AI companies are purchasing 70 percent of the world's high-end computer memory, causing shortages that have doubled hard-drive prices and increased laptop costs by up to 50 percent. Affordable entry-level computers may "disappear by 2028," and data-center capacity is planned to multiply by eight, with some firms repurposing jet engines for power. Unlike video streaming, smartphones, or cloud software, generative AI does not scale efficiently. Models have grown from 175 billion parameters in 2020 to more than 1 trillion today, yet the cost per user does not decrease. Sam Altman suggested "10 gigawatts of compute" might cure cancer, but returns are diminishing; LLMs scale quadratically, not logarithmically, making them perhaps the worst technology ever deployed by economic measures.

Ilya Sutskever noted companies prefer the brute-force approach because it offers a "very low-risk way of investing your resources." Profitability remains an open question. Epoch AI graphs show exponentially increasing costs to serve more tokens. Dario Amodei has touted "compute multipliers," but no evidence shows quadratic scaling has been overcome.

Researchers like Alexia Jolicoeur-Martineau call the approach "a bit insane." She won a $50,000 prize for a tiny recursive model that uses far fewer resources, proving efficient alternatives exist but lack funding.