How to read this
It shows what is being spent building AI infrastructure, and what has to come back to justify it, in plain numbers. Here is how those numbers are chosen, and how to read each chart.
Every number is sourced
Each figure on the site traces to a named institution: Goldman Sachs, Bain & Company, the U.S. Bureau of Economic Analysis, Slickcharts, Yahoo Finance, and a few others. Every number is checked against the source’s own wording. If a figure cannot be sourced, it is cut, not estimated. The full list is on the sources page.
Every figure shows its date
The panels do not all update on the same schedule, so each one shows its own “as of” date. The spending counter and the stock charts move on their own; the equipment figures update every quarter; the longer-range projections update only when the institution behind them publishes new work. When a number is a forecast, the panel says so.
It is a question, not a verdict
People keep asking whether AI is a bubble. The site does not answer that. It lays out the numbers that inform the question and lets you draw your own conclusion. It has a point of view about what the data shows, but it stops short of a call. You will not see the words “crash” or “mania” here.
None of this is investment advice.
The charts
The spending counter
Goldman Sachs projects that the largest tech companies will spend roughly $765 billion building AI infrastructure in 2026: chips, data centers, and power. Divide that across the seconds in a year and you get the rate the counter ticks at. Real spending is lumpy; the even pace is an estimate, and the counter’s own note shows the exact math.
Stock market concentration
Each block is a company, sized by its weight in the S&P 500. The red blocks are the Magnificent Seven; the grey field is everyone else. Together the red blocks are about a third of the entire index, and every one is a tech or AI company. Use the toggle to compare today with 2004, when no such cluster existed.
The stock market, two ways
The two lines are the same 500 companies, measured differently. One weights each company by size, the way the market actually trades; the other counts every company equally. When the size-weighted line pulls ahead, the gains are coming from the biggest companies, not the average one.
The bars below the chart estimate how much each of the Magnificent Seven added to (or subtracted from) the market’s rise: a company’s weight in the index times its own price move. A few names do most of the lifting; a few quietly drag. The panel’s note explains the calculation.
Business equipment
Each bar is one type of business equipment, and its length shows how much investment in that category grew or shrank over the past year: right for growth, left for decline. The point is the shape. Computing gear is up sharply while every other category is close to flat.
A note on the measure: these are inflation-adjusted (“real”) dollars from the BEA. For computing gear, that adjustment also accounts for how fast the hardware improves, so part of the rise reflects better chips per dollar, not only more dollars spent. The chart shows each category’s growth rate and the gap between them, not a claim about exact spend.
The gap
Two charts, each showing a distance. The first is money: how much revenue the build-out needs to earn back by 2030 (a Bain projection) against what it is on track for. The second is price: how far AI-linked stock prices ran in 2025 against the earnings expected to justify them (a Goldman figure). In both, the red line is running ahead of the grey.
Want the underlying research? See the full sources & reading list.