This model tracks actual API consumption, infrastructure, and voice costs from production usage and projects them forward at 10x, 100x, and 1,000x user scale. Margins improve dramatically with volume — model tiering, caching, and shared infrastructure compress per-user cost from ~$115/month at 1 user down to $8-20/month at 1,000.
Cost per user at each scale tier
| Line Item | Basis | Cost / Month |
|---|---|---|
| Claude API — Opus (terminal) | ~500K tokens/day | $50–80 |
| Claude API — Haiku (OC/hub) | ~50K tokens/day | $5–10 |
| Claude API — Agents & sprints | Variable | $15–30 |
| Twilio — Voice | Calls + number | $5–15 |
| ElevenLabs — TTS | Usage-based | $5–10 |
| Hosting | Mac (current) or VPS | $0–4 |
| Total | $80–150 | |
Full-power, no optimization. Founder-mode usage with Opus for all heavy lifting.
| Driver | Mechanism | Impact |
|---|---|---|
| Model tiering | Haiku for 80% of queries | -60% API cost |
| Shared infrastructure | VPS cost split 10 ways | -$0.40/user |
| Voice pooling | Shared Twilio number | -$1–3/user |
Key lever: Routing 80% of queries to Haiku at $0.0008/1K tokens vs. Opus at $0.015/1K — an 18x cost reduction per token on routine tasks.
| Driver | Mechanism | Impact |
|---|---|---|
| Multi-tenant infra | Single cluster, per-user isolation | -80% infra/user |
| API caching | Semantic cache for repeated queries | -25–40% tokens |
| Anthropic volume | Usage-based discounts begin | -10–20% API |
Key lever: Semantic caching eliminates redundant API calls. Common queries (guideline lookups, standard Q&A) hit cache instead of the model.
| Driver | Mechanism | Impact |
|---|---|---|
| Volume API pricing | Negotiated enterprise rate | -20–30% API |
| Aggressive caching | Cache hit rate 50–70% at scale | -50% tokens |
| Infra amortization | Fixed costs divided by 1K | <$0.10/user |
Key lever: At 1,000 users, Anthropic enterprise pricing + 60% cache hit rate brings effective per-token cost below $0.005/1K on average across model tiers.
Each product has a distinct cost profile based on query volume and model usage
Where 9 Enterprise sits in the market
| Product | Category | Price / Month | Notes |
|---|---|---|---|
| Devin | AI software engineer | $500 | 500 ACUs included. Usage-based above that. |
| ChatGPT Pro | AI assistant | $200 | Unlimited o1 + advanced features. Consumer-grade. |
| Cursor | AI code editor | $20–40 | Dev tool only. No workflow automation. |
| GitHub Copilot | AI code completion | $19 | Single-purpose. IDE-only context. |
| 9 Enterprise (at scale) | AI operating system | $15–60 | Full workflow automation, voice, comms, domain AI. Multi-product. |
Model tiering, semantic caching, and shared infrastructure are the three levers that make this business viable at scale.
Prepared by 9 Enterprises • March 26, 2026