AI is burning hot right now, and money is being thrown into it like there’s no tomorrow. But IBM bigwig Arvind Krishna recently did some math, and the conclusion is pretty sobering—the global AI data center bill is approaching $8 trillion, and at this rate, it’s basically impossible to break even.

How did he calculate this? Building a 1GW AI data center costs at least $8 billion, and those tech giants talking about building 20 to 30GW facilities would each need to shell out at least $1.5 trillion. The whole industry combined? About 100GW of computing demand, adding up to $8 trillion invested.

Krishna’s own words are even more blunt: “I don’t think you’ll ever make your money back.”

Why? To avoid losing money on $8 trillion, you’d need to generate $800 billion in profit every year just to make it reasonable. The problem is, which AI company’s business model can support those numbers right now? To make matters worse, GPUs need to be replaced every five years, so depreciation costs are like a bottomless pit.

He also mentioned another key point—using current large language models to build AGI (artificial general intelligence) only has about a 1% chance of success, in his estimation. That means most of the investment may never even make a splash, let alone be monetized.

Behind the capital frenzy, this math is very clear. The AI race is indeed lively, but whether the money can be made back is another question entirely.