U.S. corporate profits hit a record $4.7 trillion in the 12 months through mid-year, up 17%. Meanwhile, just 42 AI-related companies accounted for around two-thirds of the S&P 500’s gains since early 2024, according to WNYC.

Bulls call it an AI revolution. BCA Research’s Peter Berezin and Minack Advisors’ Gerard Minack see an accounting asymmetry instead.

“Nvidia (NASDAQ:NVDA) earns a big profit whenever it sells a GPU,” Berezin wrote.

“Whoever buys that chip, however, does not incur an immediate cost because the purchase is treated as capital spending.” Buyers depreciate the chips over roughly six years, while Nvidia books the full sale upfront. Cash flows don’t change, but reported earnings rise.

“This is probably the most earnings accretive capex boom in history,” Jefferies strategist Christopher Wood noted, according to Reuters.

The $1 Trillion Capex Mirage

Depreciation doesn’t start until a data center comes online. Morgan Stanley estimates that as much as 40% of this year’s AI capex won’t be operational until 2027–2029. By decade’s end, hyperscalers’ annual depreciation charges will run between $500 billion and $1 trillion, according to Capital Gearing Asset Management.

Paper gains add a second boost. Tech giants book changes in the value of their stakes in unlisted labs like OpenAI and Anthropic as “other non-operating income.” In the four quarters to mid-2026, annualized earnings at Alphabet Inc. (NASDAQ:GOOGL), Amazon.com (NASDAQ:AMZN), Microsoft Corp. (NASDAQ:MSFT), and Meta Platforms (NASDAQ:META) rose by $222 billion.

Over the same period, that non-operating line rose by $224 billion, which accounts for all of their reported profit growth.

Minsky’s ‘Ponzi Finance’ Hits the Neoclouds

This year the hyperscalers are putting almost all of their operating cash flow into capex. They’ve issued about $225 billion of unsecured bonds, roughly 30% of full-year capital spending. They also hold $1.1 trillion in lease commitments that stay off their balance sheets.

Morgan Stanley forecasts the group will have negative free cash flow of about $550 billion next year.

Economist Hyman Minsky used the term “Ponzi finance” for businesses that can’t cover their outgoings from cash flow and rely on rising valuations to raise capital. Loss-making data center operators and frontier labs fit that description.

The ‘Profits Bubble’ Trap

Tech stocks don’t look expensive relative to earnings. “The bubble is in profits, not valuations,” financial historian Edward Chancellor argued for Reuters Breakingviews. U.S. homebuilders were in the same position in 2006, and global miners were in the same position when the commodity supercycle stalled in the mid-2010s.

He points to Minack’s warning that, when the capex boom ends, earnings growth could revert to trend. Tighter credit could make the $550 billion cash shortfall hard to fund. Falling valuations for neoclouds and private labs would reverse the paper gains, and the 2027–2030 depreciation wall would start hitting income statements.

The irony, as WNYC notes, is that fierce price competition among AI labs could crush tech margins. The same competition could make automation cheap enough for cash-strapped firms to cut white-collar jobs in the downturn that follows.