AI Cost Modeling & Unit Economics

intermediate ~3 hours cost-management ai-strategy

A practical guide to AI cost modeling and unit economics for technical leaders who need to build defensible financial projections. This path bridges the gap between per-token API pricing and board-level business cases, covering macro AI economics, provider pricing structures, token consumption drivers, optimization levers, and ROI frameworks.

By the end of this path, you’ll be able to: build per-query cost models for AI features, project AI infrastructure costs at scale, identify the highest-impact cost optimization levers (which are architecture decisions, not tuning exercises), and construct ROI analyses that connect engineering costs to business value.

This path draws from a16z (market economics), Anthropic and OpenAI (provider pricing), and McKinsey (ROI frameworks) to give you both the technical cost drivers and the strategic value assessment.

Steps

  1. Who Owns the Generative AI Platform? ai-strategy article intermediate

    Analysis of value distribution across the generative AI stack — applications, models, and infrastructure — with data on margins, cost structures, and where economic value accrues.

    Start with the macro view: a16z's analysis reveals that AI applications face 50-60% gross margins because inference costs consume 20-40% of revenue. This fundamentally reframes AI product economics — your AI feature isn't zero-marginal-cost software, it's a service with real per-query costs that scale with usage. Internalize this before building your cost model, or your unit economics projections will be dangerously optimistic.

  2. Pricing anthropic-platform beginner

    Anthropic's pricing structure is your first data point for building a concrete cost model. Pay attention to three things: the per-token rates for input vs output (output is typically 3-5x more expensive), prompt caching discounts (up to 90% reduction), and batch processing rates (50% discount for non-real-time workloads). These aren't just optimization levers — they're the variables in your unit economics spreadsheet.

  3. Counting tokens openai beginner

    Count input tokens precisely for text, images, files, and tools using the Responses API — essential for cost management and context window planning.

    You can't model costs if you can't predict token consumption. The non-obvious insight: system prompts, tool schemas, and conversation context often consume more tokens than the user's actual message. For product leaders building cost projections, this means your per-query cost depends heavily on your architecture choices (how much context you include, whether you use tools, how long conversations persist), not just which model you select.

  4. Cost optimization openai advanced

    Lower your OpenAI model costs by trying our tools and strategies.

    OpenAI's comprehensive cost optimization guide ties together all the individual levers: model selection (10-30x cost range), caching, batching, structured outputs to reduce token waste, and monitoring. For technical leaders, the key takeaway is that cost optimization is an architecture decision, not a tuning exercise — the difference between a well-architected and a naive pipeline can be 10-50x in cost at scale.

  5. Batch API openai intermediate

    Learn how to use OpenAI's Batch API for processing jobs with asynchronous requests, increased rate limits, and cost efficiency.

    The Batch API offers a 50% cost discount for workloads tolerating up to 24-hour latency. For cost modeling, categorize your AI workloads into real-time (user-facing, latency-sensitive) and batch (evaluation, content generation, data processing). Routing batch workloads to batch APIs across providers is one of the highest-impact cost optimizations — and it's an architecture decision you make once, not an ongoing tuning exercise.

  6. The Economic Potential of Generative AI: The Next Productivity Frontier ai-strategy article beginner

    Comprehensive analysis of generative AI's economic impact across industries, quantifying $2.6-4.4 trillion in annual value potential and identifying which business functions will see the greatest transformation.

    Close the loop by connecting your per-query cost model to McKinsey's ROI framework. The question isn't just 'what does AI cost?' but 'does the value exceed the cost at scale?' McKinsey's analysis of $2.6-4.4 trillion in potential value gives you the macro context, while their function-by-function breakdown (customer operations, marketing, software engineering) helps you identify where AI ROI is strongest. Use this to build the business case your CFO and board will take seriously — grounded in both bottom-up cost modeling and top-down value assessment.