Standards-anchored to primary IFRS and FASB text. Worked figures are illustrative. Not accounting advice.
TCTokenCapitalisation
Amendment, US GAAP

ASU 2018-15 cloud computing implementation costs

Amendment
ASU 2018-15 addresses a customer's accounting for implementation costs incurred in a hosting arrangement that is a service contract, aligning their treatment with the internal-use software guidance in ASC 350-40 ASU 2018-15. It matters for AI consumed as a hosted service, where the model runs on a provider's infrastructure.

What it covers

When a customer buys software as a hosted service rather than acquiring a licence, the service fee is an operating expense. ASU 2018-15 addresses the separate implementation costs the customer incurs to configure and deploy that hosted service, and aligns their capitalisation with the internal-use software rules in ASC 350-40 ASU 2018-15.

How it interacts with ASC 350-40

Before the update, implementation costs in a service-contract hosting arrangement lacked clear guidance. ASU 2018-15 closed the gap by pointing to ASC 350-40 for which implementation activities are capitalised and which are expensed, so a customer applies a familiar framework ASC 350-40. The hosting fee itself remains a period cost.

Why it matters for token spend

Much AI is consumed exactly this way: a hosted model accessed through an API, billed by tokens. The metered token fee for running the service is the hosting cost and is expensed. But the implementation work to build a specific asset on top of that hosted model, and the development-phase tokens consumed doing so, can fall within the capitalisation framework this update points to. The distinction between the service fee and the implementation build is the same measure-phase discipline the rest of the site applies.

The practical takeaway

For a US GAAP reporter using hosted AI, separate three things: the recurring hosted-service token fee, expensed; the run-phase inference, expensed; and the development-phase implementation build, potentially capitalised. ASU 2018-15 is why the third is analysed under the internal-use software framework rather than left without guidance.