Strategy question

Design a pricing model spanning serverless inference, fine-tuning, and dedicated GPU clusters.

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Strategic pricing design across a spectrum of usage patterns, from casual to committed enterprise workloads.

Segment customers by usage pattern: sporadic experimenters, steady-volume production apps, and large enterprises needing guaranteed dedicated capacity.
Propose serverless inference priced per token, matching the experimenter and moderate-volume production segments where usage is unpredictable.
Propose fine-tuning priced per training hour or per job, since it is a discrete, bounded cost separate from ongoing inference.
Propose dedicated GPU clusters priced as reserved capacity with a committed-use discount, matching enterprises with predictable, high, steady load.
Design a clear upgrade path: usage alerts when serverless costs approach the breakeven point where dedicated capacity becomes cheaper, guiding self-service migration.

Matches each pricing model to the natural usage shape of that segment instead of forcing one pricing structure across all three products.
Proposes a concrete breakeven calculator or alert as the mechanism that helps customers self-select into the right tier, reducing overpayment and churn.
Names the committed-use discount lever for dedicated clusters as the way to lock in large, predictable enterprise revenue.
Flags the tradeoff: too many pricing dimensions confuses customers, so the design must keep the three tiers clearly distinct and easy to compare.

Proposing one flat pricing model across serverless, fine-tuning, and dedicated clusters, which ignores how differently each is consumed.
Ignoring the self-service migration path from serverless to dedicated, forcing customers to negotiate manually at the breakeven point.

How would you notify a customer they have crossed the breakeven point into dedicated capacity being cheaper?
How would you price fine-tuning if usage is highly variable per customer?