Sam Altman's AI dream has a mind-boggling price tag, according to the WSJ: Somewhere between $5 trillion and $7 trillion.

Why it matters: Such numbers are preposterous. The fact that they're being talked about with anything approaching a straight face is indicative of the degree to which the broader AI discourse has become unmoored from reality.

The big picture: Altman, the CEO of OpenAI, has become the most recognizable avatar of the AI revolution writ large.

• For some months now, dating back to before he was briefly ousted from OpenAI, he has been pitching some of the world's richest governments and investors on a plan to massively increase the supply of the silicon that is necessary to power the AI revolution.
• Altman is helped in his quest by his name recognition; by countries' fear of missing out on the AI rocket ship; and, of course, by the dollar signs that flash in front of investors' eyes whenever they look at something like the Nvidia share price.

Between the lines: Altman has also benefited from Elon Musk blazing the way. Musk wasn't just a co-founder of OpenAI, he also set a precedent in terms of presenting himself as being placed on this planet to achieve an objectively impossible goal.

• If Musk can be taken seriously when he talks about colonizing Mars, then Altman's multi-trillion-dollar dreams seem downright mundane in comparison.

By the numbers: $7 trillion is 219 times the $32 billion that TSMC, the world's best and largest chip manufacturer, ran up in capital expenditures in 2023.

• It's also 206 times the $34 billion cost of the Manhattan Project, adjusted for inflation.
• It's more than 8 times the entire annual budget for the U.S. Department of Defense.
• At the cheapest possible cost of funds — the U.S. Treasury risk-free rate of 4% — it would cost $280 billion per year to service. That's 10 times TSMC's 2023 net income, or 30 times the scarcity-boosted net income of Nvidia.

Be smart: As Nvidia founder Jensen Huang pointed out this week, the cost of compute has decreased by a factor of 1 million over the past decade. That has enabled the development of generative AI — and, extrapolating forward, it suggests the future can be built for much less than $7 trillion.

• Meanwhile, the practical obstacles to building vast numbers of new chip factories are even bigger than the financial ones. There's already an extreme shortage of qualified humans who can construct such things, and qualified humans, unlike computer chips, can't be churned out at scale on an assembly line.

The bottom line: All this talk of trillions should not be taken literally. Instead the best way to think about it is as a flag planted in the discourse — the idea (almost certainly false) that there is no sum of money that's too big to invest in AI.