“The demand for LLM inference and token generation currently exceeds the available supply of semiconductors, data centers, and electricity.”