What will burst the AI bubble?

```

The global AI investment boom is still accelerating, but debates over whether the "bubble is about to burst" are also heating up.

Faced with concerns about the surging capital expenditures of tech giants, the latest report from Hong Kong research institution Gavekal Research concludes: At least from the perspective of the relationship between corporate financing and investment returns, the current cycle of AI capital expenditure is not yet approaching a turning point.

Gavekal Research’s Chief US Economist Will Denyer points out in the report that the core indicator determining whether AI capital expenditure can be sustained—the "Wicksellian Spread"—is still clearly in the safe zone. This indicator measures the difference between the corporate return on invested capital (ROIC) and the weighted average cost of capital (WACC), essentially reflecting the strength of incentives for firms to expand their investments.

Denyer says that as long as ROIC continues to exceed WACC, companies have motivation to expand capital investment. Currently, this spread not only remains positive, but is at least 70 basis points higher than the median of the past twenty years, meaning that although the financing environment is tighter than before, it has not yet substantially suppressed AI investment.

In his view, what truly determines whether this AI boom can continue is not the scale of capital expenditure itself, but the profitability behind it. Once corporate AI investment cannot keep translating into revenue growth and improved profits, causing ROIC to fall significantly while financing costs remain high in a high-interest environment, the Wicksellian Spread narrows or even turns negative, and market confidence in "high AI investment, high valuation" may be shaken.

In other words, what can truly burst the AI bubble is never excessive investment scale, but when capital returns can no longer cover capital costs.

AI Capital Expenditures Near Peak Levels of Internet Bubble, But Valuations Are Far from Out of Control

US companies' capital spending on computers and peripherals—especially data centers, GPU servers, and related infrastructure—has surged to its highest level since the 1998-2000 internet bubble.

However, simply comparing the current AI cycle to the 2000 tech bubble is not accurate.

He points out that although capital expenditure growth is astonishing, the overall valuation of the tech sector "is not cheap, but is also far from the extreme stretching seen in early 2000." In other words, the market has indeed priced AI prospects highly, but has not engaged in a full-blown asset mania detached from fundamentals.

Compared to the internet era’s mode of "telling the story first, then finding profit," this round of AI investment is backed by much clearer business logic: cloud computing giants are engaged in substantial competition around computing power, model training, and enterprise AI applications, and related capital expenditure can directly translate into future market share and platform advantages.

The AI Dividend Is Spreading to the Asian Manufacturing Chain, US GDP May Not Benefit as Strongly as Expected

A key feature of this wave of AI investment is its highly globalized hardware supply chain.

Whether it’s GPUs, HBM high-bandwidth memory chips, server power supplies, advanced packaging, or data center components, the US relies heavily on imports. This means although US tech giants are the main drivers of AI capital expenditure, the direct economic dividends of that expenditure are actually spilling over to the Asian manufacturing system.

This is also why Korean memory chip companies and related Japanese tech firms have performed strongly in this cycle. Asian tech leaders including Samsung Electronics and SK Hynix have become key beneficiaries of global AI infrastructure expansion.

Meanwhile, this structure also means the direct impact of US AI investment on GDP may be lower than previously expected, because much of the equipment procurement is import-driven, not domestically produced value-added.

However, Denyer stresses that indirect effects remain significant. The wealth effect from US stock market gains, productivity improvements from greater AI-driven corporate efficiency, and investment in power and energy surrounding data center construction will still provide ongoing support for the US economy.

Two Major Risks Emerging: Inflation Shocks and Power Bottlenecks

Despite the overall optimistic trend, Gavekal highlights two potential risks facing the AI capital expenditure cycle.

The first is inflation and interest rate risk.

If Middle East tensions continue to push up oil prices—especially with risk over transportation in the Strait of Hormuz leading to further energy price increases—US inflation may come under renewed pressure. In such a situation, the Federal Reserve may keep rates high for longer, which would push up corporate financing costs and compress the Wicksellian Spread, weakening incentives to expand AI investment.

Secondly, there is the increasingly prominent power supply issue.

AI data centers' demand for electricity is growing exponentially, while the US power grid is expanding much more slowly. According to the US Energy Information Administration (EIA), US power supply growth will likely continue to lag demand growth in the coming years.

This means the spillover benefits of the AI boom will not be limited to semiconductors and cloud computing companies, but may further reach US power infrastructure, natural gas generation, and new energy fields. Investment opportunities around upgrading transmission networks, gas supply, and renewable energy construction are becoming AI cycle’s "second-layer trading logic."

Market Worries About Peaking Capital, But Potential ‘Ammo’ Remains Plentiful

To address concerns that AI capital spending cannot be sustained, Gavekal researcher Tan Kai Xian rebutted in another report.

He points out, there remains a large amount of potential funding that has not yet been activated.

First, about $7.6 trillion is parked in US money market funds. In a high-interest environment, this money has stayed in low-risk assets, but if risk appetite improves, some of it could flow back into tech and AI sectors.

Second, US banking deregulation is unlocking new credit space. Consulting firm Alvarez & Marsal estimates that deregulatory moves could add up to $2.6 trillion in new lending capability for banks, giving AI infrastructure financing room to expand.

Additionally, US government support for the AI industry is strengthening. Federal funding, tax incentives, and industry support policies may continue to flow into the AI sector in future.

Overall, Gavekal believes that although the current AI capital expenditure cycle is in a high-intensity stage, typical signs of a “bubble end” have not yet appeared. Corporate investment returns remain above financing costs, and funding sources have not dried up, meaning the AI boom still has a foundation to continue in the short term.

Risk Warning and Disclaimer ClauseThe market is risky, and investment should be approached with caution. This article does not constitute personal investment advice, nor does it take into account individual users' special investment objectives, financial conditions, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Invest at your own risk. ```