Methodology

The Leave the Cloud Calculator uses an equivalent service envelope approach -- not a naive raw-token comparison. This page explains how costs are modeled.

API Cost Model

API costs are computed as: monthly volume x (input token price x input token share + output token price x output token share). Volume grows by the configured monthly growth rate. API prices can be held flat, modeled as declining, or set to a custom trajectory.

  • Input tokens: typically 70-80% of request volume by count
  • Output tokens: typically 20-30% but can dominate cost due to higher price
  • Cache savings: optional placeholder, shown as a separate credit line
  • Price trajectory: configurable -- flat, -5%/yr, -10%/yr, or custom

Private Infrastructure Cost Model

Private costs include the full Rubin-era infrastructure stack -- not just rack cost.

  • Rack base: lease or amortized capex per rack per month
  • Networking (15% of rack base): ConnectX-9 / Quantum-X800 fabric
  • Storage and context memory (12%): AI-native inference storage tier
  • Facilities and liquid cooling (10%): power delivery and cooling allocation
  • Software and operations (8%): Mission Control, orchestration, monitoring
  • Power: kW draw x PUE x cents/kWh x hours/month
  • Capacity scales: when demand exceeds 95% of rated capacity, additional racks are added

Delivered Capacity Model

Delivered capacity is derived from benchmark cards, adjusted for utilization and availability.

Capacity (tokens/month) = benchmark TPS x utilization x availability x seconds/month x rack count

Swarm-Managed Premium

Swarm-managed cost applies a managed service multiplier (default 1.25x) to the self-managed private cost. This premium covers operations, support, SLA, and delivery management. The multiplier is quote-driven and should be replaced with a real Swarm proposal for procurement decisions.

Recommendation Logic

Recommendations weigh: cumulative economics over the full horizon, breakeven timing, capacity headroom, growth trajectory, operational appetite, privacy needs, and hybrid suitability. The product does not recommend solely on current monthly cost.

Data Provenance

Every value is tagged as one of:

  • Official: Direct from provider pricing pages
  • Benchmark: From InferenceX or published benchmark runs
  • Modeled: Derived from reference architecture assumptions
  • Quote-driven: Requires a real customer quote to confirm
  • User-entered: Values you have provided

This calculator is a planning and decision-support tool. It produces directional estimates, not guarantees. Lease pricing, benchmark transferability, and infrastructure assumptions should be verified with actual vendor quotes before committing to a procurement decision.