The Debt-Fueled AI Race – How to Deal with a Potential Bubble Burst?

One of the most profound innovations in human history, generative AI is set to drastically change how knowledge professionals work and significantly increase productivity. Indeed, annual productivity growth is projected to accelerate to 3.8% in the next five years, a near-doubling of the long-term average of 2%. But while AI’s economic impact is promising, society may not be ready for a sea change yet.

Since the launch of ChatGPT in November 2022, capital expenditures on AI infrastructure by deep-pocketed hyperscalers’ have grown more than 60%, driven by fear of missing out on what is potentially the biggest tech trend in our lifetime. What’s more, according to announcements in the most recent earnings season, spending is expected to increase by another 30% in 2026, topping $500B for the first time.

This whopping investment has become a major contributor to the economy, driving 40% of US GDP growth. It also has also boosted AI picks-and-shovels stocks, which have accounted for up 80% of the market’s YTD gain. Although the debate over spending sustainability remains, the market has put those concerns on the shelf for most of the year, with mega-techs continuing to generate enormous cashflow via their core business and start generating incrementally visible AI service revenue.

However, it turns out that even the most cash-laden mega techs can’t afford such unprecedented spending indefinitely. For two quarters in a row, the Magnificent Seven reported negative free cash flow growth, while the S&P Technology sector’s cash as a percentage of total assets dropped to its lowest level since 1996. Moreover, the market sentiment deteriorated quickly as mega-techs turned to big-time borrowing on the heels of their earnings reports in late October.

The sudden increase in debt has finally made the market uneasy. Both Oracle and Meta shares dropped more than 25% from their September peak after the newly announced debts—Oracle borrowed $56 billion in October alone. For those who remember Enron’s epic collapse, Meta and AI’s use of off-balance sheet debt ($27 and $12 billion respectively) seem particularly alarming and are seen as the canary in the coal mine.

That said, worrying about the mega-techs’ liquidity seems unnecessary, since their core businesses remain powerful cash cows. The biggest systematic overhang is that these whopping investments hugely hinge on whether OpenAI, which won’t turn profitable before 2029 at the earliest, can make good on its $1.4 trillion commitment to compute power investments. Whenever the first sign of either hardware overcapacity or OpenAI failing to secure more funding looms, the market could act very negatively and the high-flying AI arms dealer stocks’ multiple should be compressed significantly, just like what we saw in dot com era.

The opportunity and risk of AI is unprecedented, , we admittedly don’t have a crystal ball to foresee how things will play out in the near term. Therefore, in order to capture this unparalleled revolution while preserving client’s capital in case of a bubble burst, when it comes to portfolio construction, we stick to our GARP (Growth at Reasonable Price) philosophy, maintaining a reasonable AI exposure and avoiding overly high-correlation constituents.

Typically, we use the three-category framework below when allocating capital:

• Direct AI Enablers and Beneficiaries: The ones that own either quasi-monopolistic technology or a leading application platform with idiosyncratic mitigants to overcapacity risk.
• Indirect AI Beneficiaries: The ones whose underlying demand is solid regardless of incremental AI spending—for example, electrification equipment vendors, US grid builders and electricity suppliers.
• No or Low-AI Correlation Groups and Assets: Healthcare and gold.

Just like any other revolutionary innovations in history such as the PC, the Internet and the smartphone, there will certainly be turbulence on this AI journey. To get through different scenarios, sticking to companies with the strongest fundamentals and cashflow generation and applying prudent risk management to portfolio construction should still be the best way down the road.