Skip to main content
← All posts

Who pays for AI: what a16z's spending data means for API buyers

Published

a16z's Top 100 Consumer AI Apps, Seventh Edition (October 2026) added a ranking based on observed consumer spending, using YipitData's U.S. card data. A few of its numbers explain why usage-based billing exists next to flat plans. The figures below are quoted from a16z with the source linked; the charts and the reading of them are ours.

Spending is concentrated in a few heavy users

Share of all observed U.S. consumer AI spend, as of August 2026Source: a16z, Top 100 Consumer AI Apps – Seventh Edition (Olivia Moore, Oct 5, 2026)
  • Top 1% of payers19.5%
  • Bottom 50% of payers16.6%

The top 1% of payers account for 19.5% of observed spend, which is more than the bottom half of payers combined (16.6%).

Monthly consumer-card spend on AI, as of August 2026
WhoPer month
Top 1% of payers (average)$903
Median AI payer$25
Ratio (our arithmetic)about 36×
Monthly consumer-card spend on AI, as of August 2026Source: a16z, Top 100 Consumer AI Apps – Seventh Edition (Olivia Moore, Oct 5, 2026)

a16z notes that the top 1% figure is card spend and that work usage may be even higher. The gap between a typical payer and a heavy one is the point: any single flat price is a compromise between them.

How AI-native products charge

Monetization models among the 44 AI-native products in a16z's top web rankingsSource: a16z, Top 100 Consumer AI Apps – Seventh Edition (Olivia Moore, Oct 5, 2026)
  • Paid subscriptions84%
  • Usage charges / extra credits64%
  • Ads14%
  • Transaction or platform fees2%

Products can use more than one model, so the shares add up to more than 100%. Subscriptions lead, and usage charges or extra credits come second: the two are used together rather than instead of each other.

What this means if you buy tokens

  • Averages hide the distribution. A flat plan priced for the typical payer fits the heaviest users badly, and the heaviest users are where much of the spend sits.
  • The supply side sees the same effect. SemiAnalysis writes that, “Despite being just 10% of overall revenue, subscriptions can take up over 40% of inference compute” (source).
  • If your usage is heavy or automated, compute your own break-even instead of trusting an average: see LLM API vs chat subscription.
  • If you use several models, one key and published per-token prices keep the comparison honest: see the price comparison.

Caveats

  • a16z describes its spending data as “U.S.-only and drawn from panels rather than a census”, so treat the figures as indicative.
  • The numbers describe consumer subscriptions and card spend, not API usage by businesses.

Sources