Molecular design and computational screening in AI-driven drug discovery continue to accelerate, and experimental capabilities are emerging as a new competitive frontier. A growing number of companies have begun establishing their own laboratories to create data-driven R&D feedback loops. Terray Therapeutics offers a case in point: the company integrates high-throughput experimentation, automated workflows, and AI models, and has launched the EMMI platform. Using its proprietary ultra-high-density microarray chips, it generates vast amounts of small molecule–target interaction data and combines this with multiple AI modules serving different functions to form a complete R&D iteration loop. Meanwhile, the company is collaborating with NVIDIA to optimize vision-language models and extract chemical reaction information to supplement training data. To date, Terray has forged partnerships with multiple pharmaceutical companies, but its self-developed pipeline remains in the preclinical stage, and clinical progress on its first self-developed drug candidate has been delayed. Industry players in AI drug discovery have pursued differentiated development paths, and not all of them prioritize building their own laboratories. The key distinction lies in the type of data generated and whether that data can continuously feed back to optimize models. At present, the bottleneck in AI drug discovery has shifted to real-world constraints and execution on the experimental side. Laboratories are merely the entry point for R&D; ultimately, effective drugs must still be produced to validate the value of the platform.