Ed Zitron’s latest runs the bear case on AI capital spending at his usual length, and whatever you make of his framing generally, the underlying arithmetic is worth sitting with on its own. Hyperscalers will have sunk more than $1.3 trillion into generative AI by the end of this year, with another trillion already planned for next. His argument is that it would take more than $2 trillion in genuinely new revenue to make the current spend break even on its own terms, revenue that isn’t showing up in any disclosed number he or anyone else has been able to point to.
The numbers behind that claim are the part worth checking rather than just repeating. Meta, Google, Amazon and Microsoft added over $850 billion in property and equipment over four years and took on $307 billion in new on-balance-sheet debt to fund it. Nikkei’s reporting, which Zitron cites, puts another $1.35 trillion in commitments off the books entirely. Google sold $25 billion in bonds in July and got 1.6 times the demand it needed, which reads as confidence right up until you remember that oversubscribed debt is also what financing looks like immediately before the thing it’s funding stops paying for itself.
None of that is the same question as whether any individual product built on top of this spending is worth what it costs, and Zitron is careful about the distinction even while making the larger case. “Do not confuse some revenue coming out of these products with any kind of success” is the line that does the most work in the piece. Copilot is genuinely making money, he singles it out as the clearest success story, with users burning through roughly 25 times their subscription cost in actual usage. Anthropic and OpenAI are seeing similar behavior from their own API customers, some running $8,000 to $14,000 a month in usage against a $200 plan. Some of that gap is a subsidized price that isn’t built to last. None of it tells you whether a specific tool, the one somebody is actually paying for and using this week, is earning its keep.
That’s the split worth holding onto. Whether the industry’s trillion-dollar bet pays off is a different question from whether any one product built on top of it is worth its price, and the second question doesn’t need the first one resolved first. I don’t know if Zitron is right that the industry’s payoff date sits, in his words, somewhere between fuck knows and never. That’s a genuinely open question, and the honest answer is that nobody currently has the disclosure to close it either way.