Thinking / Thought Leadership

Preparing for the transition.

Agents as customers will turn software CapEx into the core growth engine: more capitalized internal software, more revenue per employee, higher margins, and more equity-based labor participation.

01 / The Ruling

On September 18, 2025, the FASB issued ASU 2025-06, rewriting when internal-use software costs can be capitalized under ASC 350-40. The development-stage trigger is gone. Costs capitalize only when management has committed funding and the probable-to-complete threshold is met — and that threshold fails whenever significant development uncertainty exists: novel or unproven features not yet resolved through coding and testing, or performance requirements still unidentified or churning.

The practical consequences: genuinely novel AI work is expensed until late in the cycle; unit-of-account scoping is a documented judgment; and prior successful delivery lowers the bar for the next project. Effective 2028, early adoption available now. The quiet effect is that the P&L now distinguishes we don't know yet (expense) from we proved it (asset) — by construction.

02 / The Firm It Anticipates

Financially, the AI transition is a recomposition of the firm. A cost base of roughly half labor, a quarter purchased inputs, and a quarter capital is heading toward thirds — with tokens and model services as the new purchased input: bought-in cognition, managed like a bill of materials. Holding labor flat in dollars while its share falls to a third means growing the firm roughly 50%. Same people, 1.5x the output, the increment produced by the doubled capital and materials shares.

The firm becomes more capital-intensive, the cost of bad capex doubles, and ROIC — not headcount — becomes the metric leadership manages. The CFO, with the auditors as enforcement, now has standing to demand the evidence this requires: stable specifications, tested demonstrations, and spend tagged by what is proven versus what is hoped.

03 / The Agentic Customer

We are moving toward a market where agents become customers at massive scale. The number of agentic customers may exceed human customers by orders of magnitude, which changes the operating model of every software-enabled business.

To serve that future, companies will need to shift capital allocation toward software CapEx: internal platforms, automation systems, agent interfaces, orchestration layers, data infrastructure, and accounting practices that allow those investments to be capitalized properly.

The prize is substantial. Companies that do this well may see CapEx rise meaningfully, but that CapEx becomes productive internal-use software on the balance sheet. Revenue can scale two to three times without a comparable increase in headcount, and margins can expand dramatically because agents create operating leverage that human labor alone cannot.

04 / The Financing Model

The financing model also changes. As capital needs rise, companies may need to compensate labor less through near-term cash increases and more through equity. Equity becomes a way to align employees with the value created by the software CapEx they help build.

05 / The Next Operating Leverage Cycle

The AI transition creates a capital allocation mandate: rent abundant intelligence, but do not merely consume it. Convert it into owned software capital, governed data assets, reusable workflows, and agent-accessible revenue channels. Recent software-accounting changes make this more board-actionable: experimentation is expense, but production-grade internal software can be capital when it meets clear authorization, completion, use, and control thresholds. The strategic prize is higher revenue per employee, higher profit per employee, and more operating leverage owned by the firm rather than rented from vendors.

Preparing for the agentic customer transition? That's the readiness engagement.

Sources: Proverify.ai TokenOps internal research; FASB ASU 2025-06, Targeted Improvements to the Accounting for Internal-Use Software.

What Auditors Are Asking (2025–2026)

Questions appearing in due diligence and audit inquiries.

What percentage of total AI spend is classified as COGS, and how is the classification methodology documented?
For AI spend classified as R&D, what is the connection (if any) to deployed product features?
Can you produce a breakdown of paid-customer inference vs. internal vs. trial traffic?
What controls prevent reclassification of historical AI spend without documentation?
For any capitalized AI assets, what specific ASC criteria did you meet, and who reviewed the classification?

The single GL account problem. Companies with one GL account for "AI costs" cannot answer any of these cleanly.

Finance Practice Today

How finance teams currently see AI spend — and where the gaps are.

Allocate
Accounts, projects, tags, labels, cost centers, and hierarchies for showback and chargeback.
Monitor
Budgets, quotas, usage exports, model metrics, request logs, token counts, and billing SKUs.
Control
Budgets, alerts, limits, routing rules, identity controls, and anomaly reviews.
Gap
Allocation does not answer treatment, margin, pricing recovery, capitalization, or capital deployment.

CFO Expectations

What CFOs need from a TokenOps function.

Ledger
Provider, product, customer, workflow, project, owner, cost center, model, and treatment.
Margin
Token cost by product, feature, support motion, contract obligation, segment, and pricing tier.
Capital
Capitalization candidates with scope, approval, evidence, eval support, useful life, and exclusions.
Control
Budget variance, concentration risk, unmanaged usage, routing leakage, cache leakage, and exceptions.
Action
Fund, price, route, reserve, move, capitalize, expense, or stop.

Founding Partner Profiles

AI Systems & Capital Strategy

The AI systems founder.

Former Big Four M&A partner and technology founder focused on AI systems, corporate finance strategy, transformation, and capital deployment.

Corporate finance strategy
AI capital deployment
Diligence and transformation
Automation systems

Transaction Advisory & PE Governance

The advisory operator.

Transaction advisory innovation leader with private equity coverage experience and a track record building repeatable advisory methods.

Transaction advisory innovation
PE client governance
Commercial advisory platforms
Board-grade operating models

Technical Accounting & Capital Markets Readiness

The technical accounting partner.

CPA and advisory partner with public-company readiness, complex transactions, revenue, leases, debt and equity, and business combinations experience.

Technical accounting policy
IPO and SPAC readiness
US GAAP and IFRS
Revenue, leases, debt and equity

Work Plan

Six-step engagement model from source inventory to CFO actions.

01 / Source Inventory

Find the spend surface.

Inventory vendors, cloud accounts, API orgs, projects, cost centers, subscriptions, agents, coding tools, and billing owners.

02 / Data Connections

Ingest the right records.

Pull usage, cost, token, model, identity, tag, project, SKU, log, and export data from enterprise systems — not just invoices.

03 / Finance Ontology

Normalize activity.

Map usage to business unit, product, customer, workflow, environment, use case, owner, and economic purpose.

04 / Rules Engine

Classify treatment.

Assign OpEx, COGS, R&D, capitalization candidate, pricing signal, waste, routing opportunity, or risk exception.

05 / Evidence Pack

Support the judgment.

Attach project approvals, evals, asset scope, intended use, owner review, control exceptions, and exclusion rationale.

06 / CFO Actions

Fund, price, route, or stop.

Produce decisions for budgets, pricing, gross margin, vendor strategy, model routing, capitalization review, and spend controls.

Enterprise AI Token Spend Is Becoming a CFO-Controlled Cost Category

Public data on GenAI spend, model API budgets, token pricing, and the finance gap. TokenOps estimates marked accordingly.

Category	Fact	Source	Finance Relevance
Enterprise GenAI Spend	U.S. enterprise generative AI spend was estimated at $37B in 2025, up from $11.5B in 2024.	Menlo Ventures, State of Generative AI in the Enterprise	AI has moved from experimentation into a material enterprise spend category.
Model API / Token Spend	Menlo's 2025 enterprise GenAI estimate included roughly $12.5B of model API spend, the closest public proxy for direct token consumption.	Menlo Ventures, reported via secondary summaries and reconciliation analysis	Token consumption is already a multibillion-dollar vendor expense category.
Large-Enterprise LLM Budgets	Average enterprise LLM spend rose from about $4.5M to $7M over two years, with surveyed enterprises expecting roughly 65% growth to about $11.6M.	Andreessen Horowitz, Global 2000 CIO survey, 2026	Large enterprises are moving toward low-double-digit-million annual LLM budgets.
AI Investment Intensity	BCG's 2026 AI Radar found corporations expected AI investment to rise from about 0.8% to 1.7% of revenue.	BCG AI Radar 2026	Total AI budgets are much larger than token budgets, but token spend is a growing controllable input.
Token Pricing Basis	Frontier model vendors price usage per million input, cached-input, and output tokens. OpenAI, Anthropic, and Google publish token-metered API pricing.	OpenAI API pricing; Anthropic Claude pricing; Google Gemini API pricing	Tokens behave like a metered, fungible consumption unit that finance can allocate, monitor, and control.
Coding-Agent Cost Pressure	Anthropic states Claude Code enterprise deployments average about $150–$250 per developer per month, with wide variation by model, codebase, and usage pattern.	Anthropic Claude Code cost documentation	Agentic coding can create a large, department-level token cost pool.
Enterprise Cost Controls	OpenAI introduced enhanced ChatGPT Enterprise usage analytics and spend controls to help enterprises manage AI consumption and credit usage.	OpenAI; Reuters, June 2026	Vendors are adding controls because enterprise AI consumption has become financially material.
Spend Volatility	Recent research on agentic coding found token usage can vary significantly across tasks and runs, making cost forecasting difficult.	Bai et al., How Do AI Agents Spend Your Money?, 2026	Token spend requires finance-grade variance analysis, not just technical observability.
F1000 Annualized Token Spend	Fortune 1000 companies likely represent $10B–$25B of annualized direct model-token/API spend in 2026.	TokenOps estimate based on Menlo, a16z, vendor pricing, and coding-agent cost benchmarks	Direct token consumption is likely already a CFO-relevant spend pool across the F1000.
Core Finance Gap	Public data shows usage, pricing, and budget growth, but does not classify token usage by margin impact, capitalization potential, pricing recovery, or capital action.	TokenOps analysis	AI spend already has data; it lacks financial treatment.

Italic rows are TokenOps estimates. All other facts sourced as noted. Figures as of mid-2026.

Spend Sources

Where AI spend originates in the enterprise.

01 / Model APIs

OpenAI and Anthropic.

Project budgets, service accounts, usage APIs, cost reports, model, token, cache, request, workspace, and user-level data where available.

02 / Cloud AI Platforms

Azure, Bedrock, Vertex.

Azure OpenAI / Foundry, Amazon Bedrock, and Google Vertex AI billing, logs, tags, identities, subscriptions, projects, SKUs, and cost exports.

03 / Enterprise Data Clouds

Snowflake and Databricks.

Consumption records for AI functions, model serving, notebooks, pipelines, warehouses, and internal data products when token work runs inside the data estate.

04 / AI Work Tools

Copilot, Cursor, Claude Code.

Seat, usage, adoption, repository, issue, pull request, workflow, and output signals where engineering and product work consume AI.

TokenOps Finance

How finance teams see AI today.

What CFOs need from Token Ops.

Token spend is a CFO cost category.

Founding partner profiles.

Six-step engagement model.

Where AI spend originates.

Enterprise TokenOps Baseline.

Ledger plus action file.

Preparing for the transition.