Wall Street has a new obsession — and it isn’t the good kind. One word — a single accounting term that practically never trends anywhere outside of finance textbooks — is suddenly the most feared in the markets. And it is striking the most sensitive spot for AI stocks: valuations. The word is “depreciation,” and over the last few weeks, it has transformed from a boring accounting line item to the central villain of the most powerful tech rally in years.
For investors who have spent the last year and a half racing up the price of just about any stock related to AI, they are now faced with an uncomfortable realization: The infrastructure behind this latest AI boom might break apart much sooner than anyone anticipated. “Whether it’s high-end GPUs, custom AI chips, servers and racks or cooling systems, the tens of billions of dollars that companies have put into data centers following a model in which they were used for five to seven years may no longer be effective,” he said. Others who are high-profile skeptics claim they won’t even make it through three. And in markets, three years and seven are a world of difference.
At issue is a growing chorus of warnings from short sellers and longtime analysts who say that Silicon Valley has been drastically underestimating the pace at which AI hardware becomes obsolete over the past year. As depreciation accelerates, companies must record larger expenses each year, cutting earnings, and when expectations for earnings fall, stock prices come down. Suddenly, the math of that A.I. trade doesn’t seem so bulletproof as it was last season in 2023-24.
Nowhere is the unease more apparent than in Big Tech’s spending. Microsoft, Google, Amazon, Meta, and Oracle have together invested hundreds of billions in AI data centers, with analysts predicting further aggressive expenditures. But the fates of those investments are now the topic that could reshape the debate. If data center hardware becomes obsolete earlier than projected, that means the costliest part of the A.I. revolution needs to be swapped out regularly — not just now and then. That turns AI from a high-margin dream into an industrial business more familiar to Wall Street analysts — one with hefty recurring costs.
Fanned by what's happening inside major cloud companies, the fear is all the greater. Executives are hurrying to update GPU clusters with the next generation of chips, which promise greater efficiency, lower power consumption, and higher throughput. Nvidia’s rapid release cycle, by itself, has been a bit of an excuse to feel a sense of urgency. Some companies that just upgraded in early 2024 are already planning updates for 2025 and 2026. For investors, that raises a tricky question: If the hardware is outperformed every time Nvidia introduces a new architecture, then all bets are off on depreciation schedules.
That hard-to-imagine future is one that some analysts say might “break” the AI-profit story. The industry’s titans might continue to own the market, but if the cost of staying competitive goes through the roof, margins will come down. The technology itself may be transformative, but the economics start to feel more like heavy manufacturing than software — and valuations at the level of software do not survive manufacturing-style expenses.
The market response has been immediate. Tech ETFs have sharply underperformed over the last few weeks as institutional investors reconsider the longer-term earnings trajectory of AI-led names. A significant factor in driving the correction is simple: The market priced AI as if enormous capital spending would plateau. And now it looks like they’re not going to level off at all — they may only accelerate.
But the fear of depreciation has another dimension that’s worrying Wall Street. When hardware life cycles shorten, it doesn’t just notch down profits; it changes the way companies invest altogether. The more an investment is depreciated, the lower a company’s reported earnings — and with them its taxable income — even if revenue would have been superstrong. Suddenly, those price-to-earnings ratios on AI stocks — which were a bit stretched to begin with — look even more puffed up.
It’s why “depreciation” has become the new dirty word in markets. It’s also a term that investors who were on the AI bandwagon are being forced to reckon with, learning anew that the AI boom isn’t entirely digital magic; it is built on physical machines that ultimately wear out, get old, and fall behind. It’s a reminder that the AI revolution is more than just futuristic narratives; it runs on literal hardware.
It also lays bare a gulf between public enthusiasm and financial reality. The typical investor thinks of AI as software — infinitely scalable, low-cost, high-margin. But real AI at scale resides in giant buildings filled with servers that suck up unbelievable amounts of electricity. And servers decay. They rust. They slow down. They break. They need maintenance. They need replacement. Meanwhile, GPUs are advancing so quickly that hardware that’s two years old is in danger of becoming obsolete. A cycle like that is brutal for any balance sheet, no matter how large you might be, as Microsoft certainly is.
Some analysts say the panic is unfounded. Big Tech, they say, has the financial muscle to swallow accelerated depreciation and keep investing at scale as if it were no big deal. But even if that remains the case today, investors are forward-looking. The question is not whether Microsoft or Amazon can afford it next quarter — it’s whether the AI boom will generate as much profit as investors had expected five years from now.
Others liken it to telecom companies from the early 2000s, who built fiber networks with abandon that took years to cover their costs. The infrastructure proved crucial in the end, but those who expected a quicker payoff were left behind. AI may eventually follow a similar path: the technology could shape the future, but its short-term payoff is likely to be much smaller than the sunniest forecasts assert.
And so this may be why depreciation has become the scariest word on Wall Street. It requires analysts to grapple with the idea that the AI trade — though real — is more financially messy than a stack of glossy pitch decks makes it out to be. The stocks could still continue to climb in the long term. The companies may yet come to dominate the world economy. But perhaps the path may be littered with costly hardware cycles, tightening margins, and erratic earnings.
The debate over deprecation is just getting started. Investors are adjusting models; chief financial officers are updating their forecasts; artificial intelligence companies are working to reassure shareholders that infrastructure spending is an investment in long-term returns. But the pressure is already altering sentiment. Or as one strategist put it to me this week: “AI is the future, but the future is expensive.”
In short, the AI boom is not ending. But the quick cash might be over. And the least-liked word on Wall Street? It’s a dull accounting term with the potential to determine the fate of an entire industry: depreciation.
