From a working fleet to a $7M/yr profit engine.
The build plan, the infrastructure, and the unit economics — what it costs to run AGIME/LIZ on our own metal, and exactly how many paying users take us to break-even, to profit, and to scale.
The roadmap
Four phases from today's proven pilot to a self-funding business.
Pilot
10-unit fleet live; register→unit onboarding; real payment→provision loop closed end-to-end (Stripe→gate→hub→LIZ unit).
Launch
Stripe live-mode, legal entity + terms, app-store builds, free-chat funnel keys, moderation. First paid cohort.
Grow
Ad campaign on the free-chat funnel → ~1,900 paying users (break-even) → ~2,700 (profit). Own-metal serving comes online.
Fleet & rail
2×8 B200 cluster, 10k+ units, AGIME Pay opened to third-party agents. ~$1.5M/mo revenue, ~41% net.
The path to profit
Unit economics (blended): ARPU $150/mo · gross margin 75% (contribution $112.50/user) · churn 4%/mo (~25-mo lifetime) · LTV ~$2,800 · CAC $400 → LTV:CAC ≈ 7:1. Fixed burn modeled at ~$180k/mo.
Contribution covers burn + churn-replacement. (Headline rev=burn ≈ 1,200 users.)
First durable profit; reinvest into growth.
≈ $7.4M/yr net. Acquisition capital to build the base ~$2.5–6M.
Highest-leverage sensitivities: churn (2% vs 6% roughly doubles/halves LTV & sustaining ad spend) and CAC (a working free→paid funnel keeps it ~$400 vs $600–1,200 paid-only). Pay-gateway 0.5%/txn revenue is excluded — pure upside.
Infrastructure — own metal
The heavy compute: a cluster of 2 × 8× NVIDIA B200 (16 GPUs) for model serving/fine-tuning. At 24/7 utilization, owning beats renting after ~12–18 months.
| Path | One-time capex | Monthly (all-in) | Notes |
|---|---|---|---|
| Buy (2×8 B200 + IB) | $0.86–1.18M | ~$29–40k | colo/power ~$5–7.5k + amortized HW ~$24–33k (36-mo) |
| Rent — reserved 1–3yr | $0 | ~$35–53k | ~$3.0–4.5/GPU-hr × 16 × 24/7 |
| Rent — on-demand | $0 | ~$58–84k | fastest to start; best for bursty use |
Core production stack (separate from the heavy cluster)
| Service | Monthly | Resale gross margin |
|---|---|---|
| pay.agime.ai gateway + HA Postgres | $85–560 | 85–96% at $1M+ GMV (0.5%/txn) |
| law.agime.ai | $10–60 | — |
| Voice: F5-TTS + faster-whisper STT (GPU) | $205–460 | 50–70% (resold as managed voice) |
| Core 14B model serving (pooled vLLM, 48GB) | $300–920 | 50–70% (resold bot hosting) |
| Core production total | ~$650–2,450 | reselling gateway/voice/bots = the recurring margin |
Blackwell supply is constrained (hyperscaler priority / waitlists) — start on reserved-rent, migrate to owned metal as utilization justifies. Own infra + resale (Pay · voice · bot hosting) is the moat: we sell the same rails we run on.
Model subscriptions
LIZ answers most turns on-device (dense-14B) and the dispatcher calls external frontier models (OpenAI · Anthropic · Google · DeepSeek · Groq · Mistral) only for hard tasks — so cloud API is a small COGS line, not a margin threat.
blended; <10% of the $275 LIZ price even at the high end
early pilot scale
reference business
after committed-use / batch / cache discounts
Cost driver = escalation rate + premium-model share (reasoning/thinking tokens dominate). Tight routing to cheap-fast tiers for medium tasks keeps per-unit cost near the low end.
Team & salaries
US 2025–26 averages, fully loaded (base × ~1.25–1.4 for taxes/benefits/equity/overhead).
~$0.84–1.2M/yr — mostly engineering: ML/infra, backend, frontend/mobile, DevOps, ½ product/growth, support
~$2.78–4.13M/yr — adds sales, more eng, ops, legal/compliance, marketing
| Role (per head, fully loaded) | Yearly |
|---|---|
| ML / Infra Engineer | $190–265k |
| Backend Engineer | $175–260k |
| DevOps / SRE | $170–250k |
| Product Manager | $175–250k |
| Frontend / Mobile Engineer | $150–225k |
| Account Executive (incl. OTE) | $155–230k |
| Growth / Marketing Manager | $125–195k |
| Customer Support Specialist | $63–90k |
Monthly burn & capital
Where the money goes at each stage — and the capital to get there.
| Line | Lean launch | Scaling (toward profit) |
|---|---|---|
| Team | ~$85k | ~$290k |
| Marketing (team + stack + ad) | ~$25k | ~$80k |
| Core production infra | ~$2k | ~$25k |
| Model APIs (dispatcher) | ~$2k | ~$15k |
| Compute cluster (B200, reserved→owned) | — (defer) | ~$35k |
| Total burn / mo | ~$115k | ~$445k |
runway to ~1,900 paying users incl. acquisition capital ($0.3–0.7M)
one-time; or defer via reserved-rent
cumulative acquisition to the max-profit base
Lean launch burn (~$115k/mo, B200 deferred) reaches break-even at ~1,900 paying users; the scaling burn is covered well before 10k. Own-metal + resale compounds margin as volume grows.
Build the hard part. Fund the scale.
The pilot is proven and the loop is closed. The plan above is what turns it into a $7M/yr engine.