On June 18, OpenAI gave ChatGPT Enterprise admins new credit usage analytics and spend controls. It’s a single view of credit consumption broken down by user, product, and model, default workspace budgets, per-group limits, and a Cost API for pulling the data into their own systems. Two days earlier, Microsoft shipped Copilot Cowork with spending limits, budget allocation, usage alerts, and user-level caps.
This is a step in the right direction. But it’s also a measure of how big the problem has become.
The spend is real but the return is elusive
Global AI spending is forecast to reach $2.59 trillion in 2026, a 47% increase over 2025 (Gartner). The ROI isn’t as clear. Fifty-six percent of CEOs report no measurable revenue gain or cost reduction from AI in the past twelve months, and an MIT NANDA study of 300 deployments found that 95% of enterprise generative AI pilots delivered no measurable return. Only about 29% of executives say they can confidently measure AI ROI (IBM). Forrester expects enterprises to defer 25% of planned AI spend to 2027 because finance teams cannot connect what they spent to what it produced.
The pattern is consistent. We see that spend continues to compound, but the accounting for it hasn’t caught up. This is less a technology failure and more a financial visibility failure. We see the result of applying legacy budgeting and accountability models to a technology whose economics are not like a traditional software subscription.
Report
Finance needs to prove AI’s return: CloudZero report
260 senior finance leaders (more than half CFOs) told us why the speed of seeing AI spend, not the size of it, separates who pulls ahead on AI from who gets burned.
What the vendors are doing, and where it stops
Vendor spend controls help. But they answer a narrow question: how many credits did your people burn inside our product this month? It may be useful for governing one tool, but it’s not the answer a CFO owes the board. Three gaps remain.
They are per-vendor and walled off. OpenAI shows you OpenAI. Microsoft shows you Microsoft. No vendor normalizes cost across its competitors’ pricing, and most enterprises now run several models from several providers plus their own.
Bottom line: they measure activity, not outcomes. Credits, tokens, and seats are the input meter. They do not tell you cost per customer, cost per product, cost per transaction, or cost per strategic bet. The meter counts inputs; the board asks about outcomes..
They cap spend but they don’t explain it. A budget limit might keep you from overspending, but it fails to separate the spend that is producing margin from wasteful spending. Rather than spending less on AI, a CFO would rather spend well and prove it.
The layer CloudZero was built for
CloudZero is the AI ROI company. We are the financial control plane for AI, the system that connects every AI dollar to the business outcome it produced, across every provider, in real time.
What that means in practice is AI outcome attribution. We map spend to the things you are accountable for (the customer served, the product shipped, the transaction processed, the strategic bet) rather than to the tokens and models that vendor dashboards count. We express it as cost-per-anything: cost per customer, cost per feature, cost per company OKR. The vendor consoles tell you what your AI did, but CloudZero tells you what it produced, and what it cost to produce.
We have been doing this class of work since 2016, processing 14 trillion cloud billing events a year for customers using AWS, Azure, GCP, and other infrastructure providers. The AI vendor dashboards launching this month are catching up to a problem we have spent a decade solving, and they can only see their own corner of it.
Three steps every CFO should take now
First, get a view across every provider, not one. You need a single, normalized picture of AI spend spanning OpenAI, Anthropic, Microsoft, your cloud, and your in-house models, because a per-vendor dashboard is a partial picture and you are accountable for the whole. This is exactly what CloudZero does: we unify spend across every provider into one view that no single vendor console can give you.
Second, attribute spend to the outcomes you are measured on. Pick the units that matter for your business (cost per customer, per product, per transaction, per key result) and connect spend directly to them. Activity metrics tell you what was used; outcome attribution tells you what it was worth. CloudZero is built to do this, joining every AI dollar to the outcome it produced so you can defend it.
Third, move from the monthly bill to a real-time signal. Do not wait for an invoice to discover a surprise. CloudZero captures spend as the work happens, so you can catch anomalies and reallocate toward what is working before the quarter closes, not after it.
The vendors watching the meter is progress. But the CFOs who separate from the field over the next two years will not be the ones who spend more on AI or less. They will be the ones who can prove what every AI dollar produced. That is the difference between a credit report and a financial control plane, and between capping AI spend and understanding AI ROI.
