The board measures the AI buildout. This page explains it in plain words — every claim here is backed by a sourced figure on the board.
A warehouse-sized building packed with computers. AI data centers are filled with specialized chips — NVIDIA's H100 is the workhorse — that train and run AI models. They need constant electricity and cooling, which is why the buildout shows up in power markets and water records, not just tech news. The biggest known AI data center holds roughly 1.1 million H100-equivalents, and a gigawatt-scale one takes about two years to build (Epoch AI).
On the board: the US map and counts
Because demand for computing is growing far faster than chips are improving. The computing used to train frontier AI has grown about 5x per year; chips get only about 1.5x better per dollar per year (Epoch AI). The gap gets filled with more machines, and every machine draws power around the clock. World data centers used about 415 TWh in 2024, projected toward 945 TWh by 2030 (IEA, all workloads) — a running average of ~108 GW, comparable to about 129 reactors the size of Three Mile Island Unit 1.
On the board: the grid chapter · US electricity prices
Two different answers. The buildings and chips: the companies — Microsoft, Alphabet, Amazon and Meta reported $513.8 billion of capital spending in the year to June 2026, from their own filings. The grid: utilities pass wholesale electricity costs into customer bills. After the PJM region's capacity price rose 9.3x in one auction, estimates for the first repriced year ran around 1.5–5% bill increases in PJM states (PJM figures, via WHYY). Other regions run their own markets.
On the board: the 22-year price table
Here is what can actually be measured: employers named AI as the reason for 120,136 announced US job cuts in the first nine months of 2026 (Challenger) — 2.2 times all of 2025. That is the employers' stated reason for announced cuts — not the same as net jobs lost, and nobody yet publishes a reliable net figure. The board reports the attribution series and says exactly what it is.
On the board: the series by year and month
Sources and formulas for every figure: the methodology. © 2026 BRAIAIN, a dnilgis project.