Conversational AI Firm CHAI Reports $68M ARR

- CHAI was founded in 2021 by William Beauchamp, a University of Cambridge economics graduate and former professional poker player. The company was among the first to launch a consumer-facing application for user-generated AI chatbots, initially utilizing the open-source GPT-J model. - The platform's technical approach includes model blending, which involves ensembling different LLMs, and reinforcement learning from human feedback (RLHF) to enhance user engagement. They also employ Proximal Policy Optimization (PPO), a reinforcement learning technique, to optimize their foundation model with the goal of reducing the probability of a chat session ending. - To manage its large-scale operations, CHAI built a 1.4 exaflops GPU cluster for in-house model inference, citing that this custom infrastructure provides nearly an order of magnitude improvement over open-source solutions like vLLM through optimizations such as custom kernels. - The company's AI safety framework is designed to align with the EU AI Act and the NIST AI Risk Management Framework. It includes a real-time classifier to detect self-harm ideation and employs anonymized auditing of conversational data to improve platform safety. - In July 2024, Belgian authorities opened an investigation into the company after a man's suicide was reportedly linked to his extensive conversations with a chatbot on the CHAI app. - The broader conversational AI market was valued at approximately $15 billion in 2024 and is projected to exceed $82 billion by 2034, with a compound annual growth rate of about 21%. - Prior to the current announcement, in July 2025, CHAI reported having over 10 million users and an annual recurring revenue of $40 million, with a lean team of 11 engineers.