The New CFO Metric: From ROI to ROC
Why “Return on Compute” (ROC) Will Redefine Business Economics
For decades, business leaders optimized around a familiar equation:
ROI = Return on Investment
Capital went in. We either used it from profits generated or borrowed it from banks.
Revenue, and ideally more profits, came out.
It appears, however, that AI is changing our underlying fundamental unit of economic production.
We are entering a world where the primary input is no longer just capital, labor, or software licenses.
I’ve been listening very closely to Jensen (Nvidia), and Dario (Anthropic), and yes, Sam (OpenAI), and many others.
I’m coming to the conclusion that we have a higher order input, which is…
Compute.
Historically, companies scaled through combinations of:
Human labor
Financial capital
Physical infrastructure
Software automation
AI is introducing a new production engine:
Intelligence generated through compute.
Every AI-generated output now consumes:
GPU cycles
Tokens
Inference costs
Model training costs
Storage
Bandwidth
Energy
AI is not “free software.” AI is an ongoing compute consumption business.
That means the CFO equation is quickly changing from:
“What return did we get on our investment?”
to:
“What return did we get on the compute consumed?”
Welcome to the Era of ROC
Return on Compute
The future operating question becomes:
How much compute was required to:
Create the product?
Deliver the service?
Make the decision?
Generate the insight?
Replace or augment labor?
Accelerate execution?
And most importantly:
What economic value did that compute create?
In CFO terms, maybe the simplest analogy is that:
Compute Is Becoming the New Cost of Goods Sold
If you are an AI-native business or need to transition to one, then
Compute becomes production
Production becomes operating margin
Operating margin becomes compute efficiency
The companies that win will not necessarily be the companies with the most AI.
They will be the companies with:
The best compute economics
The best inference efficiency
The highest-value output per unit of compute consumed
Tomorrow’s CFOs and COOs will increasingly ask questions like:
Product
How much compute does this feature consume per user?
What is the marginal compute cost of personalization?
Which customers are compute profitable?
Operations
What workflows should humans still perform?
What workflows should agents perform?
What is the compute-to-labor tradeoff?
Finance
What is our compute gross margin?
What is our revenue per GPU hour?
What is our token efficiency ratio?
Which AI workflows generate positive ROC?
Strategy
Should we buy compute or build infrastructure?
When does owning infrastructure create advantage?
Is proprietary data improving ROC?
What is our defensibility per unit of compute?
We’re Moving From Labor Efficiency to Compute Efficiency
The Industrial Era optimized labor.
The Software Era optimized workflows.
The AI Era optimizes Intelligence
That means companies must learn to measure:
Cost per inference
Cost per decision
Cost per workflow completion
Cost per AI-generated output
Cost per autonomous task execution
The CFO Becomes the Architect of Compute Economics
This is where finance leaders become central again.
We’ve been there before, right? E-commerce spending, mobile spending, Cloud spending without economic discipline quickly became:
uncontrolled experimentation
exploding costs
growth at any cost
just hire people as fast as we can, and we’ll figure it out later
For those of us who like to learn lessons and turn them into a better future, let’s not do that again.
The future CFO must understand:
compute allocation
AI compute economics
model efficiencies
inference scaling
token consumption patterns
AI operating leverage
Tomorrow’s finance dashboard may include:
Revenue per million tokens
Compute efficiency
AI contribution margin
Autonomous workflow ROI
Human-to-agent leverage ratios
It’s Time To Start Thinking About a “Compute Operating System”
This way of thinking could very well be your next competitive moat.
Amazon mastered logistics…
Netflix mastered distribution…
Google mastered search economics…
The next generation of winners will master:
compute orchestration
AI workflow economics
intelligent routing
inference optimization
autonomous execution efficiency
If compute is becoming the investment, then….
ROC (Return on Compute) should become the operating lens.
The companies that learn to measure, optimize, and govern compute economics early will have a massive strategic advantage over those still treating AI as simply another software tool.
Ask yourself?
Is Compute becoming the new labor?
Is Compute becoming the new means of production?
Should Compute become the new operating system of business itself?
Or maybe I’m just “AI Pilled”"… but I ultimately love to challenge how we need to operate in the future, and how we use our past lessons to produce even better future results.
