Recent earnings report by Oracle Corp. could be the canary in the coal mine.TOM BRENNER/Reuters
I was delighted to see the latest research from the global asset-management firm Verdad this past month, as it deals with an area I have felt strongly about and written about in the past: the long-term performance of high-asset-growth firms.
It was motivated by academic research showing that such firms underperform in the long run. For example, researchers at the Darden School of Business in 2009 looked at the stock performance of high-asset-growth firms and compared it with the performance of low-asset-growth firms between 1968 and 2007. What they found was that low-growth firms had an average return of 26 per cent annually, while high-growth firms returned a meagre 4 per cent. The low-growth firms outperformed the high-growth firms by a whopping 22 percentage points, annually on average, over that 40-year period.
High-growth stocks tend to attract a lot of attention and a lot of trading by investors, and they thus tend to become overvalued, leading to low returns going forward.
Remember the stories of the late 1990s?
During the dot-com era, telecom infrastructure needed to be built out, but at a great cost in terms of capital spending. Assets grew, but progress was slow, revenues lagged and the whole industry eventually imploded.
George Athanassakos: Is the ‘January Effect’ a slam dunk this year?
George Athanassakos: Three ugly truths that are keeping me up at night these days
That brings me to the asset growth plans of today’s artificial-intelligence-related tech companies.
It is true that companies nowadays have proven track records of growth and cash-flow generation, which was not the case in the late 1990s. But the other argument that proponents of a bull AI-driven market bring forth – namely, that these companies have solid balance sheets – is questionable.
What we have now is that companies that used to be light capex (capital expenditures) are becoming heavy capex, and companies that used to have little or no debt are now heavily indebted. A wave of debt raising by tech companies investing in AI has blanketed the private and public debt markets.
And it is not just on-balance-sheet debt, but also off-balance-sheet debt through special-purpose vehicles; does anyone remember Enron? Morgan Stanley estimates that companies such as Amazon, Meta, Oracle, Microsoft and Alphabet will need to spend a combined US$2.9-trillion on AI infrastructure by 2028 with US$1.4-trillion coming from debt.
This raises the question: Asset growth obviously will be there, but what if these hyperscalers do not earn as much revenue as expected over the next few years?
Could the recent earnings report by Oracle Corp. be the canary in the coal mine? The company reported US$248-billion of lease-payment commitments, mostly related to AI infrastructure such as data centres, which nevertheless were not included in the company’s balance sheet. This is concerning for the markets because, as Bloomberg columnist Chris Bryant recently articulated, there is “mismatch between the long duration of the property leases and much shorter contracts with key customers such as OpenAI.”
History is not on the side of market optimism about asset-growing companies’ stock performance in the long run. And Verdad’s research, examining financial metrics of light and heavy capex companies over the period from June, 1997, to June, 2025, using S&P Capital IQ data, provides further proof of that, namely that asset growth and stock returns are negatively correlated.
Verdad’s key takeaways are as follows:
- Firms with the highest capex spending tend to generate robust EBIT (earnings before interest and taxes) growth over the next three years (5.2 per cent annualized), and so do firms with the highest free cash-flow yields (1.6 per cent annualized).
- But firms with the highest capex spending tend to have lower return on assets (ROA).
- Firms with the highest free cash-flow yield tend to have higher ROA.
- And ROA is a robust predictor of returns over one-, two-, three- and four-year horizons.
It concludes by stating that, “by choosing to pursue earnings growth through aggressive spending on data centres with uncertain ROAs, [the hyperscalers] seem to be taking a gamble on their valuations.” Moreover, it goes on to say that the hyperscalers’ transitioning from light-capex models to heavy-capex models with uncertain margins could cause their ROA on new investments in data centres to be lower than the ROA of their existing software business. And this will lead to lower stock returns for these companies going forward.
So, if you want to invest in high-asset-growth firms, you must first look at the evidence and understand the history, except if you believe that this time is different!
George Athanassakos is a professor of finance and holds the Ben Graham Chair in Value Investing at the Ivey Business School, Western University. His latest book is Value Investing: From Theory to Practice.