Last month, circularity became the word keeping investors up atnight, with investors haunted by visions of Big Tech's complicated web of
circular dealmaking creating outsize risk for markets.
Enter a new boogeyman: depreciation.
Fears that expensive GPUs and semiconductorchips that firms are buying up will lose value more quickly than they expect —
therefore becoming more of a cost burden and weighing on earnings — are roiling
the AI trade. The Nasdaq 100 is down 6.3% over the last few weeks, while the
Technology Select Sector SPDR Fund has fallen more than 9%.
Famed short sellers Michael Burry and Jim Chanos have brought depreciation worries to thefore in recent weeks as a key reason they're skeptical of the AI trade.
"By my estimates they will understatedepreciation by $176 billion 2026-2028," Burry recently wrote on X about
the Big Tech hyperscalers, saying he estimates chips will have a two to three
year lifecycle, not the roughly six years firms anticipate.
It's not just big name short-sellers,though, who are nervous about the depreication issue.
"Why are stocks falling? Because ifyou take the numbers in this chart seriously, the hyperscalers will hold at
least $2.5 trillion in AI assets by the end of this decade," Peter
Berezin, the chief global strategist at BCA Research, wrote in a LinkedIn post
on Thursday. "Assuming a depreciation rate of 20%, that would generate
$500 billion in annual depreciation expense. This is more than their combined
profits for 2025."h
Kai Wu, founder and Chief InvestmentOfficer of Sparkline Capital, said in a recent report that depreciation values
could rise from $150 billion a year to $400 billion in the next half decade.
"While the Magnificent 7 are extremelyprofitable, their net income will be dragged down over the next few years once
depreciation charges from their surging capital expenditures kick in," Wu
wrote.
He continued: "Many analysts believe that the hyperscalers' 5-6 yearuseful life assumption for AI data centers is overoptimistic, with 2-3 years
more appropriate given Nvidia's accelerating GPU replacement cycle."
Adjusted for that faster depreciationtimeline, Wu put current AI spending levels in context with the railroad and
internet booms.
"Relative to GDP, current AI spendingalready exceeds the peak achieved in the Internet boom," he wrote.
"While it remains below the peak attained in the railroad buildout, the
useful life of AI chips is much shorter than that of railroads. If we adjust
for faster depreciation, today's AI buildout tops the chart."
This new depreciation argument isn't yet widely accepted onWall Street or in the AI industry itself — few mainstream strategists seem to
be warning of such a threat yet.
"GPUs can profitably run for about 6years," said Bernstein analyst Stacy Rasgon in a November 17 client note.
"The depreciation accounting of most major hyperscalers is
reasonable."
But for now, the prospect of the theorybeing correct seems to be throwing a wrench into the once red-hot AI trade.