InferMargin
Menu
Research program for AI-native margins

Know if your AI infrastructure costs are healthy.

InferMargin helps AI-native startups compare their LLM API spend against verified peer cohorts — so founders can defend their margins, pricing, and infrastructure decisions with data instead of guesses.

No prompts or logs
n ≥ 30 public reports
Written-first research
When it matters

Four moments where “is this cost normal?” becomes a business decision.

Fundraise preparation

Investors ask whether your inference costs are sustainable. Walk into the conversation with cohort percentiles instead of a guess.

Model routing decisions

Considering a switch between providers or model tiers? Compare your current cost structure to peers before re-architecting.

Enterprise contract negotiation

When a buyer pushes back on price, knowing where your COGS sits in the cohort distribution turns a defensive answer into a data point.

Cost anomaly investigation

Your OpenAI bill jumped 40% last month. Is it your traffic, your prompt design, or the whole cohort moving together?

How it works

Three steps from connection to comparison.

Connect

Read-only access to your OpenAI or Anthropic Admin API and your Stripe Restricted Key. No prompts, no logs, no user content.

Compute

We calculate your LLM API COGS / product revenue ratio and place you in a peer cohort based on your use case and stage.

Compare

See your percentile (p25 / p50 / p75 / p90) vs verified peers, updated as the cohort grows.

Full methodology, security architecture, and data policy are detailed on the methodology page →

Confidentiality

Trust is the product constraint.

The benchmark only works if founders can contribute sensitive cost data without exposing company-level metrics. That posture is visible before anyone opens the questionnaire.

Read full data policy →

InferMargin never publishes company-level data. Public reports only include aggregated statistics from cohorts of n ≥ 30. Your private dashboard shows where you stand within a cohort, but no other company can see your raw metrics.
  • Read-only API access. Revocable at any time.
  • No prompts, no logs, no user content collected, ever.
  • Cohort minimums of n ≥ 30 (public) and n ≥ 10 (private).
Positioning

Not another cost dashboard.

InferMargin is not another cost dashboard. It is a confidential peer benchmark for AI-native unit economics — the missing layer between your observability tool, your finance stack, and your investors.

We are not a replacement for observability or revenue analytics. Helicone, Langfuse, CloudZero, Vantage, ChartMogul and Maxio remain necessary tools — InferMargin sits beside them and answers a different question: how do you compare?

Public research program

The first cohort is designed to answer three concrete questions.

InferMargin is not publishing empty thought leadership. The initial research program is scoped around decisions founders already face: investor margin questions, model routing, and safe peer benchmarking.

QUESTION 01 — PUBLISHED MAY 2026
The End of the 85% Illusion

What public disclosures from ICONIQ, Bessemer, and foundation model providers reveal about AI-native gross margins.

Read article →
QUESTION 02 — PLANNED
GPT-4o vs Claude 3.5 — unit economics

Early data on how model routing decisions affect cohort gross margins.

Planned
QUESTION 03 — PLANNED
Methodology paper

Peer benchmarking under n ≥ 10 anonymity constraints.

Planned

Aggregated findings will only be published once the relevant cohort thresholds are met.

Join the research

Help shape the first peer benchmark for LLM unit economics.

InferMargin is interviewing 20+ AI-native founders to build the first peer benchmark for LLM unit economics. ~15–20 minutes, fully written, no call required.

This is independent infrastructure-engineering research on LLM unit economics with AI-native founders. The current phase is written-first, not a sales process.