Is AI really headed for bubble trouble?
With current valuations of AI hyperscalers raising eyebrows, Cormac Lucey examines concerns that a brewing AI bubble may soon upend financial markets
Chartered Accountants carry out many tasks, mainly in financial control, auditing and advisory.
The foundational skill our profession uniquely possesses is knowledge of, and interest in, the details of how transactions are recorded—and it is important that transactions be properly accounted for.
Stephen Clapham, the founder of investor research consultancy Behind the Balance Sheet, recently published a digital newsletter with the headline: “Amazon’s AI reality check— what a simple accounting change reveals about the true cost of AI hype”.
In it, Clapham examined the assumptions made by hyperscalers, such as Alphabet, Amazon, Meta and Microsoft, regarding the useful economic life of computer servers.
Last year, Amazon extended server lives from five to six years, citing continuous improvements in hardware, software and data centre designs.
Earlier this year, however, it reversed this change “due to the increased pace of technology development, particularly in the area of artificial intelligence and machine learning”.
With this about-turn, Amazon is reducing its reported earnings by over $1 billion for the full year, a number Stephen Clapham believes “significantly understates the ongoing impact”.
Meanwhile, Meta is going the other way, recently extending server lives to five-and-a-half years, telling investors, “based on the servers and network assets placed in service as of December 31, 2024, we expect this change in accounting estimate will reduce our full-year 2025 depreciation expense by approximately $2.9 billion”.
In short, the profits of hyperscalers reported under Generally Accepted Accounting Principles (GAAP) are being significantly altered by modest changes in asset life assumptions. At the same time, resulting profit measures are growing at just a modest fraction of the growth rate claimed for non-GAAP estimates of forward operating profit.
Market participants generally use these future non-GAAP estimates of operating profit to justify elevated hyperscalers valuations—rather than historic measures of profit computed under GAAP rules.
There are some obvious reasons to question such valuations. For starters, there is no evidence that AI has significantly accelerated US productivity growth overall.
Investment researcher BCA recently published a chart, tracking earnings for Standard and Poor’s benchmark 500 Index over seven years from 2018 to 2025.
Fascinatingly, the chart showed that, while the index’s GAAP earnings have increased by 2.2 percent since December 2021, its trailing operating earnings have risen by 8.8 percent and its forward earnings per share by 21.7 percent.
This mismatch is concerning, as US share prices are driven more by forward earnings estimates than by historic results.
According to BCA Research, between 1982 and 2019, annual growth in US non-farm business productivity averaged 2.0 percent per year.
Post-ChatGPT—from the fourth quarter of 2022 to the first quarter of 2025—it has grown at 2.1 percent annually.
Meanwhile, there is the question of return on investment. During a recent Robert Half earnings call, the global recruiter’s Chief Executive Keith Wadell, stated “We know definitively that so far, AI has had very little impact on our revenues”.
With annual revenues of about $20 billion, OpenAI plans to invest a reported $1.4 trillion to grow its AI business over the next eight years.
Applying simple division, this means the US company behind ChatGPT will invest 8.75 times its current annual revenue each year for the next eight years.
This is a breathtaking commitment to investment in a technology we are as yet unsure could generate discernible productivity improvements for society, let alone whether these benefits would accrue to anyone beyond the companies selling the technology.
I don’t doubt that AI will eventually change our lives as much as the internet already has. But I also have no doubt that, in attempting to put a proper value on today’s AI companies, financial markets may have lost the run of themselves.
*Disclaimer: The views expressed in this column, published in the December 2025/ January 2026 issue of Accountancy Ireland, are the author’s own. The views of contributors to Accountancy Ireland may differ from official Institute policies and do not reflect the views of Chartered Accountants Ireland, its Council, its committees, or the editor.
