Most AI products treat unit economics as a trade secret. We don't. The formula below is exact, the variables are real, and the numbers are pulled from our actual production cluster. You can plug your own numbers in and get the same kind of answer.
The point is not to look smart. The point is that an investor, a customer, or a tenant can verify whether what we say on /why-ownify and /roadmap actually pencils out — using arithmetic, not trust.
R = R_sub + R_topup
R_sub — the tier subscription (Solo €12, Duo €22, Team €32, Pro €99, Business €279)R_topup — wallet credits consumed at the customer-facing rate (Fireworks list × 1.47)Note: the 1.47 markup applies to what customers pay. Our actual cost to Fireworks is the raw list price — see C_inf below.
C = C_inf + C_fixed + C_store + C_payment C_fixed = OVH_cluster_bill / N_active_agents
C_inf — actual Fireworks invoice per agent: Σ_model (calls × avg_prompt × in_rate + calls × avg_out × out_rate)C_fixed — the real OVH cluster bill (currently €360/month, covers managed Kubernetes + nodes + load balancer + storage volumes + bandwidth) divided by the number of active agentsC_store — marginal storage cost above what's already in the cluster bill (Solo ships 5 GB of agent-specific PVC; ~€0.05/GB-month)C_payment — payment fees (Stripe + Stripe Tax) ≈ R × 2.9%C_CAC = R × 5%
We commit to a 5% of revenue CAC ceiling at steady state. That's a deliberately lean, organic-leaning growth budget — content, community, conferences, modest affiliate. It funds growth that earns its keep, not paid acquisition that races for vanity metrics. Early-stage acquisition (the first 100-300 customers) will run higher than 5% and is funded from runway, not monthly cashflow. The 5% target is the steady-state assumption; we name it as a constraint to keep future-us honest.
Burn = founder + R&D_tools + ops_misc
= €6,000 + €500 + €200
= €6,700/monthFounder comp is one full-time equivalent at modest founder pay. R&D / tools is observability, dev services, accounting tools, etc. Ops is legal + miscellaneous. We publish this because transparent burn matters more than transparent revenue — it tells you what gross margin has to clear before the platform is self-sustaining.
GM = R - C Net = (GM × N_paid) - C_CAC - Burn
Operating break-even is the smallest N_paid at whichNet ≥ 0. The break-even ladder below shows it explicitly across realistic scale points.
Solo tier (€12/mo) on the average production-traffic shape we see for our most active dogfood tenant: 27,670 prompt tokens and 976 completion tokens per call, 500 calls per month. Two scenarios shown side by side: today (N=2 active agents) — honest development-stage reality where fixed cluster cost is split across almost nobody, and at designed capacity (N=100) — where the current cluster sizing converges. The cluster bill of €360/month is real (current OVH invoice).
| Variable | Today N=2, Kimi K2.6 (pre-fix) | Today N=2, gpt-oss-120b (post-fix) | At capacity N=100, gpt-oss-120b | v3 forecast local model |
|---|---|---|---|---|
| Our cost — input (€/1M, Fireworks list) | €0.95 | €0.15 | €0.15 | — |
| Our cost — output (€/1M, Fireworks list) | €4.00 | €0.60 | €0.60 | — |
| Customer rate (€/1M, list × 1.47) | €5.88 | €0.88 | €0.88 | — |
| C_inf (inference, monthly) | €15.10 | €2.37 | €2.37 | €0.00 |
| C_fixed (€360 cluster ÷ N) | €180.00 | €180.00 | €3.60 | €3.60 |
| C_store | €0.25 | €0.25 | €0.25 | €0.50 |
| C_payment | €0.35 | €0.35 | €0.35 | €0.35 |
| C total | €195.69 | €182.97 | €6.57 | €4.45 |
| R (Solo) | €12.00 | €12.00 | €12.00 | €12.00 |
| Gross margin (per agent) | €-183.69 | €-170.97 | €5.43 | €7.55 |
| GM% | -1531% | -1425% | 45% | 63% |
This view assumes the cluster is at its designed capacity (~100 active Solo tenants). At today's 2-tenant dogfood scale, the cluster-share line is much larger and we're below break-even — that's development-stage reality, and the point of publishing this table is to show where the model converges, not to claim today's P&L is healthy. Gross margin (when we reach it) funds: salaries, the v2/v3/v4 architectural arc, support + on-call, runway buffer, and eventually profit.
Revenue assumed at the weighted-average ARPU of €30 across the Solo → Business tier mix. Trial dilution: 1 active trialer for every 3 paid agents (typical pre-conversion funnel). Cluster cost grows with N (rough linear model above the €360 baseline). 5% CAC included.
| N (paid) | + Trialers | Revenue | COGS | Gross margin | CAC (5%) | Burn | Net / mo |
|---|---|---|---|---|---|---|---|
| 100 | 33 | €3000 | €982 | €2018 | €150 | €6700 | €-4832 |
| 200 | 66 | €6000 | €1609 | €4391 | €300 | €6700 | €-2609 |
| 400← break-even | 132 | €12000 | €2861 | €9139 | €600 | €6700 | +€1839 |
| 800 | 264 | €24000 | €5367 | €18633 | €1200 | €6700 | +€10733 |
| 1500 | 495 | €45000 | €9751 | €35249 | €2250 | €6700 | +€26299 |
A consistent worry is: “if your roadmap moves model + memory to the user's device, doesn't that kill your SaaS?” Short answer: no — the unit being sold changes. Long answer in three phases.
The mistake is conflating platform with compute. Email is the precedent: every mailbox runs locally (or in iCloud / Gmail), but SMTP, DNS, deliverability, anti-abuse, and identity are paid platform layers. We're positioning to be the platform layer for personal AI agents — independent of where the inference actually runs.
Every number on this page is pulled from our own production cluster. The founder runs multiple ownify agents and depends on them daily. When something is broken, we feel it before any customer does — and you can see the receipts.
Concrete recent example: the routing fix shipped on 2026-04-28 dropped per-agent inference cost on the bulk path by ~6×. We discovered the bug because we run the system, looked at the spend, and noticed the math didn't match the page. The sequence (diagnosis → fix → verification → cost-shape change) is the kind of thing customers and investors should be able to see in real time.
A live transparency page with current cluster numbers and the founder's own agent stats is on the way. When it lands it will link from here.