Are We in an AI Earnings Bubble? 3 Warning Signs to Watch
The artificial intelligence infrastructure boom has delivered significant valuation surges in stock prices for the last few years.
Along with that growth, a major debate has taken hold between those who believe we are in an unsustainable AI bubble that is destined to pop and those who believe stocks are attractively priced.
It’s common to evaluate this economy against past cycles, with the dot-com era as the one most often invoked. A recurring argument from major investment houses and everyday investors is that the earnings of the AI companies justify these current valuations.
Sign up for Kiplinger’s Free Newsletters
Profit and prosper with the best of expert advice on investing, taxes, retirement, personal finance and more – straight to your e-mail.
Profit and prosper with the best of expert advice – straight to your e-mail.
However, I have doubts about the durability of those profits and believe the risk of an “earnings bubble” is higher than some give credit for.
During the dot-com era, P/E ratios rapidly increased. Stock prices shot up without a corresponding growth in profits, resulting in very high P/E ratios. When it became clear that earnings growth was not coming at the level the market had assumed, price had nowhere to go but down. This was a “price bubble.”
Currently, P/E ratios are nowhere near the levels of the early 2000s. Prices have gone up, but earnings have also seen an equally impressive increase. This has led many to conclude that we aren’t in a bubble.
But what if what we’re seeing is an “earnings bubble”? Unlike in the dot-com bubble, price would not drop rapidly to get back in line with net income, but price and earnings would find themselves falling instead.
Three major reasons I’m concerned with earnings sustainability are hyperscalers’ free cash flows, alleged accounting adjustments and circular financing.
Free cash flow
Hyperscalers are the large-scale technology companies that provide cloud computing and are leading this AI buildout. They have been free cash flow (FCF) machines for years. These hyperscalers have now gone FCF negative, spending more than they are bringing in.
That cash is now going to the semiconductors whose chips are powering the data centers. Earnings have skyrocketed for these semiconductors, leading many to theorize that the rapid increase in stock prices for these semiconductor companies is justified owing to their higher earnings.
If the hyperscalers don’t recognize enough AI revenue to justify their investment, that spending will eventually significantly slow. That slower spending then hits the semiconductor earnings, causing a quick and large deflation in earnings and price.
As I’ll explain, the current gap between hyperscaler spending and realized AI revenue leaves little margin for error.
Alleged accounting adjustments
Beyond negative FCFs, the quality of reported tech earnings has come under question. Hyperscalers have been accused of lengthening the assumed useful life of their servers and networking hardware from three or four years to five or six.
Spreading the cost of high-end AI infrastructure over a longer horizon would artificially depress annual depreciation expenses, resulting in higher, and questionable, reported operating profits.
And constantly swapping out hardware while income statements drag out the cost over five to six years would effectively hide the true cost of the buildout and inflate near-term earnings.
Nvidia has pushed back on the allegations, stating that “customers depreciate GPUs (graphics processing units) over four to six years based on real-world longevity and utilization patterns.”
Circular financing
Critics argue that circular financing leads to artificially increased revenue and earnings. It occurs when a company uses its financial leverage to invest in the very companies that purchase its products.
One example of circular financing in the AI sector is illustrated by Nvidia investing billions of dollars into cloud provider Nebius.
In return, Nvidia gets an equity stake in Nebius and enables Nebius to immediately turn around and spend billions on Nvidia GPUs, which Nvidia then logs as top-line revenue. The return of Nvidia’s cash causes an increase in its revenue and earnings. This arrangement is going on throughout the AI ecosystem.
This closed loop is further amplified by Nvidia’s equity stake in Nebius. As Nebius’ valuation increases, Nvidia’s shares of the company further increase its financial metrics, making Nvidia look even more attractive.
This circular financing replaces organic market growth with supplier-funded capital, making it difficult to determine how much of the earnings is genuine and sustainable.
The bottom line
AI appears to be transformative technology in the long term. However, the earnings growth seen the last few years does not, as some argue, decrease the risk of a bubble. Many great technological innovations resulted in bubbles, including railroads and the internet.
For AI to avoid a bubble, hyperscalers will need to start seeing revenue that justifies their cost of capital. They have a long way to go, though. In mid-2026, estimates put the AI revenue-to-capex ratio at 4% ($25 billion/$660 billion). That is AI revenue, not profit. It doesn’t mean it won’t happen, but the risks are highly elevated until it does.
