks at an unprecedented pace.

The 3.2 per cent decline made it the Nasdaq’s worst July performance in 20 years. At its lowest point on July 29, the index was down 10 per cent from its June peak, meeting the definition of a market in correction territory.

Hence, many analysts believe the rebound seen over the past few days is unlikely to hold, and that more volatility is possible.

Still, it would be premature to conclude that this is the end of the AI infrastructure build-up story. Such a turn would have serious implications for Singapore, whose economy has been riding high on the artificial intelligence investment boom over the past few years.

Tech research firm TrendForce estimates that in 2026, AI infrastructure spending by the top nine cloud service providers – Google, Amazon, Meta, Microsoft, Oracle, ByteDance, Tencent, Alibaba and Baidu – will surge by about 90 per cent from 2025 to US$886.7 billion (S$1.14 trillion).

Looking ahead to 2027, TrendForce believes the combined capital expenditure of the nine will hit about US$1.3 trillion.

Mike Wilson, Morgan Stanley’s chief US equity strategist, said: “The near-term path may remain uncomfortable, but Morgan Stanley sees the underlying earnings and AI infrastructure story surviving the unwind, with quality increasingly replacing pure momentum as the preferred expression.”

While July’s tech rout may have taken on a life of its own, with stock prices and indexes such as Nasdaq and South Korea’s Kospi breaching key technical support levels, this does not signal the end of the AI infrastructure investment cycle.

“The market is no longer rewarding AI investment on faith alone,” Wilson said. “Execution, monetisation and operating leverage are becoming the dividing lines between the next leaders and the companies merely spending heavily.”

Analysts said that such investor discretion is important, as the credit spread – the return that lenders demand from borrowers – continues to widen with each new data centre.

Only companies with business models that balance earnings growth, sales and spending will be able to keep investing in new infrastructure.

AI-related debt issuance has surged, with companies raising roughly US$450 billion to US$490 billion in visible bonds and loans, while cumulative off-balance-sheet hidden debt and future lease commitments across major tech giants have ballooned to approximately US$1.65 trillion.

AI is a key driver, but energy, defence and efforts to onshore industrial supply chains due to geopolitical fragmentation are equally important in driving the surge in capital expenditure and corporate debt.

Keeping that in mind, the stock market response to recent quarterly results from US tech giants was quite telling.

SpaceX’s second-quarter results released on Aug 4 – its first report as a listed company – topped analyst estimates, but did nothing to stop the free fall its stock has been on.

Its shares fell as capital spending in the quarter surged to more than U$18 billion, well above estimates, with the bulk directed towards AI compute infrastructure for SpaceXAI, the company’s AI unit that continues to post steep losses.

Just a few days earlier, the shares of Amazon and Microsoft rallied after they reported strong financial results and revealed plans with a visible runway for earning growth.

Microsoft said it would extend the useful life of data centres from 15 to 25 years, while Amazon claimed that its computing campuses can be monetised for more than 30 years.

This, said James Ooi, market strategist at Tiger Brokers, a Singapore-based digital brokerage firm, represents a significant shift in how hyperscalers frame the economics of AI infrastructure.

“While chips may become obsolete within a few years, the surrounding land, buildings, power and cooling (systems) can support multiple generations of hardware upgrades, allowing costs to be spread across multiple server cycles and potentially supporting stronger margins,” he said.

Sustained investments in AI infrastructure and repeated cycles of hardware upgrades mean that makers of advanced chips, from Taiwan Semiconductor Manufacturing Co to Semiconductor Manufacturing International Corp, and top AI chips designers such as Nvidia have their work cut out for several years to come.

The memory chip market is a bit more complex.

Chinese memory chipmaker ChangXin Memory Technologies’ (CXMT) July 27 debut was impressive, with its shares rising nearly 500 per cent on the first day of trade on the Shanghai Stock Exchange.

The future threat signalled by this initial public offering triggered a sell-off in shares of other memory chipmakers such as SK Hynix and Micron, but this was a bit of an overreaction, given that the global memory chip shortage is likely to continue until at least 2030.

Also, CXMT – for now – does not produce the high-bandwidth memory (HBM) chips sought after by AI data centre developers on a commercial scale, with the top three memory makers Samsung, SK Hynix and Micron dominating the market for these chips.

Micron is among the chipmakers that have invested tens of billions of dollars in Singapore over the past few years.

On July 31, South Korea’s main share index, the Kospi, rebounded by nearly 20 per cent after falling more than 15 per cent in the previous two days.

Still, July went on the record as the Kospi’s worst month since the height of the global financial crisis in October 2008.

With the hype unleashed by the launch of ChatGPT by OpenAI in November 2022 fading, analysts say investors will have to look at each AI-linked company on its own merits and not just as part of a sector-wide boom.

David Eiswert, portfolio manager of equity strategy at global asset management firm T. Rowe Price, said one of the biggest mistakes investors make is treating AI as the next software cycle.

“Unlike traditional software, AI becomes more capital-intensive as it scales. Every new model, inference request and application requires additional computing power, memory, networking capacity and energy infrastructure,” he said.

Eiswert said that from an investment perspective, the more important question is whether further investment continues to produce meaningfully better AI models and economic value.

“As long as these scaling relationships remain intact, companies are likely to continue investing in AI infrastructure. Governments may also maintain or increase their investment, given AI’s growing importance to productivity, economic competitiveness and national security.”

Like every capital investment cycle, the AI build-out will eventually mature. The challenge for investors is not to predict when the cycle will end but to recognise when the fundamental drivers have genuinely changed.

“In our view, the opportunity is not simply about identifying the company with the best model,” Eiswert said.

“It is about understanding how capital expenditure, scarcity and improving returns may flow through an economy being rebuilt around intelligence. The principles of investing have not changed, but the playing field has.”