Why Are They Spending Trillions on AI?
On August 10, 2026, Nvidia and six of the biggest names in finance — Apollo, Blackstone, BlackRock, Brookfield, Goldman Sachs, KKR — announced they’d try to mobilize over $500 billion for AI infrastructure. BlackRock’s Larry Fink said, on the record, that the real number over the coming years will be “trillions.” Nvidia’s Jensen Huang put a price on a single gigawatt of build-out: “something like $50, $60 billion.” Those are real, on-camera statements. Whether the belief behind them is correct is a separate question nobody has proven yet — including me.
I spent a research pass tracing the actual chain this money buys: energy becomes hardware, hardware becomes compute, compute runs models, models produce tokens, tokens become “AI working capacity” — the raw ability to get something done. That capacity gets pointed at real work. And only then, at the very last step, does it produce an outcome worth something to somebody. That last arrow is the one everyone skips. A cheap token is not a valuable token. A fast wrong answer is still wrong, just faster.
Huang calls compute an investable asset. Sam Altman calls intelligence a utility that gets cheaper and therefore gets used more. Mark Zuckerberg says the endpoint should be personal — AI you own, not AI that owns you. I don’t think any of the three is simply right or simply selling something false. Each has a stake in being believed. I laid their numbers side by side instead of picking a winner.
I wrote the whole thing up — sources, formulas, the corrections I made along the way, six different reader angles, none ranked above the others — as a free paper. Read it, poke holes in it, redo the math with your own numbers, republish it, disagree with me in public. That’s what the license is for.
Licensed under CC BY 4.0. Suggested attribution: Research synthesis and modelling by Valto Loikkanen, assisted by AI.