Case Study: Why Coinbase chose Ray and Anyscale for its ML infrastructure
Key results
The challenge
Coinbase's ML engineers were spending too much time waiting before they could iterate, with failed pull requests costing whole days of lost work. The team wanted to iterate faster, scale smarter, and operate more efficiently across its machine learning platform.
The solution
Coinbase adopted Ray and the Anyscale platform for its ML infrastructure, using distributed data processing to speed up last-mile data transformation and Anyscale Job Queues to run large batch ML workloads reliably at scale.
“Ray and Anyscale aligned with our vision: to iterate faster, scale smarter, and operate more efficiently.”
WLWenyue LiuSenior Machine Learning Platform Engineer, Coinbase
The results, in context
After moving to Anyscale, Coinbase trained 15x more jobs at the same cost compared to its original solution. Last-mile data transformation dropped from 120 minutes to 15 minutes (8x faster), the team gained the ability to process training datasets with 50x more volume, and iteration cycles fell from around two hours to seconds.