The Receipt Nobody Wants to Read
In 2024, analysts at Sequoia Capital asked a simple question. If AI companies keep spending the way they are right now, how much money would the industry need to earn every year to justify it?
Their answer: $600 billion.
Not what AI is earning today. What it would need to earn for any of this to make financial sense. Right now, generative AI is making real money , billions in revenue across OpenAI, Anthropic, and others. But billions is not $600 billion. The gap between what exists and what would need to exist is the story nobody in the industry wants to tell out loud.
The Two Numbers That Don't Match
78% of companies now report using AI in some form. 61% of all global venture capital in 2024 flowed into AI-related businesses. You would expect those numbers to be producing clear, consistent returns.
They are not. About 75% of the financial benefits from AI are flowing to just 20% of the companies. The majority are experimenting. Deploying features. Integrating tools. Not transforming their businesses.
In the early days of this cycle, companies claimed AI without really using it. Today, companies are using it but do not know how to make money from it. That is a harder problem. The stakes are higher, the expectations are enormous, and the gap between what was promised and what is happening is getting harder to ignore every quarter.
The Wall That Isn't Financial
The deeper constraint on AI isn't money. It's electricity.
By 2022, data centers were already consuming roughly 460 terawatt-hours of electricity globally every year , about the same as all of Germany, or 2% of total global demand. By this year, that number could reach between 620 and 1,000 terawatt-hours, depending on how aggressively AI keeps growing.
Here is what that energy actually goes to inside a data center: 40% runs the computing. 40% cools the machines. The remaining 20% handles everything else , moving data, maintaining stability, keeping the operation from collapsing under its own weight.
Nearly half the energy poured into AI does not make it smarter. It just keeps it alive.
Power grids in multiple US states are already being pushed toward capacity limits. New data centers are being rejected by utilities that simply cannot supply them. Microsoft, Google, and Amazon are all exploring nuclear energy partnerships , not because it is cheap, but because there is no other source large enough.
The Hidden Labor No One Advertises
Behind the polished interface is a category of cost that does not appear in most AI industry analysis: the human labor required to make these systems work.
Large language models do not self-train. They require massive amounts of human annotation , workers reviewing outputs, flagging errors, providing feedback signals that shape what the model considers a good response. This work is overwhelmingly done by contractors in low-wage markets, paid by the task, often without benefits or job security.
The automation promise runs on a workforce that nobody calls workers.
What Happens When the Bill Arrives
None of this means AI is failing. The technology is genuinely capable. The revenue is real. The companies at the top of the 20% are building durable businesses.
What it means is that the $600 billion gap between current reality and financial sustainability will have to close somehow. Either through revenue growth, or through contraction. The companies that have been deploying AI because everyone else was deploying AI , without clear ROI, without a plan for the energy costs, without accounting for the labor underneath , are the ones most exposed when the calculation comes due.
The receipt is being written. It will arrive on someone's desk.