CoreWeave offers cloud servers to large companies that train AI models and has agreed to acquire OpenPipe, a two-year-old Y-combination support startup that helps companies develop customized AI agents with reinforcement learning.
“Reinforcement learning has emerged as a vital force for enhancing model performance for agent and inference tasks,” said Brian Venturo, co-founder of CoreWeave, in a statement from TechCrunch. “By combining OpenPipe’s advanced self-learning tools with CoreWeave’s high-performance AI cloud, we are expanding our platform to provide AI Labs developers with key benefits in building scalable intelligent systems.”
CoreWeave and OpenPipe did not disclose terms of the transaction. In March 2024, Seattle-based OpenPipe attracted a $6.7 million seed round with supporters such as Costanoa Ventures, Y Combinator, Logan Kilpatrick of Google Deepmind, Github co-founder Tom Preston-Werner, and Github Copilot co-founder Alex Graveley.
The deal marks CoreWeave’s latest attempt to expand the stack up and down following the acquisition of weights and biases for the AI developer platform in March. OpenPipe develops a popular open source toolkit for creating AI agents called ART (Agent Reception Trainer). Many of CoreWeave’s biggest customers include major AI labs such as Openai, but the company is trying to appeal to small and medium-sized businesses.
More and more AI labs and startups are building enterprise products around reinforcement learning. This includes rewarding AI models for correct responses. Reinforcement learning proves a powerful way to improve the performance of AI models at specific tasks. The idea for these enterprise products is to train AI agents specialized in the needs of the company.
This kind of customer-specific training requires a lot of computing resources and by getting OpenPipe, CoreWeave wants both such services. The OpenPipe team will join CoreWeave, and OpenPipe customers will become CoreWeave customers.
