How are prepaid GPU or token credits treated on the balance sheet?

As a prepaid asset at purchase, unwound to expense or cost of revenue as the credits are consumed. Unused expiring credits are written off to expense.

Prepaid assetPrepaid inference credits

Prepaid inference credits

The posting

A prepaid block of inference or token credits is a prepaid asset at purchase, because you have paid ahead of consuming anything. As tokens are drawn down, the prepaid unwinds to the P&L: to operating expense for internal use, or to cost of revenue when the inference is a direct input to a paid product ASU 2018-15. It is not capitalized as an intangible.

At purchase

Prepaid assetIllustrative example, not client data

CapEx / Balance sheet

OpEx / P&L

DrPrepaid compute (asset)X

CrCashX

Credits bought ahead of use are set up as a prepaid asset. Figures illustrative.

As tokens are consumed

Each period, reclassify the consumed portion out of prepaid. If the inference serves a paid product it is cost of revenue; otherwise it is operating expense. Measure consumption on the cost-per-million-tokens basis you were billed on, so the release ties to actual usage rather than straight-line time.

Cost of revenueIllustrative example, not client data

CapEx / Balance sheet

OpEx / P&L

No entry

DrCost of revenueY

CrPrepaid compute (asset)Y

As credits serving a paid product are consumed. Figures illustrative.

Questions this posts answers

How are prepaid GPU or token credits treated on the balance sheet?: As a prepaid asset at purchase, unwound to expense or cost of revenue as the credits are consumed. Unused expiring credits are written off to expense.

Posts to

Primary sources

[S2] KPMG: Cloud computing implementation costs post ASU 2018-15 (US GAAP)

Ledger current as of 2026-07-24. A position and a citation, not accounting advice. See how we cite.