What does 1M tokens really cost?
On-Prem vs Cloud GPUs vs APIs
On-Premise
Your own hardware
W
€
tok/s
€/kWh
Cost per 1M tokens
€3.84
Electricity: €0.28 | Hardware: €3.56
Full Self-Hosting Equation
Electricity: (W × €/kWh) / (3.6 × tok/s)
Hardware: (CGPU × Q × 1000) / (L × U × 3.6 × tok/s)
* Excludes cooling, maintenance, and infrastructure
Cloud GPU
Rented compute
€/hr
tok/s
Cost per 1M tokens
€4.17
Formula
(€/hr / (tokens/s × 3600)) × 1,000,000
* Often 30-50% cheaper than hyperscaler on-demand pricing
API Providers
Pay per token
M tok
M tok
Monthly Cost
€60.00
Per 1M tokens
€5.00
Input: €30.00 | Output: €30.00
Current Model Pricing
Input:
€3.00/M
Output:
€15.00/M
Comparison Summary
| Option | € / 1M tokens | Monthly (10M tokens) | Savings vs API |
|---|
Assumptions: On-premise includes electricity and hardware amortization (excludes cooling, maintenance, infrastructure).
Cloud GPU pricing based on independent providers (RunPod, Lambda Labs, Vast.ai).
API pricing as of January 2026. Actual costs vary based on usage patterns, batch discounts, and commitments.
Sales Notes
- * Key Benefit 1: Helps prospects understand true TCO of AI infrastructure
- * Key Benefit 2: Shows self-hosted models can be 10-100x cheaper than APIs
- * Key Benefit 3: Demonstrates Dypsis expertise in AI cost optimization
- * Ideal Customer: Companies spending >€10k/month on API calls or considering GPU infrastructure