
Software capitalization under ASU 2025-06 is now a FY2026 decision
The FASB's rewrite of internal-use software accounting replaces stage gates with a single probability judgment, and early adoption puts reported margin, EBITDA and the asset base in play before the 2026 close.
The software capitalization ASU 2025-06 question has moved from the technical accounting backlog to the close calendar. The standard is not mandatory until annual periods beginning after Dec. 15, 2027, but early adoption is permitted as of the beginning of an annual period, which turns it into a now-or-wait election for calendar-year filers heading into the FY2026 10-K. The timing matters because the biggest thing running through internal-use software budgets right now is artificial intelligence development, and that is precisely the spend the new model treats most skeptically.
What the FASB actually changed in ASC 350-40
ASU 2025-06, Targeted Improvements to the Accounting for Internal-Use Software, removes the prescriptive project-stage framework that has governed ASC 350-40 for more than two decades. The preliminary project stage, application development stage and post-implementation stage no longer drive the expense-versus-capitalize line. In their place sits a single sequencing test: capitalization begins when management authorizes and commits funding to the project and it is probable the project will be completed and the software will be used as intended.
The second half of that test carries a condition practitioners are already circling. The standard adds a significant uncertainty consideration, meaning that where substantive development risk remains about the functionality the software is meant to deliver, the probable-to-complete threshold is not met and costs stay in the income statement. The FASB's stated aim was to fit modern development practice, where agile teams iterate rather than march through gates. The practical effect is to relocate the judgment from a project-plan artifact to a probability assessment.
Why AI builds are the hardest case under the new threshold
Experimental, model-heavy AI projects are the clearest collision point. A team training and retraining models, testing whether an approach produces usable accuracy, and pivoting architectures mid-build is the textbook description of substantive development risk. Under the old stage model, much of that work could be slotted into application development once a high-level design existed. Under ASU 2025-06, the question is blunter: at this date, is it probable this will be completed and used as intended?
That is not an engineering answer. Engineering teams are paid to be optimistic about what they are building, and funding approval is not the same as technical feasibility. The determination belongs to the controller's organization, supported by contemporaneous evidence - gate reviews, model performance benchmarks against a defined threshold, scope-change logs, and written authorization tied to a funded budget. Expect auditors to push hardest here. The Big Four technical desks have signaled that the significant uncertainty assessment on AI projects is where documentation will be tested first.
There is a second-order effect worth naming. If capitalization now hinges on a probability judgment rather than a stage, then the record of how a project was described internally becomes audit evidence. Optimistic board decks and cautious accounting memos on the same project are a reconciliation problem waiting to happen.
One standard, four metrics, zero change in cash
This is the rare pronouncement that moves reported operating margin, EBITDA, free cash flow and the asset base simultaneously without altering a dollar of cash outflow. Costs that get expensed depress operating income and sit in operating cash flow; costs that get capitalized lift the asset base, create future amortization and shift the outflow to investing. A company that capitalizes less under the new model reports lower EBITDA and higher operating cash usage while spending exactly the same money.
Transition compounds the problem. The ASU permits prospective and retrospective approaches, so two companies in the same sector, running comparable development programs, can present materially different capitalized software balances in the same reporting year purely as a function of adoption date and transition election. Analyst bridges will need to be built, not assumed. So will any covenant definition that references EBITDA, capitalized costs or tangible net worth without a frozen-GAAP provision.
The pre-close checklist for calendar-year filers
Four decisions need a position before the FY2026 books close rather than after. First, early adopt or wait: early adoption must be taken as of the beginning of an annual period, so a January election cannot be made in March. Second, the transition method, with a modeled view of what each does to the opening balance sheet and the comparative periods. Third, the capitalization trigger policy itself - what constitutes management authorization and committed funding in your governance structure, documented at a level a reviewer can test. Fourth, the significant uncertainty framework for AI and experimental work, including who signs off and on what evidence.
Then come the external conversations. Lenders need to see covenant impact before the fact. Equity analysts need a bridge that separates the accounting change from operating performance, ideally disclosed in the period of adoption rather than reconstructed under questioning. Internal audit and the audit committee need to understand that a judgment previously anchored in a project plan is now anchored in a probability estimate.
One staffing note sharpens the urgency. CFO turnover has been running hot through the fall, with new appointments announced across large caps in October. In a meaningful number of companies, this policy election and the disclosures around it will be made by a finance chief in their first close, working from a predecessor's capitalization practice. That is an argument for getting the policy documentation done early, not for deferring the decision.
Key takeaways
- ASU 2025-06 replaces the ASC 350-40 stage model with one test: management authorization and committed funding, plus probable completion and use as intended.
- Effective for annual periods beginning after Dec. 15, 2027, with early adoption permitted only as of the start of an annual period - making it a FY2026 election for calendar-year filers.
- The significant uncertainty condition keeps experimental AI development in expense, and that assessment sits with the controller, not engineering.
- Prospective and retrospective transition options mean sector peers can report materially different capitalized software balances in the same year.
- Lock the policy, transition method, covenant impact and analyst bridge before the FY2026 close, not during the audit.


