A widely shared bull case for AI hardware spending gets picked apart, and the verdict is that the numbers behind the boom rest on very optimistic guesses — a warning sign for chip investors.

• The bull case says AI demand grows on two fronts at once — more users and more usage each — so the chip buildout can run far longer than the dot-com boom did.
• The flaw: the price of AI usage is collapsing, and cheap Chinese open models like Alibaba's Qwen cost a fraction of what top labs like Anthropic and OpenAI charge for nearly the same quality.
• The bull case also assumes companies will one day pay about $1,000 a month per employee for AI, when most businesses outside software today spend closer to $12.
• It projects OpenAI and Anthropic growing revenue 40% a year through 2030 to justify roughly 15 times more computing power than they have now — a guess dressed up as a forecast.
• The real money in AI likely goes to software companies selling useful products, not to the chips or the raw AI usage, both of which get cheaper fast.

Outlook: Chip and AI stocks can keep climbing over the next six to twelve months, but if this is the thinking driving prices, a bubble pop within two to five years looks likely.