Cloud · Vercel · June 7, 2024
$96,000/wk — artists fled Meta's AI policies to Cara, and its serverless bill scaled as fast as its users
Vendor
Vercel
Service
Vercel (serverless functions + bandwidth)
Amount
$96,000/wk (reported)
Timeframe
40k to 650k users in a week
Cause
Viral growth spike
Outcome
Public appeal
Who's exposed: Anyone running server actions or serverless functions on Vercel without spend management turned on. Autoscaling keeps the app up during a viral spike but raises the bill in lockstep, with no hard cap by default.
Resolution: The founder appealed publicly for Vercel to lower the charge. Readers of the thread flagged that the $96,000 covered a single week and left out egress and Vercel Storage and Database, so the real spend ran higher than the headline number.
Widely reported from the founder's public posts and covered by TechCrunch; community threads note the $96,000 was a single week and excluded egress, storage, and database costs. TechCrunch (Cara growth report)
What happened
Cara is a portfolio and social app for artists, founded by photographer Jingna Zhang. In early June 2024 artists began leaving Instagram over Meta's policy of training AI on user content, and Cara went viral, growing from about 40,000 to more than 650,000 users in a week. The app runs on SvelteKit hosted on Vercel and leans on serverless functions and server actions, including image scanning tied to its anti-AI stance. Autoscaling kept the app online through the surge, but it scaled the bill too: about $96,000 for a single week. Developers reading the numbers pointed out that the figure was one week, not a month, and that it did not include external services, egress, or Vercel Storage and Database, so the true spend was higher. The founder appealed publicly for the charge to be reduced.
Root cause
Serverless autoscaling ties cost to traffic with no hard ceiling by default. Cara's per-request work, from server actions and API routes to image scanning on upload, multiplied across a roughly 16x jump in users. Vercel resells underlying compute and bandwidth at a markup, so a viral spike converts almost directly into a large bill. There was no spend cap or alert to catch it before a full week of usage had stacked up.
How to avoid it
•
Turn on Vercel spend management and set a hard budget with alerts. Autoscaling raises cost with traffic, so a cap and an alert catch a viral spike before a week of usage stacks up.
•
Run heavy per-request work like image scanning on a server or queue, not per-invocation serverless. One Node server or a worker pool handles the same load at a flat cost instead of billing every call.
•
Watch usage dashboards daily during a growth spike, not at month end. A 16x jump in days can run into six figures before the first invoice arrives.
CostGoat watches your Vercel bills, so a surprise like this reaches you as an alert, not an invoice.
More Bill Shock stories
February 27, 2024
$104,500 — a DDoS on one sound file ran a free static site to a $104K bill
November 3, 2023
$121,000 — a buggy auto-translate function called a paid API six billion times in two days
August 28, 2022
$213,000 — a hacked account ran Lambda in six regions overnight and support first said it was on you
May 18, 2026