How a Small Startup’s Cloud Run Misstep Led to a $72,000 Bill – Lessons Learned

Preface

In 2020, many small startups collapsed because of the pandemic, and even those that survived were on the brink of bankruptcy.

Milkie Way, an eight‑person U.S. startup founded by former Google engineer Sudeep Chauhan, built a safety‑notification platform called announce.today . They planned an AI‑generated alert service (Announce‑AI) and chose Google Cloud Functions, but its 9‑minute timeout forced them to switch to Cloud Run.

Because Cloud Run provides no storage, they used Firebase as the database, even though a traditional SQL database would have been unnecessary for their small scale.

Sudeep set a $7 budget for the GCP project, assuming the service could not exceed that amount.

After deployment, routine testing began, and the next day the nightmare started.

Sudeep first received an email about an automatic Firebase upgrade, followed by a $7‑budget‑exceeded notice. His credit‑card limit was $100, so he was not alarmed. However, when he checked Google Cloud Billing, he saw a $5,000 invoice, which quickly ballooned to $72,000 within a few hours.

The rapid cost escalation was caused by several factors:

1. Firebase automatically upgraded without explicit consent. 2. Google’s billing reports have a one‑day delay, so the massive charge was discovered late. 3. Google ignored the $100 credit‑card limit. 4. To avoid Cloud Run timeouts, the team sent POST requests with URLs as data to multiple instances, causing each instance to process a single page without timeout. This introduced an unbounded recursive loop: • No break condition, so instances never stopped. • Identical URLs in POST requests caused infinite recursion, leading to exponential growth.

The “Hello World” version deployed on Cloud Run performed 116 billion reads and 33 million writes, costing roughly $69,600 based on Firebase’s read pricing (0.06 USD per 100 k reads).

This case demonstrates that deploying a flawed algorithm on a cloud platform without full understanding can generate astronomical bills. Serverless services are powerful but can be a double‑edged sword; misuse can be disastrous.

Google eventually waived the bill after Milkie Way reported the incident, saving the company from bankruptcy. However, the team spent months learning cloud architecture to prevent future mishaps.

Key takeaways: never rely solely on budget caps, always audit what your code does in the background, and treat cloud services with the same rigor as traditional code.