I've underestimated API costs more than once — the usage-based pricing model makes it easy to scale spend faster than you scale revenue, and the bill arriving at the end of the month is usually a surprise.
APIs have transformed software development — instead of building every capability from scratch, developers buy access to infrastructure, data, intelligence, and services through programmatic interfaces. This has accelerated what teams can build enormously. It has also introduced a new class of cost that behaves differently from traditional software expenditure: it scales with usage rather than user count, can change without notice, and is remarkably easy to accumulate without realising until the billing statement arrives.
Pricing Models
API pricing falls into several distinct structures, each with different implications for cost predictability and optimisation:
Per-request pricing: You pay a fixed amount for each API call, regardless of what the call does or returns. Common for communication APIs (Twilio: approximately $0.0075 per SMS), simple data lookup APIs, and verification services. Cost is directly proportional to usage volume and easy to forecast from usage data.
Per-token or per-unit pricing: Used by AI and language model APIs (OpenAI, Anthropic, Google). Pricing is based on the amount of data processed — typically measured in tokens for language models. GPT-4 Turbo costs approximately $0.01 per 1,000 input tokens and $0.03 per 1,000 output tokens. A single complex query generating 1,000 tokens of output costs $0.03; ten thousand such queries cost $300. Costs scale with both call volume and response length, making them harder to predict without monitoring.
Tiered pricing: A fixed monthly fee for a specified usage volume, with additional charges for overages. Common for mapping APIs (Google Maps: $2-$7 per 1,000 requests depending on API type, with a $200 monthly credit), weather APIs, and financial data APIs. Predictable within tier, unpredictable if usage spikes above the tier ceiling.
Subscription with rate limits: A fixed monthly fee allowing up to a specified number of calls per month or per second. Many data enrichment and B2B APIs use this model. The cost is predictable; the risk is hitting rate limits at peak usage moments that cause application failures.
Usage Impact
The challenge with API costs is that usage is not always controlled by the product team. End user behaviour drives it. A feature that calls an AI API on each user message has costs that scale with user engagement — more engaged users produce higher API costs, which can produce a surprising dynamic where your most active users are your most expensive.
Consider a consumer app with 10,000 monthly active users, each sending an average of 20 AI-powered messages per month. Each message generates 500 tokens of AI response. Monthly API cost: 10,000 × 20 × 500 / 1,000 × $0.03 = $3,000/month. At 50,000 MAU with the same usage pattern: $15,000/month. At 200,000 MAU: $60,000/month. API costs that are manageable at early scale become significant infrastructure costs at growth scale — costs that must be incorporated into unit economics from the start.
The API Cost Calculator models monthly and annual API spend from usage parameters. Enter your daily call volume, pricing per call or per token, and expected growth rate to see projected costs at different scale milestones.
Examples
Email verification API: A SaaS product verifying new user email addresses at signup. At 500 signups/month and £0.004 per verification: £2/month. At 50,000 signups/month: £200/month. Trivial at small scale; worth monitoring as a line item at larger scale.
Mapping API: An app making 5 Google Maps API calls per session at $0.005 per call. At 2,000 daily active users making 3 sessions each: 30,000 calls/day × $0.005 = $150/day = $4,500/month. This is a material cost that is not always visible in early development when sessions are far lower.
AI writing assistant: A B2B tool making AI API calls for document summarisation. Average 2,000 tokens per summary at $0.015 per 1,000 tokens = $0.03 per use. At 500 uses per day: $15/day = $450/month. At 5,000 uses per day: $4,500/month. The product must charge enough per user to cover both this cost and CAC, margin, and infrastructure — which requires modelling these costs against the pricing and user volume plan from the outset.
Budgeting for API Costs
API costs should be treated as a variable cost of goods sold, not a fixed infrastructure cost — because they scale with usage rather than remaining constant. Include them in your cost per user or cost per transaction calculation from the earliest stages of product planning. A product whose gross margin looks healthy at current usage may have an eroding margin as volume grows if API costs are not factored in correctly.
Set up billing alerts with every API provider — most offer configurable threshold notifications. A notification at 50% and 80% of expected monthly spend gives time to investigate and respond before costs significantly overshoot forecast. The absence of billing alerts is one of the most common reasons API cost overruns are discovered only on the monthly statement rather than in real time.