ROC Framework · v0.1
Measure whether your AI compute is compounding.
Return on Compute is a strategic framework for evaluating whether AI infrastructure, API usage, model development, and internal AI tools are becoming a compounding asset — or just another infrastructure bill.
Short on time?
5Value buckets
10Calculation steps
ROC-v0.1Methodology
ComputeUseful intelligenceBusiness valueCapital
The framework
Five sources of value. One denominator.
Traditional ROI misses capability gains, productivity acceleration, efficiency improvements, and option value. ROC treats compute as a productive capital asset and measures every dollar of value it creates.
01
Production
Value from AI systems already deployed for customers and employees.
02
Capability
Risk-adjusted future value from new AI products, workflows, and model improvements.
03
Internal productivity
Captured value from employees becoming faster and more leveraged with AI.
04
Efficiency
Value from reducing cost per useful output — not just cost per token.
05
Option / reserve
Value of flexible capacity to serve upside demand without stranding cost.
Total ROC
ROC =
NPV(Vprod + Vcap + Vinternal + Veff + Voption)
NPV(Kcompute)
ROC compares the discounted economic value created by AI compute against the discounted cost of committed compute. Total ROC tells you whether the program is working. Marginal ROC tells you where the next dollar should go.
The calculator
From AI spend to board-ready ROC in under fifteen minutes.
Ten guided steps. Plain-English explanations. Visible assumptions. Every result includes a confidence range and a recommendation about where your next compute dollar should go.
- Organization profile, compute spend, and allocation
- Production, capability, internal, efficiency, and option value
- Effective compute diagnostic (utilization, fungibility, reliability, price-perf, model)
- Executive summary with marginal ROC by bucket
Sample output
ROC-v0.1
Total ROC
2.8x
Net ROC
+1.8x
Payback
1.4 yr
Efficiency4.6x
Capability4.0x
Internal productivity3.7x
Production2.5x
Option / reserve1.8x
Recommendation: Your next compute dollar belongs in efficiency. Marginal ROC is 4.6x.
Who it's for
For the people deciding where AI dollars actually go.
CFO / Finance
“Is our AI spend producing real economic value or just increasing infrastructure cost?”
CTO / Head of AI
“How should we justify and allocate compute across production, capability, and internal work?”
Transformation leader
“Which AI use cases are actually worth scaling — and which should be wound down?”
VC / Investor
“Does this company convert compute into compounding enterprise value, or just gross-margin drag?”