As AI capital spending spirals toward $600 billion annually, a growing chorus of investors warns that outdated depreciation assumptions are masking a coming profit collapse
The great AI infrastructure build-out has a dirty secret hiding in plain sight: those rows upon rows of Nvidia GPUs powering the artificial intelligence revolution may be financial time bombs on corporate balance sheets.
According to a growing body of analysis from Morgan Stanley, BCA Research, and prominent investors including “Big Short” protagonist Michael Burry, the major technology companies have dramatically underestimated how fast their AI hardware will deteriorate—and the resulting depreciation charges threaten to exceed their combined profits in each of the next five years .
The $600 Billion Question
When Amazon, Google, Meta, and Microsoft collectively cross $600 billion in annual capital expenditures, the market’s focus has shifted from model parameters to something far more mundane but potentially devastating: depreciation schedules .
The numbers are staggering. Morgan Stanley estimates that Microsoft, Oracle, Meta, and Alphabet alone will face cumulative depreciation expenses exceeding $680 billion over the next four years . This figure, derived from a modified depreciation model that accounts for the lag between capital spending and asset activation, suggests traditional forecasting methods are significantly underestimating the coming earnings drag .
BCA Research Global Chief Strategist Peter Berezin paints an even more alarming picture. By his calculation, hyperscalers will hold approximately $2.5 trillion in AI-related assets by 2030. Assuming a 20% depreciation rate—reflecting realistic five-year asset lives—annual depreciation would hit $500 billion. That exceeds the trailing 12-month cumulative profits of these companies, which stood at $377 billion as of the third quarter of 2025 .
“This figure exceeds the trailing 12-month cumulative profits of hyperscalers,” Berezin noted, warning that investors may have reached a “metaverse moment” where rising capital expenditures are finally being punished in equity markets .
The Great Depreciation Debate
At the heart of the controversy lies a seemingly technical accounting question: How long do AI chips actually last?
Currently, most hyperscalers depreciate their GPU investments over five to six years, reflecting traditional data center equipment lifetimes. But critics argue this bears no relation to reality in an industry where Nvidia releases new architectures on an accelerator clock .
Michael Burry has been characteristically blunt. In a social media post that sent ripples through the investment community, he estimated that technology giants will underestimate depreciation by $176 billion between 2026 and 2028. His reasoning: these companies are artificially inflating earnings by extending the “useful life” of assets that become obsolete in two to three years .
“Some companies even extend the depreciation cycle to six years,” Burry noted, specifically calling out Oracle, whose profits he estimates could be overstated by 26.9% by 2028 if current accounting practices continue. Meta’s profits could be inflated by 20.8% over the same period .
Sparkline Capital founder Kai Wu argues that given Nvidia’s GPU upgrade cycle, two to three years is a far more realistic estimate for AI hardware lifespan. Under that accelerated depreciation scenario, he projects annual depreciation charges could surge from $150 billion today to $400 billion within five years .
“Although the ‘Magnificent Seven’ are extremely profitable,” Wu wrote, “as depreciation expenses from surging capital expenditures begin to be recognized, their net profits will be dragged down in coming years” .
Who’s Most at Risk?
Not all tech giants face the same exposure. A quantitative audit by Tradingkey, reported through Mitrade, reveals stark divergence in financial health across the major players .
Microsoft emerges as perhaps the most vulnerable among the top-tier names. While the company boasts impressive incremental operating margins of 58.9%, its asset burden ratio—depreciation and amortization as a percentage of revenue—stands at 14.63%, dramatically higher than Google’s 4.91%. The report describes Microsoft as facing “severe growing pains from its transition to an asset-heavy model,” with profit margins “being eroded by rising infrastructure depreciation” .
By contrast, Google (Alphabet) appears better positioned, thanks to its self-developed TPU chips that insulate it from third-party hardware price hikes. With an asset burden ratio of just 4.91% and revenue conversion velocity of 92%, Google earns the highest financial health score among the group .
Meta sits in the middle, with strong incremental margins of 42.18% but warning signs that its AI ad recommendation returns may be approaching a short-term ceiling .
Oracle represents the most extreme case. With an incremental margin of just 9.70% and free cash flow coverage ratio of only 0.65x—well below the 1.0x survival line in a high-interest environment—the company is described as engaging in “hemorrhaging expansion supported by leverage” .
Amazon shows the most intriguing profile: a low static score of 4.7 but evidence of “bottom reversal” momentum, suggesting its massive investments may be crossing the “most difficult depreciation adjustment period” .
Beyond the Magnificent Seven
The depreciation dilemma extends beyond the largest U.S. tech names. Analysis of the Australian tech sector by The Investor Standard shows that high-quality names like Xero, WiseTech Global, and NextDC suffered drawdowns ranging from 20% to more than 50% during 2025, despite resilient earnings .
For infrastructure-focused companies like NextDC, which continues heavy investment while free cash flow remains negative, the market has imposed a new rulebook. “Free cash flow discipline now matters more than top line momentum, particularly while rates remain restrictive,” the analysis notes .
Historical Parallels
Kai Wu draws provocative historical parallels, comparing current AI infrastructure spending to the railroad building craze and the internet bubble. Relative to GDP, AI capital expenditure already exceeds internet bubble-era peaks. When adjusted for the rapid obsolescence of AI chips—far faster than railroad assets—today’s investment intensity ranks at the very top of historical comparisons .
The implication is sobering. Railroad overbuilding led to massive write-downs and consolidation. The internet bubble produced a decade of indigestion for telecommunications companies that overbuilt fiber networks. AI may prove no different.
Wall Street’s Blind Spot?
Despite these warnings, the depreciation thesis has yet to gain widespread acceptance on Wall Street. Bernstein analyst Stacy Rasgon maintains that “GPU profitable operating cycles are about six years” and that most hyperscalers’ current depreciation methods are reasonable .
But the skeptics are growing louder. Bank of America analyst Justin Post argues that Wall Street has been “slow to react” to the coming depreciation wave. He estimates that for Google, Meta, and Amazon alone, consensus depreciation forecasts could be understated by nearly $16.4 billion by 2027, meaning actual profitability will fall significantly short of current expectations .
The Bottom Line
For investors, the depreciation debate is not an accounting abstraction. If the skeptics are right, tech giants face a multiyear period in which reported earnings are crushed by catch-up depreciation charges, even as cash flows remain strong.
Morgan Stanley’s Katy Huberty emphasizes that the shift from relatively asset-light business models to infrastructure-heavy operations has profound accounting implications that markets have yet to fully price .
Peter Berezin of BCA Research puts it more starkly: when you see dangerous amounts of snow accumulating at the summit, you must decide whether to leave or continue enjoying the view .
With the Nasdaq 100 already down 6.3% and the Technology Select Sector SPDR fund off more than 9% in recent weeks, some investors appear to be heading for the slopes . The question is whether the avalanche is still gathering force.
The cumulative picture emerging from this analysis suggests total depreciation-driven losses across major tech companies could approach or exceed $1 trillion over the 2026-2030 period, potentially wiping out a significant portion of sector profits in each of those years. For Microsoft, Oracle, and other heavily exposed names, the earnings drag from realistic depreciation schedules could transform current margin heroes into cautionary tales of capital allocation.