Concept 1: AI Spend Has Become a Major Operating Expense
What's happening?
AI is no longer just a small experiment. It has grown into a large, recurring cost for businesses.
Key fact:
Global AI spend reached $1.5 trillion in 2025
Why it matters to CFOs:
When something shifts from a small pilot budget to a major operating line item, finance leaders must treat it like any other significant expense — it needs to be measured, justified, and optimized.
The core problem:
- 88% of organizations have deployed AI in at least one function
- But only 39% can connect that investment to actual profit impact (EBIT)
💡 Simple analogy: Imagine spending heavily on a new factory floor but not being able to tell if it's producing more goods. That's where most companies are with AI right now.
Concept 2: AI Usage Correlates With Revenue Growth (But Unevenly)
The finding:
A BCG analysis found a clear pattern:
| Token Usage Level | Revenue Growth (YoY) |
|---|---|
| Highest quintile (top 20%) | 16.5% |
| Lowest quintile (bottom 20%) | 5.1% |
What this tells us:
Companies that use AI more intensively tend to grow faster. Intelligence is showing up in real business outcomes.
The Jevons Paradox connection:
When AI models get better, people use them more — not less.
- After major model improvements in late 2025, workers sent 44% more agent messages per week
- For high-complexity work, messages rose 68%
💡 Key insight: Better AI doesn't reduce demand — it expands what teams are willing to attempt. This is called a Jevons-style dynamic.
Concept 3: The Returns on AI Are Highly Concentrated
The finding:
Not everyone benefits equally. The data shows extreme inequality in who gets value from AI.
Striking statistics:
- p99 developers (top 1%) produced 46x more AI-assisted lines per day than the median user
- They merged 15x more pull requests per week than the median
How unequal is this?
Measured by Gini coefficient (a standard inequality measure), the distribution of AI benefits is more unequal than income distribution in any country in the world
What this means practically:
- A small number of people are getting enormous leverage from AI
- Most users are not getting meaningful value
- Simply buying AI tools for everyone does not guarantee returns
💡 Takeaway for CFOs: You can't assume AI spend is working just because licenses are deployed. You need to look at who is actually using it effectively.
Concept 4: Cost Per Unit of Work Varies Widely
The problem:
Even when AI is working, how much you pay per unit of output varies dramatically.
Key data:
- Cost per agent request varied by nearly 9x across different model families
- Cost per accepted line of code varied by roughly 7x
Why does this happen?
Different AI models are built for different tasks:
| Task Type | Best Fit |
|---|---|
| Planning | Larger, more capable models |
| Frontend development | Mid-tier models |
| Debugging | Specialized models |
| Simple/repetitive execution | Lower-cost models |
The solution — model optionality:
84% of power users already use multiple models each week
Using the right model for the right task = significant cost savings.
The pricing shift making this more urgent:
AI providers are moving toward usage-based pricing, which means:
- Costs become variable (harder to predict)
- Routing work to the wrong model wastes money at scale
💡 Analogy: It's like shipping — you wouldn't use overnight express for every package. Matching the shipping method to the urgency saves money. Same logic applies to AI models.
Concept 5: A New Discipline Is Needed — AI Economics
The gap:
There is currently no established framework for making AI investments:
- Measurable
- Predictable
- Efficient
What needs to be built:
The article identifies three key areas finance leaders must develop:
- Shared benchmarks for AI productivity
- Frameworks for measuring return on intelligence (ROI)
- Practical approaches to model allocation and cost management
The CFO Council response:
Cursor is launching a CFO Council — a working group of finance leaders meeting quarterly to answer:
"How do you keep AI spend tied to value?"
Summary: The 5 Core Concepts at a Glance
| # | Concept | Core Idea |
|---|---|---|
| 1 | AI as a major expense | AI has moved from experiment to major operating cost — but most can't measure its impact |
| 2 | AI usage drives revenue | Higher AI usage correlates with 3x higher revenue growth |
| 3 | Returns are concentrated | A tiny minority captures most of the value — deployment ≠ adoption |
| 4 | Cost varies by model | 7–9x cost differences exist; matching work to model is critical |
| 5 | A new discipline is needed | AI economics requires new benchmarks, frameworks, and cost management approaches |
🎯 The big picture: AI is becoming a core financial variable — like headcount or infrastructure. CFOs who build the discipline to measure, allocate, and optimize it will have a significant competitive advantage over those who treat it as a flat subscription cost.