As research involving large language models (LLMs), particularly agentic AI applications, gathers pace, universities are finding it increasingly difficult to secure affordable and reliable access to the computing capacity they need.