"Impairment" is about to show up in a lot of AI headlines. Here's what it means in plain words, and what Gartner actually said.
Three words to know
- Capex: cash spent on long-lived things like servers, GPUs and data centers.
- Depreciation: that cost, spread out over the asset's useful life as a yearly expense. A $5M server rack with a 5-year life costs $1M a year on paper.
- Impairment: a one-time write-down when an asset is worth less than its remaining book value, for example if chips go obsolete early or a data center sits empty. It hits profit right away, but the cash was already spent.
What Gartner said (Oct 14, 2026)
From Gartner's press release:
- AI providers are expected to spend nearly $1 trillion on AI infrastructure in 2026 and $6 trillion more from 2027 to 2030.
- Data center owners would need $8.5 trillion of cumulative revenue through 2030 to cover that spending, financing and a 12% return on invested capital.
- Gartner forecasts AI revenue attributable to data centers of no more than $3.6 trillion over that period.
- Result: over $3.5 trillion of spending could be exposed to impairment by the end of 2030, and more than $4 trillion of asset value could face write-down risk through 2032.
- It predicts over 80% of neocloud providers will cease independent operations by 2030 (neoclouds explained).
"Failure to achieve sustainable ROIC on rapidly depreciating assets would almost certainly necessitate write-downs," Gartner wrote. It doesn't expect the 12% ROIC hurdle to be reached until the end of 2032 at the earliest. It's a scenario based on the current pace of spending, not a call on any one company.
Why useful life matters so much
How long a company says an asset lasts decides how fast its cost hits earnings.
- On its Jul 29, 2026 call, Microsoft said it's extending the useful lives of data centers and office buildings from 15 to 25 years, effective fiscal 2027. It said the change affects only the timing of depreciation, with "a minimal benefit" to FY27 operating income; the bigger effect is that more future data center leases become operating leases, which aren't counted in capex.
- Alphabet said on its Jul 22 call that about 60% of its Q2 technical-infrastructure spending went to servers, which wear out much faster than buildings, and warned of higher depreciation.
GPUs are the fast-aging part. When a new generation ships, older chips can fall in value. That's the risk inside GPU-backed financing and the off-balance-sheet deals in the $10 trillion financing question.
How to spot it in a report
- Search the 10-Q or 10-K for "impairment" and "useful lives." Investor.gov explains where the financial statements and MD&A sit.
- Compare depreciation growth to revenue growth.
- Watch for non-GAAP figures that exclude impairments (EPS and adjusted numbers).
- Check lease commitments and guarantees in the notes.
For the bigger debate on returns, see the AI bubble capex math.
