aicost.ai
VC/PE Diligence Memo
2026-07-22
Pricing as of 2026-07-20
aicost.burn-multiple-diligence
Burn Multiple Stress-Test
Verdict
EFFICIENT
Burn multiple 1.2x is under 1.5x — efficient growth by 2026 Series A standards.
Key figures
| Burn multiple |
1.2x |
| Reported gross margin |
62% |
| True gross margin |
62% |
| Margin erased |
0 pts |
| Inference disguised as growth |
$0 |
| Share of burn that is subsidy |
0% |
| ARR needed for an efficient 1.5x |
$2,800,000 |
| Inference basis |
reported |
Assessment
Reported burn multiple is 1.2x (efficient) on $4,200,000 net burn against $3,500,000 net new ARR. Inference appears booked in COGS; gross margin holds at 62%. True margin is above the ~52% 2026 AI median. Ask for the GL classification of inference spend and the NRR of the cohort driving the ARR.
Questions for the founder
- Show the GL: how much inference/compute cost is booked in COGS vs sales & marketing vs R&D?
- What is net revenue retention on the cohort generating this net new ARR? (120%+ is the 2026 floor.)
- Is any usage sold at or below cost to drive logos — and what share of ARR is on such terms?
- What is the gross-margin trend over the last 4 quarters, net of all inference cost?
- What is the burn multiple trend quarter-over-quarter — is capital efficiency improving or decaying?
- Are there customer contracts with unlimited usage for fixed fees or price guarantees that cap expansion revenue?
Assumptions & method
- Inference uses the reported annual figure ($700,000); supply a workload to price it vendor-exact. *
- 0% of inference assumed mis-booked as growth/S&M — confirm against the GL. *
- Burn multiple = net burn ÷ net new ARR over the same period; thresholds per 2026 Series A benchmarks.
What moves the answer
- inferenceBookedAsGrowthPct: The distortion knob — the higher this is, the more the reported burn multiple and gross margin flatter reality.
- netNewArrUsd: Denominator of the burn multiple; verify it is NET new ARR (after churn), not gross bookings.