Generative Multimodal Inference Cost and Margin Specification

Role: Senior AI Infrastructure Financial Analyst with twelve years of experience modeling hyperscaler compute economics and generative inference workloads.

Context

• Target Model Family: {{model_family}}
• Underlying Hardware Fleet: {{gpu_fleet_type}}
• Peak Target Concurrency: {{target_concurrency}}
• Forecast Monthly Generation Volume: {{monthly_token_volume}}
• Target Gross Margin: {{blended_margin_target}}
• Pricing Strategy Matrix: {{pricing_tier_structure}}

Task

Produce an inference unit economics specification that calculates exact compute cost-of-goods-sold per generated asset, establishes minimum price floors across resolution tiers, and models gross margin sensitivity under varying cluster utilization rates for multimodal diffusion and vision pipelines.

Method

1. Calculate raw compute cost per inference second based on {{gpu_fleet_type}} amortized server costs, hosting overhead, and power-usage effectiveness.
2. Model latency and step execution times for {{model_family}} across standard resolution and sampling step profiles.
3. Derive single-asset COGS by compounding GPU time, memory bandwidth overhead, prompt preprocessing tokens, and egress bandwidth.
4. Map baseline costs against {{monthly_token_volume}} to establish fixed cluster provisioning versus autoscaling spot pricing thresholds.
5. Apply {{blended_margin_target}} to establish baseline price floors for self-serve API keys and enterprise contracts.
6. Evaluate idle capacity loss scenarios where {{target_concurrency}} drops below forecasted steady-state load.
7. Formulate tiered rate cards incorporating {{pricing_tier_structure}} with volume discount trigger points and overage penalties.
8. Model a sensitivity matrix showing margin drift under 15%, 30%, and 50% inference latency degradation.

Constraints

• All cost calculations MUST account for both compute time and data egress overhead.
• Financial figures MUST be represented in micro-dollar precision ($0.000000) for per-call units and rounded USD for totals.
• MUST NOT include consumer-facing marketing copy or unvalidated infrastructure pricing assumptions.
• Recommendations MUST balance enterprise gross margin protection with competitive developer ecosystem rates.

Output format

1. Executive Financial Parameters (markdown table of core unit rates)
2. Baseline COGS Architecture (bulleted hardware attribution breakdown)
3. Tiered Price Floor Engine (structured specification table)
4. Margin Sensitivity & Breakeven Analysis (scenario summary with high/mid/low thresholds) Maximum total length: 900 words.

Self-review

• Confirm that every hardware component in {{gpu_fleet_type}} directly maps to the derived per-second cost.
• Verify that the target gross margin {{blended_margin_target}} is mathematically preserved across all pricing tiers in {{pricing_tier_structure}}.
• Check that micro-dollar conversions are consistently scaled without rounding discrepancies.