The current generative AI boom has become as much an engineering and infrastructure challenge as a technological breakthrough. His central claim is not that AI itself is failing, but that today's large language models are extraordinarily resource-intensive, consuming enormous amounts of computing power, memory, storage, electricity, and water. As AI companies race to expand their models, they are absorbing a significant share of the global supply of high-end computer memory, contributing to shortages and driving up prices for laptops, storage devices, and other computing hardware used by businesses and consumers.
The article argues that these costs extend well beyond AI companies. Building and operating AI models requires massive investments in data centers, GPUs, networking equipment, and energy infrastructure. This surge in demand is straining global supply chains and increasing costs across the technology industry. According to the article, the AI boom is creating a trillion-dollar infrastructure buildout whose economic effects are being felt even by people who do not use generative AI directly, through higher hardware prices and growing pressure on electricity and digital infrastructure.
Reisner also questions whether the industry's current trajectory is sustainable. He suggests that many leading AI companies continue scaling their models despite diminishing improvements relative to the rapidly increasing costs of training and inference. Investors and technology firms are committing unprecedented amounts of capital to larger models and new data centers, but the article argues that the engineering challenge is shifting from simply building more powerful AI to making AI more efficient, affordable, and sustainable. Without major breakthroughs in algorithms or hardware efficiency, the economic and environmental costs of continued scaling could become increasingly difficult to justify.
The article concludes that the future of generative AI will depend not only on producing smarter models but also on overcoming the engineering constraints that underpin them. Advances in specialized chips, memory technology, energy-efficient computing, and optimized AI architectures will be essential if AI is to scale sustainably. Rather than viewing AI solely as a software revolution, Reisner argues that its long-term success will hinge on solving the physical infrastructure challenges that support modern AI systems.
