Poolside, a Paris-based startup that plans to launch a large language model for developers, is reportedly in talks to raise a nine-figure funding round.

TechCrunch today cited sources as saying that the investment is expected to be worth over $400 million. According to the tipsters, Bain Capital Ventures and DST are holding discussions to co-led the round. The raise is expected to value Poolside at $2 billion.

Poolside is led by Chief Executive Officer Jason Warner, former chief technology officer of GitHub. The company’s other co-founder, Eiso Kant, previously launched a developer tooling startup called Athenian that was acquired by the Linux Foundation in 2023. 

Poolside reportedly moved its headquarters from the U.S. to Paris last August and raised $126 million in initial funding around the same time. The round is believed to have been backed by Bain Capital Ventures, one of the investors that could reportedly co-lead the company’s next raise. The firm was joined by Redpoint, Felicis and more than a half-dozen other backers.

According to its website, Poolside is developing an LLM optimized for coding tasks. The company plans to give the model an edge over other AI systems that target programming use cases through a training approach it calls reinforcement learning from code execution feedback. This approach will enable the LLM to “improve by completing millions of tasks in tens of thousands of real-world software projects,” Poolside detailed. 

The company’s website also indicates that it plans to build a custom stack, or software toolkit, for AI training tasks. Such toolkits typically include features that help researchers manage the datasets they use in machine learning projects. In many cases, AI training software also automates related tasks such as measuring the accuracy of newly developed LLMs.

Poolside detailed on its website that it plans to place a particular emphasis on training data quality in its development efforts. “Synthetic data generation, while seemingly counterintuitive, works and works particularly well for code,” the company added. Synthetic data is information generated by an AI for use in the development of other neural networks. 

Earlier this year, Poolside inked a deal to rent AI training infrastructure from Australian data center operator Iren. The contract is for a cluster of 504 H100 graphics processing units. The H100 was Nvidia Corp.’s flagship data center GPU until a product line refresh last November.

Building an LLM that can make developers more productive is only the initial focus of Poolside’s long-term development roadmap. Down the line, it plans to train AI models that will allow business users to create software. From there, Poolside hopes to apply its language models’ reasoning capabilities to other areas. 

Photo: Unsplash