Here's an uncomfortable math problem for the AI revolution: What happens when the bill comes due?
Peter Berezin, Chief Global Strategist at BCA Research, dropped a sobering analysis on X late Thursday that's got investors reconsidering the economics of the AI arms race. His calculation is straightforward but striking. By 2030, major cloud and AI infrastructure providers - the hyperscalers like Microsoft (MSFT), Alphabet (GOOG), and Amazon (AMZN) - could be sitting on more than $2.5 trillion in AI assets.
When Depreciation Exceeds Profits
Apply a standard 20% depreciation rate to that mountain of hardware, and you're looking at around $500 billion per year in depreciation expenses later this decade. That's not a small number. In fact, Berezin points out it would actually exceed the combined projected profits for 2025 across these companies. Which raises an obvious question: How sustainable is this spending spree?
The timing of Berezin's warning was notable. It came on the heels of a brutal Thursday session that saw tech stocks tumble despite Nvidia (NVDA) delivering strong earnings. The initial euphoria over Nvidia's results quickly faded as investors refocused on concerns about sky-high AI valuations and the breakneck pace of infrastructure investment by Amazon, Meta (META), Oracle (ORCL), and others building out the data centers needed for generative AI.
The Nasdaq Composite closed down 2.2% after an early rally evaporated, while the S&P 500 dropped 1.6%. The selling pressure rippled into Asian markets Friday, where major indexes also retreated.
Dance While the Music's Playing
So why would companies keep pouring money into AI infrastructure if the economics look this shaky? When one X user posed exactly that question, Berezin offered a knowing response: "If the music is playing, you have to keep dancing." It's a reference that anyone who remembers the financial crisis will recognize - former Citigroup CEO Chuck Prince famously used those words in 2007, right before everything collapsed.
The reaction to Berezin's projection was predictably mixed. Some observers worried that depreciation could crush profit margins if AI revenue growth doesn't materialize as expected. Others pushed back, noting that hyperscalers have options. They can extend the useful life of equipment beyond standard depreciation schedules, or repurpose older hardware for less demanding computational tasks, effectively spreading out or reducing the financial impact.
Still, Berezin's analysis adds another layer to the growing debate about whether the AI boom is built on solid economics or inflated expectations. The companies making these massive bets are banking on AI generating enough revenue to justify the spending. But with depreciation costs potentially outpacing profits, the margin for error looks uncomfortably thin.
Wall Street is now wrestling with a classic problem: Nobody wants to be the first to stop dancing, but somebody's going to be holding a very expensive bag when the music stops.