Bring on the spending caps · ↗ simonwillison.net

Simon Willison has a good post arguing for hard budget caps by default on pay-for-usage services and APIs. Large compute providers like AWS and Google Cloud have traditionally used soft caps, i.e., you receive an email at 2 AM warning that your spending has reached 80% of your limit, and then you wake up to a $2,000 bill after some bot finds your demo or (more optimistically) it goes viral.

Thankfully, both cloud providers have recently introduced hard spending caps. For disposable projects, demos, experiments, etc., I’ve always liked the prepaid-credit and per-key spending caps of services like OpenRouter, but these changes make the big cloud providers more viable for these use cases.

There’s a discussion of the article on Hacker News, with commenter motionlessveloc arguing that hard budget caps are a nightmare for businesses that need to provide reliable services:

It was, unfortunately, a nightmare. There were tons of tickets and even threats of lawsuits from customers whose service got cut off hard at the worst possible time due to organic growth/going viral/big event/nobody knew about the limit/etc. Not only did they lose all the leads and revenue they would have gotten from that bump, but they also pissed off their own existing users who suddenly couldn’t use the service either.

Generally speaking, it’s much better to use alerts instead of hard limit. Even in the worst case (hackers pwn your credentials and mine Bitcoin or whatever) the rest of your business is unaffected and you can negotiate with the billing department at comparative leisure.

I can buy this argument for businesses where continuity of service is paramount, but for individuals and hobbyists, there’s no question that hard caps should be the default. Individual users hate spending uncertainty; no one needs to be bankrupted when their repo that turns your GitHub profile into a medieval illuminated manuscript goes viral on Hacker News. And anyhow, individual users don’t have much leverage to negotiate with big providers if spending goes haywire.

This is also why most individual LLM users prefer fixed-cost subscriptions rather than gambling on APIs, especially for agentic use, since the connection between user intent and token consumption is still so hazily understood.

Bring on the spending caps!