In 2024 the data-center industry told Elon Musk what it told everyone: a computer the size he wanted takes 18 to 24 months to build. On September 2nd of that year, 122 days after construction began, xAI switched on a training cluster of 100,000 Nvidia GPUs, the specialised chips on which AI models are trained, inside a derelict refrigerator factory in South Memphis. It was the largest coherent AI computer ever assembled, and 92 days later it had doubled. How that happened is the clearest lesson available in what actually limits artificial intelligence today. It is not chips, talent or money. It is the queue.