In the current gold rush of AI adoption, the conversation has been dominated by two themes: productivity gains and total cost of ownership (TCO). We talk about how many hours are saved by Cursor or Anthropic’s Claude Code or how many lines of code an agentic workflow can generate.
However, there is a massive, silent shift happening in the background that almost no one is talking about: The transformation of the software development balance sheet.
As AI begins to dramatically reduce the labor cost associated with software development, it is simultaneously introducing a new, significant cost input: token consumption. For forward-thinking CFOs and heads of accounting, the question is no longer just "How much are we spending on AI?" but rather "How do we account for it?"
Historically, software development has been a labor-intensive endeavor. Under standard accounting principles (such as FASB ASC 350-40), the labor hours spent during the application development stage are capitalized. This allows a company to move those costs to the balance sheet as a durable asset, depreciating them over the software's useful life and protecting their EBITDA in the process.
But as AI takes over more of the "heavy lifting," the ratio of human labor to machine consumption is flipping.
A company may reduce its developer headcount or hours but start to spend millions of dollars on high-reasoning tokens to build the same software. If the organization continues to treat those token costs as OpEx (operating expense), they are effectively punishing their own bottom line. They are turning what should be a capital investment into a "utility bill," leading to lower reported margins and a potentially lower share price.
From an accounting perspective, software has always been a bit of an anomaly. It is an intangible asset built almost entirely of "services." However, in the world of AI, tokens function much more like direct materials.
If you were building a physical factory, the steel and concrete used in construction would be capitalized. In the "digital factory" of 2026, tokens are the materials consumed to create code, test cases, and architectural documentation. If these tokens are consumed to build a software asset with a durable life, there is a powerful argument—and a GAAP-compliant pathway—to capitalize those expenses.
What are the accounting best practices for treating AI token consumption as a CapEx asset?
To successfully transition AI spending from a "black hole" of OpEx to a strategic CapEx asset, organizations need to implement three core pillars of governance:
You cannot capitalize what you cannot track. Organizations must move beyond "bucketed" AI billing. It is a best practice to use AI gateways to tag every token request with a project ID and a resource ID. This creates the "nexus of cost" required by auditors to prove that the expense was directly related to the creation of a specific asset.
Not all token spending is capitalizable. Tokens used for brainstorming or "preliminary project" research must still be expensed. A system like Clarity by Broadcom, the leading strategic portfolio management (SPM) solution for the enterprise, is vital in this regard. That’s because it provides the framework to ensure tokens are only capitalized when the project enters the "application development" phase.
We are entering an era in which the "timesheet" is no longer just for humans. Financial systems must now ingest token consumption files from providers like OpenAI, Anthropic, or Google, treating token units with the same rigor as labor hours.
For a large enterprise, the impact of this shift is not trivial. For a company spending $50M annually on tokens for software development, the ability to capitalize that expenditure can result in a significant boost to adjusted EBITDA and a more accurate representation of the company’s investment in intellectual property. (See a prior post to learn more about why AI success hinges on financial discipline.)
For teams in most organizations, the barrier to capitalizing AI tokens isn’t a lack of desire—it’s a lack of evidentiary proof. Auditors require a "nexus of cost," a clear, unbreakable link between a specific dollar spent and the creation of a durable asset. Clarity is indispensable in addressing these requirements.
How does the FASB ASC 350-40 standard apply to new AI token-driven software projects?
With the 2026 FASB ASU 2025-06 update, the rigid "stage-gate" model has been modernized. Capitalization now hinges on two criteria: Management authorization and probable completion. Clarity is uniquely positioned to help manage the compliance for both.
Clarity’s Financial Transaction Framework can integrate usage files from AI providers (such as OpenAI, Anthropic, and Google Cloud) or internal AI gateways. Rather than manual entries, Clarity’s extensible framework can map thousands of granular token requests to specific projects, tasks, and resources.
The audit trail: Every token-related expense can be associated with a Resource and a Project in Clarity. This creates a transparent ledger that shows exactly which AI token was used to build which software asset, and whether that work was related to new feature development or maintenance.
Not all AI use is created equal. A developer using an LLM to fix a legacy bug (maintenance/OpEx) is different from a developer using an LLM to architect a new microservice (development/CapEx).
Granular mapping: By using Clarity’s Charge Codes, organizations can bifurcate token spending based on the nature of the task. If the task is coded as "capitalizable," the associated token costs are automatically routed to the balance sheet.
The new 2026 accounting standards require proof that management has committed to the project.
The system of record: Clarity stores the formal project approvals, budget allocations, and probability of completion assessments in one place. When an auditor asks, "Why did you start capitalizing these tokens on March 15th?," Clarity provides the timestamped project status report and the budget approval that authorized the expense.
Perhaps most importantly for the CFO, Clarity provides real-time visibility into how AI spending is affecting the bottom line. By separating AI "utility" spending from AI "investment" spending, leaders can report accurate adjusted EBITDA margins that reflect the true value being created by their engineering teams.
The companies that win the AI era won't just be the ones that write code the fastest; they will be the ones that govern their AI spending most intelligently. By treating AI tokens as a capitalizable input rather than a sunk operating cost, CFOs can ensure that their digital transformation efforts are reflected where they matter most: in the company’s valuation.
It’s time to stop treating AI like the electric bill and start treating it like the investment it truly is.
To learn more, please contact the ValueOps team to continue the discussion.
The profitability hole occurs when companies reduce labor costs but treat the millions spent on high-reasoning tokens for software development as OpEx (operating expense) instead of capitalizing them. This hurts the company's bottom line by reducing reported margins and adjusted EBITDA.
Tokens function much like direct materials in the “digital factory,” as they are consumed to create the code, test cases, and documentation for a software asset with a durable life. This aligns with GAAP-compliant pathways like FASB ASC 350-40, which traditionally allow for the capitalization of labor hours during the application development stage.
Organizations must implement granular attribution (tagging tokens with project/resource IDs for an audit trail), phase-gate mapping (capitalizing spend only during the "application development" phase), and consumption-based timesheets (ingesting token files and treating token units like labor hours).
Under the 2026 FASB ASU 2025-06 update, capitalization has been modernized and hinges on two criteria: proof of management authorization and evidence of probable completion of the project.
Disclaimer: This article is intended for informational and educational purposes only and does not constitute formal accounting, tax, or legal advice. The accounting treatment of AI-related expenditures is subject to evolving regulatory interpretations and depends on the specific facts and circumstances of each organization. While the strategies discussed align with current GAAP and FASB frameworks, Broadcom recommends that companies consult with their own internal financial leadership and independent auditors to determine the appropriate capitalization and reporting policies for their specific business needs.