ComputePrices: real-time GPU pricing across 8+ cloud providers
I built a product from scratch that helps ML engineers find the cheapest GPU instances across cloud providers — with real-time pricing, filterable comparison, and affiliate monetization.
Posted by
Gene KobilanskyRelated reading
Headline Goat: A/B testing in a single binary
A self-hosted A/B testing tool for headlines — one Go binary with embedded SQLite, no external dependencies. Add it to any site, test your copy, and get statistically significant results.
Objectives
- 🙋🏻♀️ Help ML practitioners quickly compare GPU pricing across cloud providers
- 🙋🏻♀️ Build a sustainable product with affiliate revenue
- 🙋🏻♀️ Create a fast, SEO-optimized comparison tool
- 🙋🏻♀️ Aggregate data from providers with wildly different APIs and pricing models
Results
- ✅ Live at computeprices.com with organic search traffic
- ✅ Real-time pricing from 8+ providers (AWS, RunPod, Vast.ai, Lambda, CoreWeave, and more)
- ✅ Affiliate partnerships generating revenue
- ✅ 215+ commits — built from domain purchase to production in weekends
Highlights
Multi-Provider Aggregation
Each cloud provider structures their pricing differently — some have APIs, some have HTML tables, some change formats without notice. I built a scraping pipeline that normalizes all of this into a single, filterable interface so you can compare an H100 across every provider in seconds.
Affiliate Monetization
Rather than charge users, I partnered with GPU cloud providers directly. Every "Visit Website" click carries a referral link — the site earns commission when users sign up through ComputePrices. Revenue without friction.
Challenges and Solutions
Data Reliability
Cloud providers change their pricing pages, API schemas, and availability constantly. I built resilient scrapers with error handling and fallback strategies so the site stays accurate even when upstream sources break.
SEO for Data-Heavy Pages
Comparison pages with hundreds of GPU configurations need careful optimization — proper metadata, structured data, fast load times, and useful filtering so both search engines and users can navigate the data effectively.
Built in public, built to last
ComputePrices started as a weekend experiment — "just bought computeprices.com, let's see if I can make something useful." Two weekends later, the first version was live. It's now a real product with real users and real revenue. Need a product built from zero to one? Let's chat.