Mission

The physical infrastructure for AI is being built — data centers, substations, fabs, and the mines and refineries under them — faster than the statistics can count it. Silicon Genome builds instruments for measuring it. Our flagship index, the Hyperscaler Buildout Index, tracks the combined capital expenditure of the six largest builders, monthly, as filed. Every series we publish is constructed to be reconstructed: sourced to the filing, dated to the filing, nothing interpolated or estimated.

The closest precedent is the railroads. The network went from 9,000 miles in 1850 to 254,000 by 1916, financed largely through the securities markets and corrected by the bankruptcies of 1873 and 1893. The Interstate Highway System spanned 42,795 miles and took over three decades to build, with the last stretch of road — through Glenwood Canyon in Colorado — taking 12 years and half a billion dollars. The fiber buildout of the late 1990s was laid against demand forecasts nobody could verify, and the reckoning came when announced capacity and lit capacity turned out to be different numbers.

Each of those buildouts was, at the time, the largest capital project in American history. Each was also counted badly, which led to mismanagement of funds, cost overruns, capital drawn against unmet milestones, and — in the railroads and in fiber both — receivership. The compute buildout is larger than all of them, and it's still in the early stages. Nothing exempts it from the same fiscal failures and, in a worst-case scenario, bankruptcy court. Reliable data is, therefore, the buildout's most critical asset.