The race to build artificial intelligence has a staggering new price tag. According to Wccftech, Foxconn — the Taiwanese manufacturing giant best known for assembling iPhones — has pegged the cost of an NVIDIA "Vera Rubin" AI datacenter at roughly $47 billion for every gigawatt of capacity.
That figure is for the buildout itself. On top of it, Wccftech reports that the electricity to run such a facility would run about $1.3 billion per year. In other words, the power bill alone for a single gigawatt-scale site rivals the annual revenue of a mid-sized company.
Vera Rubin is the codename for one of NVIDIA's next-generation AI computing platforms, named after the pioneering astronomer. Datacenters built around it are designed to train and run the kind of large AI models that have driven much of the recent tech spending frenzy.
Why does a single company's cost estimate matter? Because it helps quantify something that has so far been described mostly in vague superlatives: just how expensive the AI infrastructure boom really is. When industry leaders talk about needing gigawatts of computing power, the Foxconn numbers translate that abstract goal into hard dollars — tens of billions to build, and more than a billion a year just to keep the lights on.
Those figures also underline why electricity, not just chips, has become a central constraint on AI's growth. A $1.3 billion annual power bill points to enormous energy demand, raising questions about where that power will come from and what it means for grids and emissions.
For now, this is a single estimate reported by one outlet, and the broader picture will depend on how many such facilities actually get built. But it offers a concrete sense of scale.
The story matters because it puts a real number on the AI arms race — showing that the cost of competing is measured not in millions, but in tens of billions per gigawatt.