Hyperscalers Face $1 Trillion Depreciation Bomb by 2030

Capital Gearing Asset Management forecasts that hyperscalers' annual depreciation expenses will reach $500 billion to $1 trillion by the decade's end. Depreciation begins only once a data center starts operating. Morgan Stanley estimates that up to 40% of this year's AI capital spending will not become operational until 2027-2029.

U.S. companies generated record profits of $4.7 trillion over the 12 months through mid-year, an increase of 17%. WNYC reported that 42 AI-related companies produced around two-thirds of the S&P 500's gains since early 2024. Optimistic investors view the developments as an AI revolution. However, BCA Research's Peter Berezin and Minack Advisors' Gerard Minack see an imbalance in accounting treatment instead.

Berezin wrote that Nvidia earns substantial profits from GPU sales, while buyers classify chip purchases as capital spending rather than an immediate expense. Buyers spread depreciation over roughly six years, while Nvidia recognizes the entire sale immediately. The accounting treatment increases reported earnings without changing cash flows. Reuters reported that Jefferies strategist Christopher Wood described the capital spending boom as probably the most earnings-enhancing in history.

Technology giants also record valuation changes in stakes in privately held labs, including OpenAI and Anthropic, as other non-operating income. Across the four quarters through mid-2026, annualized earnings at Alphabet, Amazon.com, Microsoft and Meta Platforms increased by $222 billion. Other non-operating income at those companies increased by $224 billion over the same period, accounting for the entirety of reported profit growth.

Hyperscalers are directing almost all operating cash flow into capital spending this year. Hyperscalers have issued about $225 billion in unsecured bonds, equivalent to roughly 30% of full-year capital spending. Hyperscalers also have $1.1 trillion in lease commitments outside their balance sheets. Morgan Stanley forecasts that hyperscalers will generate about $550 billion in negative free cash flow next year.

Economist Hyman Minsky defined "Ponzi finance" as businesses relying on rising valuations to raise capital because cash flow cannot cover expenses. Loss-making data center operators and frontier labs meet that description.

Technology shares do not appear expensive compared with earnings. However, financial historian Edward Chancellor argued in Reuters Breakingviews that profits, rather than valuations, contain the bubble. U.S. homebuilders faced a similar situation in 2006. Global mining companies faced a similar situation when the commodity supercycle faltered in the mid-2010s.

Chancellor cited Minack's warning that earnings growth could return to trend once the capital spending boom ends. Tighter credit could make financing the $550 billion cash shortfall difficult. Lower valuations for neoclouds and private labs would undo the gains recorded on those investments. Depreciation charges would begin affecting income statements during 2027-2030.

WNYC noted that intense price competition among AI labs could sharply reduce technology companies' margins. WNYC also noted that the competition could make automation affordable enough for financially strained companies to eliminate white-collar jobs during the subsequent downturn.