Projections & Profitability¶
This page covers the empirical cost model, per-user economics, tiered cost estimates, and breakeven analysis. For infrastructure line items, see Infrastructure.
Empirical Cost Model¶
Assumptions¶
Based on SaaS industry benchmarks and trading platform characteristics:
| Parameter | Value | Rationale |
|---|---|---|
| Pareto (80/20) | 20% power users drive 80% of compute | Industry standard |
| DAU/MAU | 50% of subscribers active daily | Conservative SaaS benchmark |
| Peak concurrency | 35% of total users running workers | DAU × 70% peak online |
| Worker idle timeout | 30 minutes | Balances cost vs UX |
| Pool startup | 897ms | Sub-second, transparent to users |
Market Coverage (~20 hr/day)¶
00 02 04 06 08 10 12 14 16 18 20 22 24 (UTC+8)
├─────────────────────── Crypto 24/7 ────────────────────────────────┤
├── TW Night ──┤ ├──── TW Night Session ────────┤
├── TW Day ──┤
├── US Market ────┤
Cost Formula¶
Peak workers = Total users × 0.50 (DAU) × 0.70 (peak) = 0.35 × N
Peak instances = ⌈ 0.35N / workers_per_instance ⌉
Off-peak inst. = max(1, ⌈ 0.05N / workers_per_instance ⌉)
Daily instance-hours:
Peak (8 hr): peak_instances × 8
Off-peak (16 hr): off_peak_instances × 16
Monthly cost = daily_instance_hours × 26 days × $0.258/hr
Example: 100 Users
Per-User Unit Economics¶
| Model | Per-User Cost | Gross Margin (NT$999) | Use Case |
|---|---|---|---|
| Worst case (24/7) | $3.15/mo (NT$101) | 89.9% | Theoretical upper bound |
| Empirical (Pareto) | $1.61/mo (NT$52) | 94.8% | Operating estimate |
| Best case (low activity) | $0.80/mo (NT$26) | 97.4% | Low engagement |
Tiered Cost Estimates¶
All-In Monthly Cost by Tier¶
| Tier | Peak EC2 | Total Cost (USD) | Per-User | Revenue (TWD) | Net Margin |
|---|---|---|---|---|---|
| 10 | 1 (fixed) | $485 | $48.50 | NT$9,990 | -55% |
| 16 | 1 (fixed) | $485 | $30.31 | NT$15,984 | Breakeven |
| 30 | 1 (fixed) | $485 | $16.17 | NT$29,970 | 48% |
| 100 | 1 | $495 | $4.95 | NT$99,900 | 84% |
| 500 | 3 | $695 | $1.39 | NT$499,500 | 96% |
| 1,000 | 6 | $1,210 | $1.21 | NT$999,000 | 96% |
| 5,000 | 15 (2xl) | $4,300 | $0.86 | NT$4,995,000 | 97% |
| 10,000 | 14 (4xl) | $7,920 | $0.79 | NT$9,990,000 | 97.5% |
| 50,000 | 35 (8xl) | $34,000 | $0.68 | NT$49,950,000 | 97.8% |
| 100,000 | 70 (8xl) | $64,000 | $0.64 | NT$99,900,000 | 98.0% |
Detailed Breakdown: 10,000 Users¶
| Component | Spec | Monthly (USD) |
|---|---|---|
| EC2 Workers | 14 × r7g.4xlarge (peak) | $3,871 |
| API EC2 | 4 × m7g.xlarge (16 tasks) | $596 |
| ALB | ~60 LCU | $365 |
| Lambda Orchestrator | ~10M invocations | $100 |
| ElastiCache Valkey | Serverless 10 GB, 100K ECPU | $1,000 |
| RDS Aurora Serverless v2 | 8-16 ACU (Multi-AZ) | $850 |
| SES | ~7.8M emails/month | $780 |
| CloudWatch | Full monitoring | $200 |
| WAF | Security rules | $100 |
| Other | VPC Endpoint, Secrets, ECR, backup | $58 |
| Total | $7,920 |
Revenue & Profitability¶
Annual Projections¶
| Users | Annual Revenue (TWD) | Annual Cost (TWD) | Annual Profit (TWD) | Net Margin |
|---|---|---|---|---|
| 100 | NT$1,198,800 | NT$190,080 | NT$1,008,720 | 84% |
| 500 | NT$5,994,000 | NT$266,880 | NT$5,727,120 | 96% |
| 1,000 | NT$11,988,000 | NT$464,640 | NT$11,523,360 | 96% |
| 5,000 | NT$59,940,000 | NT$1,651,200 | NT$58,288,800 | 97% |
| 10,000 | NT$119,880,000 | NT$3,041,280 | NT$116,838,720 | 97.5% |
Breakeven Analysis¶
Fixed costs (regardless of user count):
EC2 Worker (r7g.xlarge × 1): $189/month (always-on baseline)
API EC2 (m7g.large × 1): $74/month
ElastiCache Valkey Serverless: $91/month (1 GB minimum)
RDS Aurora Serverless v2: $55/month (0.5 ACU minimum)
Lambda Orchestrator: $9/month
ALB: $16/month
VPC Endpoint (3 AZs): $28/month
WAF + SES + Other: $23/month
──────────────────────────────────────
Total fixed: ~$485/month = NT$15,520
Breakeven: NT$15,520 ÷ NT$999 = 15.5 → 16 paying users
Sensitivity Analysis¶
The empirical model assumes 50% DAU and 70% peak concurrency. Varying these assumptions shows the model is robust:
| Scenario | DAU | Peak Factor | Per-User Cost | Gross Margin |
|---|---|---|---|---|
| Conservative (baseline) | 50% | 70% | $1.61/mo | 94.8% |
| Low engagement | 40% | 60% | ~$1.05/mo | 96.6% |
| High engagement | 60% | 80% | ~$2.15/mo | 93.1% |
| Power-user heavy | 50% | 90% | ~$2.45/mo | 92.1% |
Even in the worst-case scenario (high engagement, power-user heavy), gross margin stays above 90%. The 30-minute idle timeout is the primary cost control lever: shortening it reduces per-user cost at the expense of startup latency for returning users; lengthening it improves UX but increases compute cost.
See Optimization for Reserved Instances, Graviton, and Spot strategies.
Report updated: 2026-03 | Region: ap-southeast-1 | Exchange rate: 1 USD = 32 TWD