Post your token spend to the right ledger line
Describe the instrument you just signed. The LedgerPost writes the double-entry posting, stamps the verdict, and cites the paragraph that governs it, under US GAAP and IFRS.
Book the commitment as prepaid at signing, then reclassify the drawn-down capacity used in the application-development stage of an internal-use build to the software asset. The rest hits expense as consumed.
Token CapEx is treating LLM token, compute and inference spend as capital expenditure that can sit on the balance sheet, capitalized and amortized over its useful life, rather than defaulting to operating expense. Under US GAAP, ASC 350-40 permits it for internal-use software in the application-development stage, a stage model ASU 2025-06 replaces from 2028 for calendar-year companies; production inference is expensed or booked to cost of revenue. IFRS routes it through IAS 38.
Three ways in, one ledger
Enter by instrument
You signed a specific spend item. Start from the instrument and get its default posting.
All spend instruments→the CFO entry pointEnter by treatment
You care what a spend item does to operating margin. Start from the ledger line.
All treatments→the auditor entry pointEnter by standard
You need the paragraph. Start from the source-of-record index of the six standards.
Source-of-record index→The instrument index
Reserved GPU commitment
Multi-year committed-use GPU capacity
Prepaid inference credits
A prepaid block of inference or token credits
On-demand inference
Metered pay-as-you-go inference
One-off training run
A large training run provisioned as a project
Fine-tuning run
Adapting a base model to an owned artifact
Committed-use discount
CUDs, savings plans and reserved instances
Rented model access
Third-party LLM API subscriptions
Eval-suite build
Building an evaluation and test harness
Prompt registry
A managed store of prompts and prompt artifacts
Integration code
Glue and integration code around a model
Vector DB and retrieval
RAG infrastructure, embeddings and vector storage
MLOps pipeline
MLOps tooling and orchestration
Mechanics we own
Worked scenarios
- $2M reserved GPU commitment
- Prepaid inference credit block
- Internal copilot build
- Production inference to COGS
Every figure in a scenario is an illustrative example, not client data. See how we cite.
Questions a FinOps lead types
- What is Token CapEx?
- Token CapEx is treating LLM token, compute and inference spend as capital expenditure that can sit on the balance sheet, capitalized and amortized over its useful life or held as prepaid compute that unwinds over the commitment, rather than defaulting to operating expense. Development-window build cost can capitalize under ASC 350-40 and IAS 38; production inference is expensed or booked to cost of revenue.
- Is a reserved compute commitment CapEx or OpEx?
- Neither at signing. A signed commitment is a prepaid asset, not consumption. It unwinds to expense, cost of revenue or a capitalized asset as capacity is drawn down, under ASU 2018-15.
- Can we capitalize LLM training costs?
- Yes, when the run develops buyer-owned internal-use software, funding is committed and completion is probable; after adopting ASU 2025-06, no significant development uncertainty may remain. Exploratory or novel runs are expensed as research.
- How does US GAAP differ from IFRS on the same token spend?
- US GAAP uses the ASC 350-40 internal-use model, amended by ASU 2025-06 from annual periods beginning after 15 December 2027; IFRS uses the IAS 38 research-versus-development split. Cloud configuration diverges most: ASU 2018-15 permits capitalizing implementation, while the IFRIC March 2021 decision generally expenses it.