# What is the journal entry for AI token usage?

Published 2026-10-05 · Brian Diamond

Track: finops

Segment: controller

**The journal entry for AI token usage is a debit to AI services expense, split by the cost center whose agents consumed the tokens, and a credit to an accrued liability for usage metered but not yet invoiced; when the invoice arrives, the accrual is reversed against accounts payable and any difference is booked as a true-up.** Tokens are a unit of consumption, like kilowatt-hours; the entry prices them and assigns them.

## Pricing the tokens

Providers bill per million tokens with separate rates for input, output, cached input, and sometimes per-request tool calls. The accrual uses your contracted rates applied to metered counts. If you buy through a cloud marketplace or have committed-spend discounts, use the effective rate, not the list price.

## The entries

**At period end (usage metered, no invoice)**

| Account | Debit | Credit |
| --- | --- | --- |
| AI services expense (by cost center) | X | |
| Accrued AI services | | X |

**On invoice receipt**

| Account | Debit | Credit |
| --- | --- | --- |
| Accrued AI services | X | |
| AI services expense (true-up, by cost center) | Δ | |
| Accounts payable | | X + Δ |

If Δ is negative, the true-up is a credit to expense.

**On payment**

| Account | Debit | Credit |
| --- | --- | --- |
| Accounts payable | X + Δ | |
| Cash | | X + Δ |

## Worked example

September usage for one provider: 38.0M input tokens at $3.00/M, 9.5M output tokens at $15.00/M, 12.0M cached input tokens at $0.30/M.

| Component | Tokens | Rate per M | Amount |
| --- | --- | --- | --- |
| Input | 38.0M | $3.00 | $114.00 |
| Output | 9.5M | $15.00 | $142.50 |
| Cached input | 12.0M | $0.30 | $3.60 |
| **Total** | | | **$260.10** |

Attribution from the gateway: Sales agent 55%, Support agent 45%.

**September 30**

| Account | Debit | Credit | Cost center |
| --- | --- | --- | --- |
| AI services expense | 143.06 | | Sales |
| AI services expense | 117.04 | | Support |
| Accrued AI services | | 260.10 | |

**October 6, invoice $261.40**

| Account | Debit | Credit | Cost center |
| --- | --- | --- | --- |
| Accrued AI services | 260.10 | | |
| AI services expense | 0.72 | | Sales |
| AI services expense | 0.58 | | Support |
| Accounts payable | | 261.40 | |

## Three choices to make once, in policy

- **Capitalize or expense.** Token usage is almost always a period expense. Tokens consumed to build an internal-use software asset may be capitalizable under your software development policy; decide the rule once with your auditor rather than per invoice.
- **Materiality threshold for accruals.** Many controllers skip the accrual below a threshold and book on invoice. Set the threshold in writing and apply it consistently.
- **Prepaid credits.** If you buy credits upfront, the purchase is a prepaid asset; usage draws it down to expense by cost center each period.

## Related

[How do I account for AI agent spend in the general ledger?](https://www.onaro.io/blog/how-to-account-for-ai-agent-spend-in-the-general-ledger) · [How to book AI subscriptions in the general ledger](https://www.onaro.io/blog/how-to-book-ai-subscriptions-general-ledger)

*Worked examples are illustrative. Rates shown are examples, not any provider's current pricing. Confirm treatment with your controller and auditor.*

Onaro Meridian is [FinOps for agentic AI](https://www.onaro.io/finops-for-agentic-ai): the system of record that attributes, controls and books what AI agents spend.

Canonical: https://www.onaro.io/blog/journal-entry-for-ai-token-usage