PortfolioC3-3

AI Investment Strategy and Portfolio Governance

Inference Run Rate Forecaster

SIMULATEDVerified Jul 2, 2026

Inference strategy is a run rate decision. This artifact compares API based usage and self hosted capacity over time, then identifies where the economics cross and which assumption moves the crossover most. (Per-call economics live in GAP-06.)

Same instrument · three industries pick a use case to reconfigure the run

Prefer to read? The two minute case study · problem → approach → metric → outcome

Problem

API usage offers flexibility, but cost scales with volume. Self hosting can reduce marginal cost after utilization reaches the right level, but it introduces fixed capacity, operations, infrastructure, and talent requirements, and the decision depends on utilization more than sticker price.

Approach

The forecaster projects API and self hosted costs across 24 months. It marks the crossover point and shows how growth, token volume, frontier model share, utilization, and staffing assumptions move the decision.

Why this way

This connects AI operating strategy to budget planning, unit economics, platform investment, infrastructure commitments, and cost governance.

The metric

The crossover month; 24-month cumulative cost each way.

The trade-off

API is flexible pay per use; self host is fixed capacity that only amortizes past the crossover.

Outcome

The crossover month with the assumption that moves it most made explicit.

0.5M/mo
6%
3,000
40%
60%
1.5

Monthly run rate · 24 months

API Self host
mo 1mo 12mo 24$185k
The cliff
N/A

Beyond 24 mo

API · 24-mo total
$686k

Cumulative

Self host · 24 mo total
$3.22M

Cumulative

What pulls the break even forward · crossover under each single move

  • Growth → 12%/mono cliff
  • Frontier share → 20%no cliff
  • Tokens/call → 4,500no cliff
  • Utilization → 85%no cliff

Each row recomputes the crossover with one assumption changed, earlier (green) means self host pays off sooner.

No cliff inside 24 months

At these assumptions API stays cheaper for all 24 months. Raise growth or lower the frontier model share to bring a cliff into view, or accept that self host doesn't pay yet.

If you act on this · the call → expected lift → how you'd measure it

The call

Revisit inference strategy when volume and utilization make the run rate materially different from the pilot economics.

Expected lift · illustrative

Avoids both premature self hosting and uncontrolled API spend.

How you'd measure it

Monthly run rate, cumulative 24 month cost, utilization, cost per task, crossover month.

Steering committee takeaway: The crossover is not decided by vendor claims. It is decided by utilization, growth, and the share of workloads that truly need higher cost models.

How this is built & assumptions

API/mo = volume × tokens/call × blended price ($3 to $18/1M tokens by frontier share). Volume compounds at the monthly growth rate.

Self host/mo = ⌈tokens ÷ (cluster capacity 2.5B × utilization)⌉ × $38k amortized + ops FTE × $22k. The cliff is the first month self-host < API.

Stack: Next.js (static) + shared design system; deterministic client side.

Limitations: this model uses simplified cost assumptions. Production forecasting would require current pricing, workload profiles, infrastructure benchmarks, reliability requirements, and finance approved cost allocation.