Skip to content

Anthropic and OpenAI Have Found Product-Market Fit

Source: I think Anthropic and OpenAI have found product-market fit \ Date Published: 2026-05-27 \ Author/Org: Simon Willison \ Time to read: 15 min

TL;DR

Anthropic and OpenAI have achieved genuine product-market fit — but not with ChatGPT or Claude chat. The breakthrough is coding agents (Claude Code/Cowork, Codex). Evidence: enterprise pricing has shifted to full API rates ($200+/month/seat), power users burn $1,000+/month in tokens, and the SpaceX-Anthropic cloud deal commits to $1.25 billion per month through May 2029. The narrative of "AI hitting a wall" is mostly press distortion of mild executive quotes.

The Pricing Earthquake

Simon Willison noticed that his personal $200/month Claude Max plan gave him about $2,180 worth of tokens — a massive discount. He assumed enterprises got similar deals. He was wrong.

Anthropic's Pivot

  • Enterprise plan moved to $20/seat/month + API pricing for usage
  • Change reportedly occurred November 2025, but customers discover it upon renewal
  • The original plan ("Claude seats include enough usage for a typical workday") is gone

OpenAI's Pivot

  • Codex pricing aligned with API token usage, effective April 2026
  • Costs quoted in "credits" that exactly match API token costs

New Model Pricing (April 2026)

Model Price Relative to Predecessor
GPT-5.5 2× the API price of GPT-5.4
Opus 4.7 ~1.4× the price of Opus 4.6

Net effect: Enterprise customers locked into year-long deals at higher API prices, removing previous extreme discounts.

Why This Signals Product-Market Fit

The ChatGPT Math Problem

  • 900M weekly active users, but only 50M paying (5.6%)
  • $10–20/month/user is decent, but doesn't cover trillion-dollar infrastructure costs

The Agent Solution

  • $200+/month/user scales much faster
  • Power users burn $1,000+/month in API costs per vendor
  • Coding agents burn vastly more tokens and are daily drivers for well-compensated professionals
  • Agents automate anything you can do by typing commands into a computer — applicable far beyond software engineering

Hiring Spree

Both labs are hiring heavily for enterprise sales: - OpenAI: 703 open jobs, 32.6% enterprise-oriented (account executives, Forward Deployed Engineers) - Anthropic: 390 open jobs, 26.9% enterprise-oriented - Irony: AI labs picked a business model with heavy demand on human labour

(Willison ran this analysis by scraping job sites with Claude Code, piping into Datasette Cloud via JSON API, and analysing with Datasette Agent — dogfooding all the way.)

The "AI Failure" Narrative Is Thin

Uber

  • Headline: "Uber maxed out its full year AI budget just a few months into 2026, mostly thanks to Claude Code"
  • What the COO actually said (Rapid Response podcast): "25% of our code commits were via Claude Code last quarter… it's hard to draw a line between that stat and producing 25% more useful consumer features"
  • Willison's take: A budget set in 2025 failing to predict 2026 demand is unsurprising. The COO's quote was notably mild, but press blew it up.

Microsoft

  • Cancelling Claude Code licenses to encourage dogfooding Copilot CLI
  • The Verge: "sources tell me the decision is also a financial one", triggered by June 30th end of financial year

"The best advice I ever heard on pricing a product was that your customer should suck air through their teeth and then say yes. Uber's budget overrun and Microsoft's seat cancellations look like that effect playing out in practice."

The SpaceX Deal — $1.25B/Month

"In May 2026, we entered into Cloud Services Agreements with Anthropic PBC… with respect to access to compute capacity across COLOSSUS and COLOSSUS II. Pursuant to these agreements, the customer has agreed to pay us $1.25 billion per month through May 2029…"

  • Implied to be for inference, not training
  • This is from one vendor; Anthropic has vast compute from others too

Cutting Out the Middlemen

  • Historically, Cursor and GitHub Copilot generated $1.2B of Anthropic's ~$4B revenue (Aug 2025)
  • Today, Claude Code and Codex directly compete with those middlemen
  • No wonder Cursor is investing in their own models

Key Takeaways

  1. Anthropic and OpenAI have found product-market fit through coding agents (Claude Code, Codex), not general chatbots — evidenced by enterprise pricing shifts and massive token consumption by high-value professionals
  2. The "AI failure" narrative is largely press distortion: Uber's COO made a mild observation that was blown out of proportion, and Microsoft's Claude Code cancellations were financially motivated end-of-year decisions
  3. The SpaceX-Anthropic deal ($1.25B/month through 2029) signals that inference demand for frontier models is orders of magnitude larger than most analysts assume