• Post author:
  • Post category:AI World
  • Post last modified:March 30, 2026
  • Reading time:4 mins read

AI’s ARR illusion: decoding GMV, run-rate, and real revenue math

What Changed and Why It Matters

AI startups are reporting “ARR” that often isn’t recurring, contracted, or even revenue. In the rush to show traction, teams are annualizing volatile usage, counting marketplace GMV as topline, and bundling one‑off services into “software.”

“ARR is no longer a measure of durable revenue, it’s a storytelling device.” — Career Candour

The signal is everywhere: investors and operators are calling out GMV‑as‑ARR, “eventized” run rates, and creative metrics. Breakingviews even flagged how AI vendor revenues can swing fast with enterprise behavior and concentration.

Here’s the part most people miss. This isn’t just accounting. It’s a structural clash between usage‑based AI economics and the SaaS mental model.

The Actual Move

Across the ecosystem, several patterns are inflating the idea of “ARR”:

  • Marketplaces and staffing platforms counting gross payments (GMV) as “ARR,” instead of their net take rate or platform revenue.
  • API and model providers annualizing a strong month or quarter of usage into a “run rate,” then branding it as ARR.
  • Resellers and integrators including low‑margin pass‑through compute or services in a single “ARR” figure.
  • Enterprise minimum commitments or credits getting treated like guaranteed recurring revenue, despite uncertain consumption.
  • One‑time implementation, data labeling, or custom model work folded into “software ARR.”

“AI startups are boasting about ARR numbers that are actually low‑margin GMV. This is not just accounting; it is a structural issue.” — Oswarld

“That deserves a very different multiple than a SaaS company with $500M of true recurring revenue.” — Mostly Metrics

Recent flashpoints made this visible:

  • A viral debate questioned whether a fast‑growing AI marketplace’s “$500M ARR” was actually GMV rather than net revenue (CJ Gustafson, LinkedIn).
  • Breakingviews highlighted how AI vendor revenue narratives can jump quickly on enterprise dynamics and concentration.

“Big businesses account for 80% of [an AI vendor’s] revenue…” — Reuters Breakingviews

  • Operators proposed a new filter—ERR vs. ARR—to spot “eventized run rate” dressed up as durable revenue.

“Signals that suggest revenue might be ERR rather than ARR” — AIPM Guru (Substack)

  • Fortune chronicled founders using “creative accounting” to inflate ARR as AI blurs lines between software, services, and infra.

“That’s really not revenue.” — Fortune

  • Practitioners reiterated first principles: true ARR is predictable, contracted, and excludes one‑time or project work (Medium).

The Why Behind the Move

AI is breaking the SaaS scoreboard. Usage is spiky, infra is expensive, and marketplaces run on GMV math. Pressure to show momentum pushes teams to map new economics onto old metrics.

• Model

  • LLM/API usage is variable. Annualizing a hot month looks impressive but isn’t recurring by default.

  • Marketplaces earn take‑rate on GMV; gross volume ≠ company revenue.

• Traction

  • Demos, POCs, and credit‑fueled trials create short bursts that don’t persist.

  • Concentrated enterprise adoption amplifies swings.

• Valuation / Funding

  • The market still prices “ARR.” Blurry definitions unlock better multiples—until diligence lands.

  • As Mostly Metrics notes, GMV‑like businesses deserve different comps than SaaS.

• Distribution

  • Fast land‑and‑expand can tempt teams to count commitments and pipelines as ARR.

  • Marketplaces highlight liquidity (GMV) to attract both sides, masking thin take rates.

• Partnerships & Ecosystem Fit

  • Resale and integrations blur net vs. gross. Without clear gross margin disclosure, capacity resale can look like software.

• Timing

  • Tight capital markets reward narrative clarity. That’s pushing some founders to over‑simplify or over‑state.

• Competitive Dynamics

  • Everyone claims “platform” status. Inflated ARR becomes table stakes in board decks, press, and recruiting.

• Strategic Risks

  • Mismatched metrics break trust with customers, investors, and candidates.

  • Over‑promising locks teams into growth-at-all-costs decisions that crush unit economics later.

What Builders Should Notice

  • Separate the stacks. Report GMV, net revenue, and gross margin distinctly. Never bundle pass‑through with ARR.
  • Define ARR narrowly. Contracted, renewable revenue only. Exclude services, one‑offs, credits, and pilots.
  • Kill run‑rate theater. Share trailing 3–6 month averages and logo‑level retention to prove durability.
  • Publish a revenue ladder. Bookings → Billings → Recognized revenue → Net revenue → Gross margin. Make leakage explicit.
  • Price to value, not to optics. If your model is usage‑based, lean into usage KPIs plus cohort retention—not faux SaaS labels.

What most people miss: ARR is a contract, not a vibe.

Buildloop reflection

“Trust compounds faster than growth—especially in AI.”

Sources

Tags: , , , , , , ,