a16z: AI Apps Shouldn't Price Per Token — Price at the Highest Value Layer
aleximm · x · 2026-11-07
a16z investors Tugce Erten and Sarah Wang argue in a new essay that application-layer companies importing token-based pricing from model providers is a common mistake.
Key points:
- Token pricing made sense in 2020 when OpenAI's API metered raw inference;
- Post-ChatGPT applications bundle proprietary data, tools, orchestration, integrations and workflow logic to complete work for customers;
- Pricing in tokens imports the model provider's cost structure and anchors value to a unit whose cost keeps falling.
Their prescription: price at the highest layer of value a company can claim, not in tokens.
More from Venture
- Rippling acquires Reprompt AI to scale its agentic AI efforts — andreisavu · 2026-11-07
- Exploits now take hours while fixes take weeks — a16z maps 6 security startups — a16z · 2026-11-07
- Stripe makes incorporating a company with a bank account free via $500 Atlas credit — jeff_weinstein · 2026-11-07
- Investor: ~60% of my recent investments came from a16z portfolio alumni — tkexpress11 · 2026-11-07
- Hudson River Trading starts trading on Kalshi as Wall Street piles into prediction markets — ZeroStateReflex · 2026-11-07
- Parse playbook repeats: Meta's Conductor deal kills cloud 16 days after acquisition — Arindam_1729 · 2026-11-07